The Nine Hundred KS: A Review of AMD's Latest CPU

In our latest review, we put AMD's new 9900 K and its variant, the 3900 X, through their paces to see how they stack up against each other. To start, we'll take a look at the 9900 K. We've been testing this chip for some time now, and it's proven to be one of the most powerful CPUs on the market. With its high clock speed and large number of cores, it's no surprise that it's been able to edge out the competition in our benchmarking tests.

We've tested the 9900 K with various workloads, including a render test using Blender and a synthetic test using Unigine Heaven 4. In both cases, the 9900 K was able to produce some impressive results. However, we also want to compare it to its variant, the 3900 X, which is priced slightly lower. The difference between these two CPUs isn't huge, but it's still worth noting.

One of the things that stands out about the 9900 K is its ability to overclock. We've been able to push this chip up to some pretty high frequencies, and it's been able to maintain its performance even at those higher speeds. This makes it a great option for anyone who wants to build a PC that can handle demanding workloads.

Another thing we want to talk about is the 3900 X itself. Like its sibling, the 9900 K, this chip is also very powerful and is well-suited for gaming. It's priced lower than the 9900 K, but it still offers some impressive performance. In terms of pricing, the 3900 X is a great option if you want to get into CPU overclocking without breaking the bank.

Now that we've covered both CPUs in detail, let's talk about their performance in different workloads. As we mentioned earlier, the 9900 K was able to edge out the competition in our benchmarking tests. However, this isn't entirely surprising - after all, it's a high-end CPU designed specifically for gaming.

However, the 3900 X is also very good at what it does, and it's worth considering if you're looking for a more affordable option that still offers some impressive performance. In terms of its performance in different workloads, the 3900 X actually comes out ahead of the 9900 K in some cases - specifically, in our Blender render test.

For example, the 3900 X was able to produce some impressively good results in this test, even though it's priced lower than the 9900 K. This is a great option for anyone who wants to build a PC that can handle demanding workloads like video editing or 3D modeling.

So what should you buy? Well, it really depends on your specific needs and priorities. If you're looking for a high-end CPU designed specifically for gaming, the 9900 K might be the better choice. However, if you want a more affordable option that still offers some impressive performance, the 3900 X is definitely worth considering.

One thing to keep in mind, however, is that there are other CPUs on the market that offer similar performance at lower prices. For example, the AMD Ryzen 5 3600 and Intel Core i7-9700 K are both great options for gaming and content creation.

Ultimately, the choice between these CPUs will depend on your specific needs and priorities. We recommend taking a closer look at each of them in our benchmarks to see which one is right for you.

In terms of the 9900 K, it's worth noting that we've been able to push this chip up to some pretty high frequencies using liquid nitrogen overclocking. This makes it a great option for anyone who wants to build a PC that can handle demanding workloads.

However, we also want to note that the 3900 X is just as good at overclocking, and it's priced lower than the 9900 K. So if you're looking for a CPU that can handle high-end workloads without breaking the bank, the 3900 X might be the better choice.

Another thing worth noting is that the 9900 K and the 3900 X are essentially the same chip - they just have slightly different specifications and prices. The main difference between them is the frequency and voltage at which they're clocked.

For example, the 9900 K has a higher clock speed than the 3900 X, but it also consumes more power. This means that the 9900 K might be better suited for builds that require more power, while the 3900 X might be better suited for builds that require less power.

Ultimately, the choice between these CPUs will depend on your specific needs and priorities. We recommend taking a closer look at each of them in our benchmarks to see which one is right for you.

In conclusion, we've put the AMD 9900 K and its variant, the 3900 X, through their paces to see how they stack up against each other. While both CPUs offer some impressive performance, there are some key differences between them that should be considered before making a purchase decision.

"WEBVTTKind: captionsLanguage: enin total I $9.99 hundred KS launches today abend CPU that should allow for better frequencies at a given voltage the CPU is best known for its marketing heavy five gigahertz claims something that will reopen the can of worms relating to MC e TDP and what really counts as five gigahertz intel has also dropped some of its prices recently slotting me ninety-nine hundred KS into the spot where the 99 hundred K previously sat it then brought its top-tier CPUs down a few bucks as well the end result is a sort of kind of new ish product that claims higher clocks and will exist alongside the 9900 K mostly because it is a 9900 K but it's a bit faster out of the box critically the KS has been silicon that should mostly guarantee higher clocks at a lower given voltage or vice-versa well look into that today as long as whether five gigahertz really holds at TDP and some benchmarks for power and gaming and production workloads before that this video is brought to you by the be quiet dark rock slim keeping a high-performance focused but reducing the footprints to accommodate full memory slot compatibility the dark rock slim comes with one of be quiet silent wins three 120 millimeter fans built for low noise operation the dark rock slim at ver ties is cooling capability up to 180 watts EDP but mixes in a matte black out color design to combine performance and looks learn more at the link in the description below the three most important bullet points first the night I heard KS should be 513 ash dollars or so the 9900 k9s that has existed for about a year now we're a little more actually should come down to about 485 and it still has the IGP it can be overclocked to kind of match a KS but it's not as good and the 9900 KF without the IGP should be $470 and that's the second time and 48 hours have had to reference notes because of silly naming schemes so first off this discussion restarts in important points about TDP and MCE or multi core enhancement which is what motherboard makers used to refer to their version of removing things like power limitations and pulling a more power requiring more cooling as a result to give you better performance out of the box Intel's definition of TDP is a little bit closer to wand one with power if it's followed but it still has some massive asterisks beside it she probably probably careful with the phrase massive asterisks it's kind of like big buttons where some massive asterisk on the internet is going to crop it out and he says out of context so we'll move on as quickly as possible officially Intel's power limit one is 127 watts which you'll notice is directly related to the TDP 127 watts for the CPU so like previous Intel CPUs what this really means is that the power budget recommended to motherboard manufacturers as the limitation for pl1 is 127 watch they don't have to follow these guidelines we consider it basically a spec but it's not a formal requirement to follow that that spec or guidance so a limited duration turbo is allowed to exist they can even extend this to infinity so it's not limited duration anymore and that provides a greater power budget for higher clocks exiting the 127 watt pl-1 in the instance of the 9900 Kaos the formal guidance from Intel is a 28 second to 8 second P l2 which defaults to 1.25 x PL one or one point two five times one 27 equals 150 nine ish watts for default PL - it's under PL - that the 99 hundred KS will boost to five gigahertz all core upon PL - expiry or again 28 seconds on boards that play by the guidelines like the ACS Maximus 11 hero we used a test the CPU will drop down to its 127 watt PL one limiter and will clock accordingly in all core applications like blender this might be something like a range of 4.4 to 4.6 gigahertz and we do have a chart on the screen if your tabbed away to help see that it's not 5.0 gigahertz so it will drop as P l1 is enforced along with the turbo limits the duration limits on a board like the gigabyte z3 90 extreme and overclocking focus board with all power limits functionally disabled the CPU will stick to 5.0 gigahertz all core at the cost of significantly increased power the gigabytes 90 extreme sits at an EPS 12-volt power consumption that's that's measured at the cables before vrm efficiency losses but after the wall so are more accurate here that power consumption is about 196 to 2 or 3 watts on the gigabyte z 390 extreme sustained or about 16 amps EPS 12-volt power that's not 1:27 clearly it's it's quite a bit different the HDS board meanwhile will start at about 15 amps maybe 15 point 5 or 186 watts and fall to 127 Watts within 30 seconds following the Intel guidance with MCE disabled almost precisely if not precisely you could bypass this just like is done on the gigabyte board but it's done with a few prompts so at this Junction then it is not possible to both achieve 5.0 gigahertz all core and achieve a 127 watt power consumption for the CPU acknowledging that TDP is not literally power consumption yes but Intel's is pretty close typically once you get past the initial turbo limits and you enter steady-state with turbo durations enforced it is mostly one-to-one in this scenario it's either 5 gigahertz or it's 127 watts and which is true will depend on the motherboard not on Intel so Intel is not creating this behavior they provide a guideline the motherboards execute asus gets credit for following Intel guidance and sticking to the 127 watt number and 28 second number gigabyte gets credit for sticking through what's been marketed 5 gigahertz and following what consumers are expecting out of their product who's right and who's wrong it's it's arguable until probably should look into firmer guidance in the future because of this discrepancy and because of confusion that it creates in the market otherwise we're gonna have to address this for eternity so anyway let's move on all of our testing in this instance that's a 127 watt number with PL 2 turbo duration of its enforced we considered exiting PL 2 to be a violation of the spec for stock testing and as long as the consumer is made aware of what's happening it's fine to exit the power limit guidance but for purposes of cooler and motherboard selection the power does play a role obviously the power level and we do prefer to see boards that stick to MCE equivalents off and power targets in forest overclocking boards a little bit more leeway there but we're gonna show blender first this is more of a gaming CPU so it makes more sense to show gaming first because it'll do better they're comparatively but we're showing blender and that's because it'll allow us to better illustrate the difference of MCE on and off between the two boards we talked about earlier so with the GL monkeyhead stress test the 900k stock Seaview completed the render in 20 minutes decreased by eight point nine percent with the KS stock CPU to eighteen point five minutes the nine hundred K and KS at five point one baath hit about seventeen minutes with the nine hundred K at five point two gigahertz at sixteen point seven minutes that said running with gigabyte and allowing the power to run amok without restriction the KS improved it says seventeen point three minutes for the monkey had render a drop of six point five percent versus the Asus board where MCE was off and power limits were enforced the 3900 acts outpaces both thanks to its higher core count which is just the nature of tile-based renderers like blender frequency matters a bit but threads count the most is here and the extra four cores pushes the 3900 X to twelve point eight minutes the GM logo render is a bit more abusive in this render the 9rk stock cpu manages render completion in 26 minutes with a stock KS at twenty three point eight minutes the time reduction is about nine percent from k2 KS stock to stock similar in scaling to the previous result overclocking the CPUs puts them at around 21 minutes with the 3900 X at about 16 minutes for the gigabyte results the CPU ended up finishing in twenty one point eight minutes when unrestrained by silly things like Intel TDP and recommended power levels and so it finishes eight point four percent faster than the Asus board with MCE off and that's not say either is better it's just a show hey if someone tests or benchmarks this product with MCE on versus off there will be a difference in performance results and you can see that here we plan to put the night I heard KS or one of them under liquid nitrogen soon so make sure you check back for that we have a delivery of liquid nitrogen coming in actually when this video goes up and we have some basics to discuss you there for normal people use a crack in AK 62 280 millimeter CLC is our standard liquid cooler for CPT testing and we choose this because it's something that's achievable it's normal and we could get more overclocking Headroom out of the KS if we upgraded the cooler to something like say a 360 with really loud fans but that's no longer really that realistic so we'll save that for later so it's also kind of cheating because we use the 284 the 900 K and everything else anyway so unfortunately the ASUS Maximus 11 hero that we use had issues with V core on this one we couldn't seem to we were unable to really control it in a way that was that followed what we were typing in it's very frustrating so we tried things like different boot voltages for V core we set V core itself we tried offsets and adaptive configurations all kinds of things it didn't really play along so working with those parameters of maybe a buggy bios we ended up settling on 5.2 gigahertz at what it provided us which was 1.3 - 3 8 volts that was the max we could get out of that board for today anyway with the 48 X on-court this is a good voltage and a good frequency combo but we had limited control so we took it over to an EVGA Z 390 dark which had a lot of its own problems behind you and lacked some of the BIOS options we really liked on the Maximus but we took it over there we were able to get 5.4 gigahertz with an AV X offset of 2 so we couldn't hold beyond 5 points here for something like blender it was pegged at 95 98 C and because the dark lacks some of the options for temperature tuning that was a limitation that we ran into so 5.2 is the max we could reasonably hold with this cooling solution for AVX workloads like blender for non AVX games we could kind of do 5.4 but it was not stable to the extent that we did encounter a couple voltage drops every now and then that would introduce instability or skew the results in a negative fashion to be worse than stock things of that nature so unfortunately we did end up settling for 5.2 5.3 didn't quite hold we need a better cooling solution for that as well so and we have them but it's not within the scope of what we're trying to do so we settled at 5.2 that's the end of it 1.3 G 3 volts pretty damn good but we'll have to do more with hydrogen later and that's what our benchmarks will be with 5.1 for the 900k won't be in all of the tests at this point it needs to be rerun in some of them we have it in a couple but it frankly doesn't matter because in 900k at x frequency should equate a 900 KS at X frequency as well it's the same CPU after all let's get into some of the testing and finally power consumption is up now for power consumption benchmarks we test that the EPS 12 volt rail to get a relatively accurate preview RM power measurement this is more accurate than power draw at the wall as it isolates the CPU before vrm efficiency losses and we don't have to worry about all the other overhead from the rest of the system components we're pretty close to true CPU power draw here starting with blender the Intel I 999 hundred KS CPU had stock ended up at 126 watts after the turbo duration expired despite running out of fixed 5.0 gigahertz all core the 99 hundred KS still follows the Intel turbo window power policies and other Intel power policies when MCE is off before the turbo window expires the 1900 KS was drawing about 186 watts in the same workload but this is only for the first 20 seconds or so after that the power consumption Falls to 126 watts this is still higher than the original 99 hundred K it's a no one's surprise and actually does follow the TDP metric pretty closely as a one-to-one we just published a massive deep dive on what TDP is for an D and how it works and how it isn't related to power consumption directly on AMD so check that out for more information on TDP intel's TDP measurements though are different from how antes work and in this instance tend to follow the actual power consumption more closely as long as the motherboard follows the Intel power policies and as long as you're measuring the power after any turbo duration windows expire Intel denotes a TDP of 127 watts so we're right on top of that after the turbo window this wouldn't be true with MCE on as that put us in the 186 watt scenario similarly the 9900 k9s model sits at 91 watts when following the Intel spec limits so it too is at about the original TDP the power consumption increase here is for the frequency increase which can sometimes also require additional volt ability in this case though we're weakened with a specially bend chip that's the entire KS line so the required voltage I was at given frequency should be lower on the whole it's just that the frequencies higher - and finally note that these are all taken on the same motherboard so that matters where some boards will push higher Auto voltage than others but we use the same one for this for comparison the r9 3900 X in this test sits at 148 watt stock with 171 at 4.3 gigahertz all core tonight a hundred K Paul's 268 watts when overclocked at one point for one volts but the KS can manage this same frequency at a lot less voltage in fact our 9900 KS was doing 5.2 gigahertz with zero AVX offset so straight v 2 at 1.3 to 3 V core the end result was 246 watts for 5.2 gigahertz as opposed to our previous one point 4 1 volts five point one gigahertz configuration at 268 Watts on the non SS q that's a big improvement and voltage requirement reduction from the bending alone if you're curious about the inverse look at this single threaded power consumption we have a more slim down chart for that we only added this one for the Andy rise in 3000 series so we have a lot fewer CPUs on the chart only the 99 K original and the binding I had a KS or from Intel here as the rest haven't been rerun through Cinebench R xx just yet the Cinebench 1t results position that I entered K at twenty eight point eight watts originally or 30 watts with the 99 hundred KS there's no meaningful difference here but that's also somewhat expected remember the 99 hundred K could still turbo up to five gigahertz and with 1t loads that's a common occurence the 900 KS is special for again that lower voltage requirement but primarily because it's configured to hold 5.0 out of the box these results then are pretty predictable power consumption for Cinebench R xx and T is up next for another multi-core look in this one again with the limited amount of parts on the table the night I heard KS stock CPU strikes 182 Watts during the short work load the 900 k9s is at 146 Watts both of these results are higher because the CPUs are completing the test before the duration expiry so unless running this test back-to-back and rapid succession the result will be more of the fully boosted power matric blender shows lower figures because it's a longer test and we wait to see what the steady-state power consumption is that's the real difference between the Cinebench test and blender in this test the 9900 KS at 5.2 gigahertz and 1.3 to 3 volts draws 234 watts which is actually pretty good for a 5.2 gigahertz clock thanks to those lower voltages the 3900 X at 4.3 gigahertz is running 173 watts with the stock 3900 X at 148 watts and close to the stock 9900 K original CPU we're going to pull up one last older chart this one doesn't have the modern AMD parts on it but it has an older Intel and previous gen AMD parts that can be worth looking at this gives a representation of a CV to heavy workload in a gaming environment using a heavy physics simulation as the low generator we're able to see what the CPUs look like when they're doing something other than tile-based rendering in this scenario the 9900 KS is pulling 144 Watts when stock the 9900 K is at 127 Watts when stock and the 9900 K with a 5.1 gigahertz 1.41 volt o C is at 210 watts the 99 hundred KS can hold a 100 megahertz higher frequency at 1.3 to 3 volts a big drop in voltage requirement for a 180 watt power reading that's a reduction in power consumption of 14 percent versus the technically slightly slower 5.1 gigahertz 99 K original sample and it speaks to how much intentional bidding can help with power consumption and with voltage moving onto some gaming benchmarks next civilization 6 will start us out for this one we're looking at average AI player turn time rather than the usual FPS metric each second increase in turn time becomes palpable as player count increases so this can become significant as campaigns thrown on and the game gets more complex for this one the 1900 KS stock CPU managed a 29 point 1 second average turn completion time which has it right around the 9700 K at five point one gigahertz within 9,900 case stock CPU not the s not far behind at twenty nine point seven seconds keep in mind two things one tests are imperfect and have some variance but sieve is really consistent and we've controlled it as best we can to and more importantly Civilization six really likes frequency this most evident when looking at these sets of results just mentioned with the 9900 k stock sieve you can leverage its limited thread boosting frequencies to climb the charts since the cores aren't as heavily and constantly loaded overclocking the que ésta 5.2 gigahertz landed it at twenty eight point five seconds for average turn completion time it turned time reduction from the stock chaos of two percent the stock ninety nine hundred K at five point one gigahertz was holding 29 seconds average sitting right behind the five point two gigahertz at ninety nine hundred KS GTA 5 at 1080p is up now at 1080p the Intel I $9.99 hundred K stock CPU originally landed at 1:22 FPS average with lows at about eighty eight and seventy five the stock and ninety nine hundred KS configuration improved about six point six percent over baseline at 131 FPS average from 120 to the 1900 Kaden on s at five point one gigahertz our previous stable OC at one point four volts managed 132 FPS average lows between the two are functionally identical and are within error for lows which is a wider error the 1900 KS at five point two gigahertz hit 135 FPS average a boost of two point seven percent over the five point one gigahertz results for reference the r9 3900 AK stock CPU runs at 110 FPS average allowing the 99 hundred KS stock CPU a lead of nineteen percent over the r9 3900 ax when both are stock the 3900 X at four point three gigahertz all core with SMT on hit 113 FPS average but that doesn't move the needle much when we're comparing to overclock versus overclocked numbers disabled in SMT and screwing around with things like that helps but we don't recommend actually disabling SMT because then you're thrown away half of what you bought the CPU for so it's just a technology demonstration not something we recommend for GTA 5 at 1440p the 99 hundred KS maintains its positioning and the hierarchy remains largely unchanged the KS ends up at 132 FPS average when I overclocked to the coolers limits that's our limitation here or 129 FPS average one stock the RO 939 hundred X remains around 108 FPS average keeping about the same distance as before as for the 9900 k stock CPU non ass it's at 122 FPS average this same as 1080p and that's because it's CPU limited here not GPU limited so this illustrates that even at a high resolution until you really start pushing something like 4k the CPU can play a role in performance the KS is about six point three percent better stock to stock although overclocking both equalizes them albeit with a higher voltage for the non ass CPU the campaign benchmark for total war Warhammer - had the 9900 KS at 189 FPS average stock the highest result for any stock CPU currently on this chart the overclocked in 9900 K did slightly better with 5.1 gigahertz across all cores vs. 5.0 for the stock 9900 KS that's with the uncor set to 45 that was at 191 FPS average to 189 while the 5.2 9900 K s predictably did best at 196 FPS average it's impressive but this is also a test that doesn't care much for extra cores the eight core eight thread 9700 K averaged at 179 fps without any overclock at all and it's a great deal cheaper and these highest score on the chart is the 3900 X with a 4.4 gigahertz overclock and SMT disabled not a configuration we recommend but a good test and that brought it to 171 FPS average f1 20:18 gives us a dx11 title to look at but we'll move back to dx12 next for this one the 900 K and KS have finally found our GPU bottleneck at the top and the 900 KS at 5.2 gigahertz and 900 K s stock we're both stuck at 320 FPS average which happens to me where the 20 ATT I can no longer keep up with the CPUs this is all at 1080p mind you so it'll only get more GPU bound as the resolution goes up we can't really draw firm conclusions here since a limit imposes a barrier that cuts off the total gains possible from the CPU upgrade but the night I heard KS manages to outstrip the nine hundred K stock CPU by a few FPS we just don't have enough test resolution or Headroom here to define the absolute gains the r9 3900 X for reference runs at about 276 FPS average stock that's plenty clearly that's that's a lot of frames but if you're more interested in relative scaling it's about a 16% jump to the GPU bound 900 KS the next chart demonstrates our hard GPU bottleneck if you're curious in this instance if you're playing at higher resolutions the GPU fast become a limiter that can act as a great equalizer for higher-end CPUs the 9600 K 700 K 900k 700 K there sorry nine hundred KS can't mix those up we're all functionally equal in this chart the difference is you're seeing those plus or minus 3 FPS swings are not actual determiners of performance but rather normal test variance from bouncing off of a GPU limit and averaging a more random set of numbers because we're limited by another device the 3900 X gets closer in instances of GP limitations at the top-end naturally and closes at 2:30 FPS average it's still behind at 1440p though moving up in resolution further might begin to lock the tooth heightens warrants that hitman 2 is up next tested in dx12 and first at 1080p the nine hundred KS stock runs at 142 FPS average out of the box with Louis at 77 and 47 1% a 0.1% lows respectively this puts the KS at 4% over the K stock CPU with hyper-threading on overclocking the KS puts it to 144 FPS average about a 5% gain for the increase in all core frequency and uncor although the latter does last for framerate the night I heard K hyper-threading off result hyper-threading disabled result helps demonstrate that we still have Headroom before hitting GP limitations as for the r9 3900 X the stock result runs at 121 FPS average allowing the 9900 KS a lead of about 17.6% over there are nine 3900 acts on this workload shadow the tomb Raider's next another DX 12 title that's fairly decent for for testing and it's new that is 500 KS stock CPU ended up at 172 FPS average of lead of about 1% over the united case stock cpu and this charge the KS at 5.2 did little for performance a mixture of nearing GPU limiters and diminishing returns on cpu clock into this title so it's a bit of both let's throw a Dobby premier in here to Adobe premier will give us another look at a production workload accompanying our earlier blender results for this one the r9 3900 ax is back in the driver's seat leading the pack in a 1080p show floor render with a 3.4 minute render time requirement to get the job done the 900k stock CPU required 3.8 minutes allowing the 3900 X an advantage of 10.5% the united 900k asks the asus board doesn't really improve much moving to three point seven minutes required if you did this with the gigabyte board or enabled MCE then yeah I'd be faster but that's that's cheating this is not running stock overclocking the 5.1 gigahertz helps a bit more but it's still not in the lead in this instance the 3900 X is holding its own for 4k render results it's more or less the same but the scale stretches a bit the difference is the 1950 X sort of closets way into the results and displaces the 9900 K by another rank the nine hundred KS ends up at 11.15 minutes for this render with the 9900 K at eleven point nine minutes we're gonna cap it there we do technically have a whole lot more production and synthetic results but frankly should just look at one of our reviews with the night ahead of K in it because it's basically that when we overclock it and there's no point in adding 15 more minutes to the charts to say that also it is presently four hours away from embargo lifts and I need to edit and upload this in time to hit embargo lifts so we're gonna cap it there we're planning to revisit this chip with liquid nitrogen overclocking just for some fun see how it scales it should be better than our 900 K in theory it's Bend and then as for whether you should buy this first of all let's address something the night on our K is objectively a good processor the r9 3900 X is objectively a good processor where they're good depends a bit on which one you're talking about then I heard K and the KS because it's the same thing are good in gaming they are topping the charts you cannot presently get faster than those based on our charts HDD T CPUs don't really add value here in fact a lot of them will drag down the performance if you go to a 99 80 XE and probably a 10 9 8 exe you're very likely going to end up in a lot of games with less performance than a higher frequency lower core count 8 core 900 K so the 9900 K and the KS are at the top of the charts for gaming and that's really all there is to that that doesn't mean the 3900 X is bad or that you shouldn't buy it the 3900 X is at the top of the charts for things like blender it is also impressively good and Adobe Premiere and and that's frankly not something we really expected out of this generation from AMD didn't even expect together frequencies at got out of Zen too so that was pretty cool to see the way it breaks down then is the same as before nothing's really changed it's just a bit faster now if you're buying a gaming only building a gaming only computer and you really really care about high frame rate and you're using the highest end possible GPU and you don't really care about things like blender rendering with cycles on a CPU it's a common use case but it's not that common or you don't care that much about premiere or maybe you care a little bit about from here but you care more about gaming and deny outer case tool is one premiere anyway so you weigh them and gaming comes out ahead whatever the case may be the Niantic K has those use cases in mind the 3900 X is a better choice if you have $500 to spend and that's what you're planning to spend for something like blender for tile based rendering for v-ray for decompression with 7-zip if those things sound appealing to you you should buy a 3900 X and if gaming at super high frame rates isn't that important maybe you play games sometimes but you don't care if it's 16 17 percent well that maths funny but even a 900 K is 16 or 17 percent ahead of a 3900 X if that doesn't matter to you then 3900 X is fine so you have to very well-defined options here this is not us being wishy-washy this is not us waffling between them we've given you two very firm recommendations depending on what your workload is and now it's up to you to figure that part out and choose based on that also as a reminder there are a lot of very good CPUs that aren't $500 and are cheaper and you don't have to come up this high in price to get something that's good for gaming or for production see also r5 3600 and well for just about everything really or i7 9700 K for gaming so that should wrap us up KS is a K just with its Bend and you'll get better frequency voltage curves VF curves out of it I'm just repeating myself at this point if you jumped here you want to tell the our then go away go back to the beginning and watch the whole thing not gonna recap it any further than that thanks for watching subscribe for more go to store that game is exes dotnet to support us directly like by buying one of our mod mats the large ones are in stock and shipping now we just received the most recent batch so they're no longer on backward thank you for your patience and you go to patreon.com/scishow hearings Nexus you would like to support us there as well have a lot to edit I'll see you all next timein total I $9.99 hundred KS launches today abend CPU that should allow for better frequencies at a given voltage the CPU is best known for its marketing heavy five gigahertz claims something that will reopen the can of worms relating to MC e TDP and what really counts as five gigahertz intel has also dropped some of its prices recently slotting me ninety-nine hundred KS into the spot where the 99 hundred K previously sat it then brought its top-tier CPUs down a few bucks as well the end result is a sort of kind of new ish product that claims higher clocks and will exist alongside the 9900 K mostly because it is a 9900 K but it's a bit faster out of the box critically the KS has been silicon that should mostly guarantee higher clocks at a lower given voltage or vice-versa well look into that today as long as whether five gigahertz really holds at TDP and some benchmarks for power and gaming and production workloads before that this video is brought to you by the be quiet dark rock slim keeping a high-performance focused but reducing the footprints to accommodate full memory slot compatibility the dark rock slim comes with one of be quiet silent wins three 120 millimeter fans built for low noise operation the dark rock slim at ver ties is cooling capability up to 180 watts EDP but mixes in a matte black out color design to combine performance and looks learn more at the link in the description below the three most important bullet points first the night I heard KS should be 513 ash dollars or so the 9900 k9s that has existed for about a year now we're a little more actually should come down to about 485 and it still has the IGP it can be overclocked to kind of match a KS but it's not as good and the 9900 KF without the IGP should be $470 and that's the second time and 48 hours have had to reference notes because of silly naming schemes so first off this discussion restarts in important points about TDP and MCE or multi core enhancement which is what motherboard makers used to refer to their version of removing things like power limitations and pulling a more power requiring more cooling as a result to give you better performance out of the box Intel's definition of TDP is a little bit closer to wand one with power if it's followed but it still has some massive asterisks beside it she probably probably careful with the phrase massive asterisks it's kind of like big buttons where some massive asterisk on the internet is going to crop it out and he says out of context so we'll move on as quickly as possible officially Intel's power limit one is 127 watts which you'll notice is directly related to the TDP 127 watts for the CPU so like previous Intel CPUs what this really means is that the power budget recommended to motherboard manufacturers as the limitation for pl1 is 127 watch they don't have to follow these guidelines we consider it basically a spec but it's not a formal requirement to follow that that spec or guidance so a limited duration turbo is allowed to exist they can even extend this to infinity so it's not limited duration anymore and that provides a greater power budget for higher clocks exiting the 127 watt pl-1 in the instance of the 9900 Kaos the formal guidance from Intel is a 28 second to 8 second P l2 which defaults to 1.25 x PL one or one point two five times one 27 equals 150 nine ish watts for default PL - it's under PL - that the 99 hundred KS will boost to five gigahertz all core upon PL - expiry or again 28 seconds on boards that play by the guidelines like the ACS Maximus 11 hero we used a test the CPU will drop down to its 127 watt PL one limiter and will clock accordingly in all core applications like blender this might be something like a range of 4.4 to 4.6 gigahertz and we do have a chart on the screen if your tabbed away to help see that it's not 5.0 gigahertz so it will drop as P l1 is enforced along with the turbo limits the duration limits on a board like the gigabyte z3 90 extreme and overclocking focus board with all power limits functionally disabled the CPU will stick to 5.0 gigahertz all core at the cost of significantly increased power the gigabytes 90 extreme sits at an EPS 12-volt power consumption that's that's measured at the cables before vrm efficiency losses but after the wall so are more accurate here that power consumption is about 196 to 2 or 3 watts on the gigabyte z 390 extreme sustained or about 16 amps EPS 12-volt power that's not 1:27 clearly it's it's quite a bit different the HDS board meanwhile will start at about 15 amps maybe 15 point 5 or 186 watts and fall to 127 Watts within 30 seconds following the Intel guidance with MCE disabled almost precisely if not precisely you could bypass this just like is done on the gigabyte board but it's done with a few prompts so at this Junction then it is not possible to both achieve 5.0 gigahertz all core and achieve a 127 watt power consumption for the CPU acknowledging that TDP is not literally power consumption yes but Intel's is pretty close typically once you get past the initial turbo limits and you enter steady-state with turbo durations enforced it is mostly one-to-one in this scenario it's either 5 gigahertz or it's 127 watts and which is true will depend on the motherboard not on Intel so Intel is not creating this behavior they provide a guideline the motherboards execute asus gets credit for following Intel guidance and sticking to the 127 watt number and 28 second number gigabyte gets credit for sticking through what's been marketed 5 gigahertz and following what consumers are expecting out of their product who's right and who's wrong it's it's arguable until probably should look into firmer guidance in the future because of this discrepancy and because of confusion that it creates in the market otherwise we're gonna have to address this for eternity so anyway let's move on all of our testing in this instance that's a 127 watt number with PL 2 turbo duration of its enforced we considered exiting PL 2 to be a violation of the spec for stock testing and as long as the consumer is made aware of what's happening it's fine to exit the power limit guidance but for purposes of cooler and motherboard selection the power does play a role obviously the power level and we do prefer to see boards that stick to MCE equivalents off and power targets in forest overclocking boards a little bit more leeway there but we're gonna show blender first this is more of a gaming CPU so it makes more sense to show gaming first because it'll do better they're comparatively but we're showing blender and that's because it'll allow us to better illustrate the difference of MCE on and off between the two boards we talked about earlier so with the GL monkeyhead stress test the 900k stock Seaview completed the render in 20 minutes decreased by eight point nine percent with the KS stock CPU to eighteen point five minutes the nine hundred K and KS at five point one baath hit about seventeen minutes with the nine hundred K at five point two gigahertz at sixteen point seven minutes that said running with gigabyte and allowing the power to run amok without restriction the KS improved it says seventeen point three minutes for the monkey had render a drop of six point five percent versus the Asus board where MCE was off and power limits were enforced the 3900 acts outpaces both thanks to its higher core count which is just the nature of tile-based renderers like blender frequency matters a bit but threads count the most is here and the extra four cores pushes the 3900 X to twelve point eight minutes the GM logo render is a bit more abusive in this render the 9rk stock cpu manages render completion in 26 minutes with a stock KS at twenty three point eight minutes the time reduction is about nine percent from k2 KS stock to stock similar in scaling to the previous result overclocking the CPUs puts them at around 21 minutes with the 3900 X at about 16 minutes for the gigabyte results the CPU ended up finishing in twenty one point eight minutes when unrestrained by silly things like Intel TDP and recommended power levels and so it finishes eight point four percent faster than the Asus board with MCE off and that's not say either is better it's just a show hey if someone tests or benchmarks this product with MCE on versus off there will be a difference in performance results and you can see that here we plan to put the night I heard KS or one of them under liquid nitrogen soon so make sure you check back for that we have a delivery of liquid nitrogen coming in actually when this video goes up and we have some basics to discuss you there for normal people use a crack in AK 62 280 millimeter CLC is our standard liquid cooler for CPT testing and we choose this because it's something that's achievable it's normal and we could get more overclocking Headroom out of the KS if we upgraded the cooler to something like say a 360 with really loud fans but that's no longer really that realistic so we'll save that for later so it's also kind of cheating because we use the 284 the 900 K and everything else anyway so unfortunately the ASUS Maximus 11 hero that we use had issues with V core on this one we couldn't seem to we were unable to really control it in a way that was that followed what we were typing in it's very frustrating so we tried things like different boot voltages for V core we set V core itself we tried offsets and adaptive configurations all kinds of things it didn't really play along so working with those parameters of maybe a buggy bios we ended up settling on 5.2 gigahertz at what it provided us which was 1.3 - 3 8 volts that was the max we could get out of that board for today anyway with the 48 X on-court this is a good voltage and a good frequency combo but we had limited control so we took it over to an EVGA Z 390 dark which had a lot of its own problems behind you and lacked some of the BIOS options we really liked on the Maximus but we took it over there we were able to get 5.4 gigahertz with an AV X offset of 2 so we couldn't hold beyond 5 points here for something like blender it was pegged at 95 98 C and because the dark lacks some of the options for temperature tuning that was a limitation that we ran into so 5.2 is the max we could reasonably hold with this cooling solution for AVX workloads like blender for non AVX games we could kind of do 5.4 but it was not stable to the extent that we did encounter a couple voltage drops every now and then that would introduce instability or skew the results in a negative fashion to be worse than stock things of that nature so unfortunately we did end up settling for 5.2 5.3 didn't quite hold we need a better cooling solution for that as well so and we have them but it's not within the scope of what we're trying to do so we settled at 5.2 that's the end of it 1.3 G 3 volts pretty damn good but we'll have to do more with hydrogen later and that's what our benchmarks will be with 5.1 for the 900k won't be in all of the tests at this point it needs to be rerun in some of them we have it in a couple but it frankly doesn't matter because in 900k at x frequency should equate a 900 KS at X frequency as well it's the same CPU after all let's get into some of the testing and finally power consumption is up now for power consumption benchmarks we test that the EPS 12 volt rail to get a relatively accurate preview RM power measurement this is more accurate than power draw at the wall as it isolates the CPU before vrm efficiency losses and we don't have to worry about all the other overhead from the rest of the system components we're pretty close to true CPU power draw here starting with blender the Intel I 999 hundred KS CPU had stock ended up at 126 watts after the turbo duration expired despite running out of fixed 5.0 gigahertz all core the 99 hundred KS still follows the Intel turbo window power policies and other Intel power policies when MCE is off before the turbo window expires the 1900 KS was drawing about 186 watts in the same workload but this is only for the first 20 seconds or so after that the power consumption Falls to 126 watts this is still higher than the original 99 hundred K it's a no one's surprise and actually does follow the TDP metric pretty closely as a one-to-one we just published a massive deep dive on what TDP is for an D and how it works and how it isn't related to power consumption directly on AMD so check that out for more information on TDP intel's TDP measurements though are different from how antes work and in this instance tend to follow the actual power consumption more closely as long as the motherboard follows the Intel power policies and as long as you're measuring the power after any turbo duration windows expire Intel denotes a TDP of 127 watts so we're right on top of that after the turbo window this wouldn't be true with MCE on as that put us in the 186 watt scenario similarly the 9900 k9s model sits at 91 watts when following the Intel spec limits so it too is at about the original TDP the power consumption increase here is for the frequency increase which can sometimes also require additional volt ability in this case though we're weakened with a specially bend chip that's the entire KS line so the required voltage I was at given frequency should be lower on the whole it's just that the frequencies higher - and finally note that these are all taken on the same motherboard so that matters where some boards will push higher Auto voltage than others but we use the same one for this for comparison the r9 3900 X in this test sits at 148 watt stock with 171 at 4.3 gigahertz all core tonight a hundred K Paul's 268 watts when overclocked at one point for one volts but the KS can manage this same frequency at a lot less voltage in fact our 9900 KS was doing 5.2 gigahertz with zero AVX offset so straight v 2 at 1.3 to 3 V core the end result was 246 watts for 5.2 gigahertz as opposed to our previous one point 4 1 volts five point one gigahertz configuration at 268 Watts on the non SS q that's a big improvement and voltage requirement reduction from the bending alone if you're curious about the inverse look at this single threaded power consumption we have a more slim down chart for that we only added this one for the Andy rise in 3000 series so we have a lot fewer CPUs on the chart only the 99 K original and the binding I had a KS or from Intel here as the rest haven't been rerun through Cinebench R xx just yet the Cinebench 1t results position that I entered K at twenty eight point eight watts originally or 30 watts with the 99 hundred KS there's no meaningful difference here but that's also somewhat expected remember the 99 hundred K could still turbo up to five gigahertz and with 1t loads that's a common occurence the 900 KS is special for again that lower voltage requirement but primarily because it's configured to hold 5.0 out of the box these results then are pretty predictable power consumption for Cinebench R xx and T is up next for another multi-core look in this one again with the limited amount of parts on the table the night I heard KS stock CPU strikes 182 Watts during the short work load the 900 k9s is at 146 Watts both of these results are higher because the CPUs are completing the test before the duration expiry so unless running this test back-to-back and rapid succession the result will be more of the fully boosted power matric blender shows lower figures because it's a longer test and we wait to see what the steady-state power consumption is that's the real difference between the Cinebench test and blender in this test the 9900 KS at 5.2 gigahertz and 1.3 to 3 volts draws 234 watts which is actually pretty good for a 5.2 gigahertz clock thanks to those lower voltages the 3900 X at 4.3 gigahertz is running 173 watts with the stock 3900 X at 148 watts and close to the stock 9900 K original CPU we're going to pull up one last older chart this one doesn't have the modern AMD parts on it but it has an older Intel and previous gen AMD parts that can be worth looking at this gives a representation of a CV to heavy workload in a gaming environment using a heavy physics simulation as the low generator we're able to see what the CPUs look like when they're doing something other than tile-based rendering in this scenario the 9900 KS is pulling 144 Watts when stock the 9900 K is at 127 Watts when stock and the 9900 K with a 5.1 gigahertz 1.41 volt o C is at 210 watts the 99 hundred KS can hold a 100 megahertz higher frequency at 1.3 to 3 volts a big drop in voltage requirement for a 180 watt power reading that's a reduction in power consumption of 14 percent versus the technically slightly slower 5.1 gigahertz 99 K original sample and it speaks to how much intentional bidding can help with power consumption and with voltage moving onto some gaming benchmarks next civilization 6 will start us out for this one we're looking at average AI player turn time rather than the usual FPS metric each second increase in turn time becomes palpable as player count increases so this can become significant as campaigns thrown on and the game gets more complex for this one the 1900 KS stock CPU managed a 29 point 1 second average turn completion time which has it right around the 9700 K at five point one gigahertz within 9,900 case stock CPU not the s not far behind at twenty nine point seven seconds keep in mind two things one tests are imperfect and have some variance but sieve is really consistent and we've controlled it as best we can to and more importantly Civilization six really likes frequency this most evident when looking at these sets of results just mentioned with the 9900 k stock sieve you can leverage its limited thread boosting frequencies to climb the charts since the cores aren't as heavily and constantly loaded overclocking the que ésta 5.2 gigahertz landed it at twenty eight point five seconds for average turn completion time it turned time reduction from the stock chaos of two percent the stock ninety nine hundred K at five point one gigahertz was holding 29 seconds average sitting right behind the five point two gigahertz at ninety nine hundred KS GTA 5 at 1080p is up now at 1080p the Intel I $9.99 hundred K stock CPU originally landed at 1:22 FPS average with lows at about eighty eight and seventy five the stock and ninety nine hundred KS configuration improved about six point six percent over baseline at 131 FPS average from 120 to the 1900 Kaden on s at five point one gigahertz our previous stable OC at one point four volts managed 132 FPS average lows between the two are functionally identical and are within error for lows which is a wider error the 1900 KS at five point two gigahertz hit 135 FPS average a boost of two point seven percent over the five point one gigahertz results for reference the r9 3900 AK stock CPU runs at 110 FPS average allowing the 99 hundred KS stock CPU a lead of nineteen percent over the r9 3900 ax when both are stock the 3900 X at four point three gigahertz all core with SMT on hit 113 FPS average but that doesn't move the needle much when we're comparing to overclock versus overclocked numbers disabled in SMT and screwing around with things like that helps but we don't recommend actually disabling SMT because then you're thrown away half of what you bought the CPU for so it's just a technology demonstration not something we recommend for GTA 5 at 1440p the 99 hundred KS maintains its positioning and the hierarchy remains largely unchanged the KS ends up at 132 FPS average when I overclocked to the coolers limits that's our limitation here or 129 FPS average one stock the RO 939 hundred X remains around 108 FPS average keeping about the same distance as before as for the 9900 k stock CPU non ass it's at 122 FPS average this same as 1080p and that's because it's CPU limited here not GPU limited so this illustrates that even at a high resolution until you really start pushing something like 4k the CPU can play a role in performance the KS is about six point three percent better stock to stock although overclocking both equalizes them albeit with a higher voltage for the non ass CPU the campaign benchmark for total war Warhammer - had the 9900 KS at 189 FPS average stock the highest result for any stock CPU currently on this chart the overclocked in 9900 K did slightly better with 5.1 gigahertz across all cores vs. 5.0 for the stock 9900 KS that's with the uncor set to 45 that was at 191 FPS average to 189 while the 5.2 9900 K s predictably did best at 196 FPS average it's impressive but this is also a test that doesn't care much for extra cores the eight core eight thread 9700 K averaged at 179 fps without any overclock at all and it's a great deal cheaper and these highest score on the chart is the 3900 X with a 4.4 gigahertz overclock and SMT disabled not a configuration we recommend but a good test and that brought it to 171 FPS average f1 20:18 gives us a dx11 title to look at but we'll move back to dx12 next for this one the 900 K and KS have finally found our GPU bottleneck at the top and the 900 KS at 5.2 gigahertz and 900 K s stock we're both stuck at 320 FPS average which happens to me where the 20 ATT I can no longer keep up with the CPUs this is all at 1080p mind you so it'll only get more GPU bound as the resolution goes up we can't really draw firm conclusions here since a limit imposes a barrier that cuts off the total gains possible from the CPU upgrade but the night I heard KS manages to outstrip the nine hundred K stock CPU by a few FPS we just don't have enough test resolution or Headroom here to define the absolute gains the r9 3900 X for reference runs at about 276 FPS average stock that's plenty clearly that's that's a lot of frames but if you're more interested in relative scaling it's about a 16% jump to the GPU bound 900 KS the next chart demonstrates our hard GPU bottleneck if you're curious in this instance if you're playing at higher resolutions the GPU fast become a limiter that can act as a great equalizer for higher-end CPUs the 9600 K 700 K 900k 700 K there sorry nine hundred KS can't mix those up we're all functionally equal in this chart the difference is you're seeing those plus or minus 3 FPS swings are not actual determiners of performance but rather normal test variance from bouncing off of a GPU limit and averaging a more random set of numbers because we're limited by another device the 3900 X gets closer in instances of GP limitations at the top-end naturally and closes at 2:30 FPS average it's still behind at 1440p though moving up in resolution further might begin to lock the tooth heightens warrants that hitman 2 is up next tested in dx12 and first at 1080p the nine hundred KS stock runs at 142 FPS average out of the box with Louis at 77 and 47 1% a 0.1% lows respectively this puts the KS at 4% over the K stock CPU with hyper-threading on overclocking the KS puts it to 144 FPS average about a 5% gain for the increase in all core frequency and uncor although the latter does last for framerate the night I heard K hyper-threading off result hyper-threading disabled result helps demonstrate that we still have Headroom before hitting GP limitations as for the r9 3900 X the stock result runs at 121 FPS average allowing the 9900 KS a lead of about 17.6% over there are nine 3900 acts on this workload shadow the tomb Raider's next another DX 12 title that's fairly decent for for testing and it's new that is 500 KS stock CPU ended up at 172 FPS average of lead of about 1% over the united case stock cpu and this charge the KS at 5.2 did little for performance a mixture of nearing GPU limiters and diminishing returns on cpu clock into this title so it's a bit of both let's throw a Dobby premier in here to Adobe premier will give us another look at a production workload accompanying our earlier blender results for this one the r9 3900 ax is back in the driver's seat leading the pack in a 1080p show floor render with a 3.4 minute render time requirement to get the job done the 900k stock CPU required 3.8 minutes allowing the 3900 X an advantage of 10.5% the united 900k asks the asus board doesn't really improve much moving to three point seven minutes required if you did this with the gigabyte board or enabled MCE then yeah I'd be faster but that's that's cheating this is not running stock overclocking the 5.1 gigahertz helps a bit more but it's still not in the lead in this instance the 3900 X is holding its own for 4k render results it's more or less the same but the scale stretches a bit the difference is the 1950 X sort of closets way into the results and displaces the 9900 K by another rank the nine hundred KS ends up at 11.15 minutes for this render with the 9900 K at eleven point nine minutes we're gonna cap it there we do technically have a whole lot more production and synthetic results but frankly should just look at one of our reviews with the night ahead of K in it because it's basically that when we overclock it and there's no point in adding 15 more minutes to the charts to say that also it is presently four hours away from embargo lifts and I need to edit and upload this in time to hit embargo lifts so we're gonna cap it there we're planning to revisit this chip with liquid nitrogen overclocking just for some fun see how it scales it should be better than our 900 K in theory it's Bend and then as for whether you should buy this first of all let's address something the night on our K is objectively a good processor the r9 3900 X is objectively a good processor where they're good depends a bit on which one you're talking about then I heard K and the KS because it's the same thing are good in gaming they are topping the charts you cannot presently get faster than those based on our charts HDD T CPUs don't really add value here in fact a lot of them will drag down the performance if you go to a 99 80 XE and probably a 10 9 8 exe you're very likely going to end up in a lot of games with less performance than a higher frequency lower core count 8 core 900 K so the 9900 K and the KS are at the top of the charts for gaming and that's really all there is to that that doesn't mean the 3900 X is bad or that you shouldn't buy it the 3900 X is at the top of the charts for things like blender it is also impressively good and Adobe Premiere and and that's frankly not something we really expected out of this generation from AMD didn't even expect together frequencies at got out of Zen too so that was pretty cool to see the way it breaks down then is the same as before nothing's really changed it's just a bit faster now if you're buying a gaming only building a gaming only computer and you really really care about high frame rate and you're using the highest end possible GPU and you don't really care about things like blender rendering with cycles on a CPU it's a common use case but it's not that common or you don't care that much about premiere or maybe you care a little bit about from here but you care more about gaming and deny outer case tool is one premiere anyway so you weigh them and gaming comes out ahead whatever the case may be the Niantic K has those use cases in mind the 3900 X is a better choice if you have $500 to spend and that's what you're planning to spend for something like blender for tile based rendering for v-ray for decompression with 7-zip if those things sound appealing to you you should buy a 3900 X and if gaming at super high frame rates isn't that important maybe you play games sometimes but you don't care if it's 16 17 percent well that maths funny but even a 900 K is 16 or 17 percent ahead of a 3900 X if that doesn't matter to you then 3900 X is fine so you have to very well-defined options here this is not us being wishy-washy this is not us waffling between them we've given you two very firm recommendations depending on what your workload is and now it's up to you to figure that part out and choose based on that also as a reminder there are a lot of very good CPUs that aren't $500 and are cheaper and you don't have to come up this high in price to get something that's good for gaming or for production see also r5 3600 and well for just about everything really or i7 9700 K for gaming so that should wrap us up KS is a K just with its Bend and you'll get better frequency voltage curves VF curves out of it I'm just repeating myself at this point if you jumped here you want to tell the our then go away go back to the beginning and watch the whole thing not gonna recap it any further than that thanks for watching subscribe for more go to store that game is exes dotnet to support us directly like by buying one of our mod mats the large ones are in stock and shipping now we just received the most recent batch so they're no longer on backward thank you for your patience and you go to patreon.com/scishow hearings Nexus you would like to support us there as well have a lot to edit I'll see you all next time\n"