The Most Powerful PC Gaming Hardware Known To Man
### Article: Building the Fastest Gaming PC Hardware Setup: A Deep Dive into High-Performance Overclocking
#### Introduction
People who build their own gaming PCs often strive to find the perfect balance between cost, performance, and functionality. However, this video is about none of that—it’s purely focused on achieving the highest tier of performance using the fastest gaming PC hardware available. Practical considerations like budget, power draw, temperatures, and aesthetics take a backseat in this pursuit of ultimate speed.
The goal here is to explore how far we can push modern hardware to achieve the best possible scores in 3D Mark Port Royal, a synthetic benchmark that tests GPU and system performance. This article will walk through the entire process, from selecting components to testing and overclocking, using the latest high-end hardware.
---
#### The Hardware Setup
The centerpiece of this build is the **NVIDIA GeForce RTX 3090 Ti** graphics card. Known for its exceptional performance, the RTX 3090 Ti was tested in a two-way SLI configuration. To ensure compatibility and optimal cooling, we used mismatched cards—one from MSI and another from Palit—and faced challenges with sli bridge alignment.
To overcome these issues, **PCIe 4.0 riser cables** were utilized to position the cards further away from the motherboard. However, the initial motherboard (ASUS ROG Maximus Z690 Hero) lacked SLI support, leading us to switch to the **ASUS ROG Maximus Z690 Extreme**, which supports four-slot spacing and Intel’s 12th Gen CPUs like the **Intel Core i9-12900K**.
For AMD enthusiasts, the **Ryzen 7 5800X 3D** was also tested as a contender. This CPU features 3D V-Cache technology, which theoretically enhances gaming performance by improving memory latency.
The storage solution included **Samsung 970 Evo SSDs** and **MP600 Pro XT NVMe drives**, while the cooling system relied on **Corsair H150i AIO coolers** to keep temperatures in check.
---
#### Testing and Overclocking
Testing was conducted using **3D Mark Port Royal**, a synthetic benchmark that’s particularly demanding for high-end systems. The focus was on achieving the highest possible score on the leaderboard, which requires careful overclocking and optimization of system settings.
Key steps included:
1. **Baseline Tests**: Running the system at stock settings to establish a baseline score.
2. **Overclocking the GPU**: Using EVGA Precision X1 software to boost GPU frequencies while monitoring temperatures and voltage.
3. **Memory Overclocking**: Pushing memory speeds beyond default limits to extract more performance.
4. **Fan Speed Adjustments**: Controlling fan speeds to maintain optimal card temperatures for stable high-frequency operation.
The system was also tested in a garage environment, with ambient temperature playing a significant role in GPU performance. To mitigate this, the setup was positioned near an AC unit, and fans were run at maximum speed.
---
#### Results and Achievements
After extensive testing, the **Intel i9-12900K** configuration achieved a top score of **30,930**, placing it in the **top 17 on the Port Royal leaderboard**. The AMD Ryzen 7 5800X 3D, while slightly behind, still performed admirably with scores around **30,665**.
Notable achievements included:
- Breaking the **30k mark** for the first time in baseline testing.
- Overclocking the RTX 3090 Ti to run at frequencies between **2145 MHz and 2175 MHz** during peak performance.
- Dialing in memory speeds of up to **1,421 MHz** for optimal GPU performance.
---
#### Challenges and Lessons Learned
Several challenges were encountered during the build:
1. **Sli Bridge Misalignment**: The RTX 3090 Ti cards had mismatched sli fingers, requiring riser cables and careful alignment.
2. **Motherboard Limitations**: Initial testing was hampered by the lack of SLI support on the Z690 Hero motherboard.
3. **Overclocking Balance**: Finding the sweet spot between voltage, frequency, and cooling to avoid instability.
Despite these hurdles, the project demonstrated that with the right hardware and optimization, consumer-grade systems can achieve professional-level benchmark scores.
---
#### Future Plans
The success of this build has sparked interest in future projects. With the release of new hardware later in the year, there may be opportunities to revisit this setup with updated components. The focus remains on pushing performance boundaries while maintaining practicality.
---
#### Conclusion
This article has provided an in-depth look at building and testing a high-end gaming PC capable of achieving top-tier scores in 3D Mark Port Royal. By leveraging cutting-edge hardware, careful overclocking, and innovative cooling solutions, we’ve demonstrated what’s possible with today’s technology.
Whether you’re aiming for the leaderboard or just want to build a powerhouse gaming rig, this guide offers valuable insights into maximizing performance. Stay tuned for more updates as new hardware becomes available!
"WEBVTTKind: captionsLanguage: enpeople who build their own gaming pcs are often looking for that perfect balance between the amount of money they have to spend the amount of power they want out of their system what they're using it for if it's gaming or mixed use and then other factors might come into play like power draw temperatures and of course aesthetics this video is about none of that though because this video is about the fastest gaming pc hardware that is currently on the market and at that highest tier of performance a lot of those practical considerations just go out the window starting with budget a couple weeks ago i already tested the fastest gpu configuration that's on the market nvidia geforce rtx 3090 ti graphics cards in two-way sli you guys really enjoyed that video and you asked me for a follow-up so that's what this video is all about but this time i got some extra help excellent today's video is brought to you by kyoccia's family of nvme ssds featuring their latest bix 3d flash memory the bg5 now supports pcie 4.0 and is still available in the incredibly small m.2 2230 form factor so it's a great fit for gaming pcs laptops and compatible handheld gaming devices and for enterprise or hyperscale data center use the cd7 and xd6 feature the new edsff or enterprise and data center standard form factor for ease of integration while the cd7 supports pcie gen 5 for maximum performance when paired with the latest amd epic or intel xeon server hardware for more on keoxia ssds click the sponsor link in the video description so back at the end of march nvidia launched what many of us assume is going to be the last card in the rtx 30 series the geforce rtx 3090ti we're also assuming that that's going to be the fastest card from the current generation that they're coming out with and i happen to end up with two of them i have the msi supreme x card which i tested in my launch review and then i also got this palette game rock version which is very blingy and shiny and reminds me a lot of a g skill trident z royal rgb kit i have a build planned for this palette card as part of my how to build a pc series which i am revamping in 2022 but part of the reason i didn't get that series underway in april is because i was distracted by the fact that i had two of these cards and even though two-way sli configurations are mostly deprecated by nvidia they do not receive the same amount of support that they used to and with the 30 series nvidia limited sli support to only the top card in the stack the rtx 3090 and i already tested the rtx 3090 in two-way sli as part of the rip j rip gn series that i briefly participated in going as far as to do ice water bath cooling and there's a whole video series on that if you're interested so my first thought with the ti now is how easy would it be to beat my own score from back when we attempted that in 2020 and i'm not trying to go after jay or gamer's nexus again this time around or reboot that little rivalry because let's be honest once they moved into the exotic cooling methods such as ln2 pretty much i wasn't really able to continue competing also if we're looking at the benchmarks that i'm going to be running today which is 3dmark port royal i'm not even planning to get too high up the leaderboard because at the top of the leaderboard you have professional overclockers who spent extensive amounts of time with stuff like lm2 to get the scores that they achieved so yes my focus today is on a new personal best and if you checked out the video from a couple weeks back you might have seen that i was down at the very bottom of the leaderboard i was actually in spot 100 but with this configuration right here i was able to beat my 30 90 score and i got up into the top 50. but many of you asked for a follow-up on that video and that's for multiple reasons a lot of that video was about troubleshooting because i'm using two mismatched cards and i was limited to the hardware that i had on hand and while i do have a lot of hardware on hand what i actually needed was a motherboard that had four slot spacing between the two primary pci express slots that the graphics cards slot into prior to now three slot spacing has been just fine but the 39 dtis are pretty big boys they're fairly thick as we like to say and if we're going beyond a three slot cooler and many of these cards do have three plus slot coolers if not full four slot coolers then you need the space to be able to plug those into the motherboard and since i only had three slot spacing motherboards i ended up using these pcie 4.0 riser cables to position the cards a little bit further away from the board and that also helped me get around the sli bridge misalignment problem that i had because the sli fingers on these two boards are actually positioned differently so the bracket on this top card is actually about a centimeter further out this way than the bracket on the lower card so even if i had a board with proper four slot spacing i couldn't plug these two cards in with an sli bridge without using these riser cables because the sli bridge simply wouldn't line up with the fingers but even with these pci express riser cables and this gaff tape and wood block assembly that i put together to try to get a little bit more structural support and my perfectly engineered plastic block framework that i put together to hold these up in the right position what i actually discovered is that this motherboard i was using the asus rog maximus z690 hero which is a 600 motherboard and part of the rog line actually does not support sli it does not have the proper sli keys for it to work so when i finally did get it working on my amd testbed i only had a very limited amount of time to run a few tests and i did get some success like i said i jumped into the top 50 but i didn't have a lot of time to do benchmarking and overclocking so what's going to be different this time around first of all despite my vocal complaints about the z690 hero not supporting sli asus was kind enough to send over the maximus z690 extreme which is a crazy overbuilt board it's super heavy it has rgb scattered throughout it costs 1100 us dollars but it also supports the intel 12th gen core processors and has four slot spacing between the two by 16 pci express slots for graphics cards that meant i could bust out some new cpus and on the intel side we now have the 12900 ks which is a slight upgrade from the 12 900 k intel's charging 800 bucks instead of 600 bucks for it so it's really not the most practical cpu for most people but like i said at the beginning of this video we're not all about practicality here we're about getting a bit of extra performance no matter what the cost and with a 5.5 gigahertz turbo boost this cpu will run faster without any tweaking or tuning from me and that is a good thing i will note that when i first installed the cpu we did have some issues getting booted up but thankfully asus includes usb bios flashback and most of their motherboards these days from my understanding that allows you to update the uefi or the bios and you don't even need a cpu or memory installed so i did that and then the system booted up just fine and recognized our 12 900 ks but another contender also entered the ring since i did my initial video on the 30 90 ti sli and that is the amd ryzen 7 5800 x 3d which is now amd's fastest cpu for gaming and that's all due to the addition of 3d v cache other than that it's just a 5800x which actually runs a slightly lower frequency than the original 5800x so the question is does the 3d v cache help in these synthetic benchmarks that i'm running because these 3dmark synthetic benchmarks are some of the only somewhat practical uses for 3090 ti sli setups these days anyway and for that purpose evga sent over their x570 dark motherboard this is one of evga's first x570 or amd motherboards and they did an amazing job on it it's designed for overclocking so it has some unique features like a rotated cpu socket only two memory slots but they're positioned as close to the cpu socket as possible which is helpful for memory overclocking a host of other features on this board as well evga always does a great job designing motherboards specifically for overclocking and hitting those world records and yes four slot spacing between the two pcie slots which means i should hopefully not have to use those riser cables whether i want to test the intel platform or the amd platform but you're probably thinking paul you just showed us how the sli fingers on these cards do not line up so even with four slot spacing i still couldn't plug both of these cards in and sli them up without using the riser cables and the answer there is yes you're absolutely correct and that is why evga sent other stuff too in their little care package and uh two of the things were graphics cards these are evga for the win rtx 390 ti's and these actually are slightly different cards even though they look very much the same they have the same pcb design so the sli fingers on them will line up but one of these is an ultra card which has a little bit of an overclock and one of these is a black card which runs at a slightly lower frequency it's a difference of like 60 megahertz or something like that but this means that first i have like 6 500 or so worth of graphics cards stacked here in front of me which i'm only slightly nervous about but it also means that i can plug both of those cards into my four slot space motherboards use an sli bridge hopefully get rid of a lot of those complications so i can just focus on getting some benchmarks run and evga just within the past week or so over on their forums has posted has posted xoc or extreme overclocking v-biosis for these cards that upped the power limit from 107 to 115 so that should allow us to overclock that much more and so here are the two test beds for my initial comparison for intel we have the 12 900 ks cpu of course the asus rg maximus z690 extreme motherboard and a g-skill trident z5 rgb 32 gigabyte memory kit that's ddr5 6000 speed and for storage we have a 500 gig samsung 970 evo ssd the ssd shouldn't really affect performance but just wanted to include that for our amd testbed we have the 5800x3d cpu of course the evga x570 dark motherboard a g skill trident z royal rgb 32 gig ddr4 4400 memory kit but i did dial back that speed to 3800 so that the infinity fabric and the memory could be running at the same frequency 1900 megahertz which is kind of where infinity fabric tops out and then the ssd is of course their mp600 pro xt 2 terabyte uh i did have to remove the heatsink that came with it in order to install it on this motherboard but it still worked out on both test beds i have a corsair h150i elite lcd 360 millimeter all-in-one liquid cooler and the power supply is a corsair ax1600i 1600 watt 80 plus platinum power supply which should have more than enough wattage for our needs but i think i will be switching to a two power supply test setup at some point as well of course it still wasn't just a simple plug everything and fire it up situation this test bed over here is a liam lee pct 60 which i've had for quite some time but it has a handle that goes across the top and that can often block my access for installing and removing graphics cards so i usually just don't have it there the aluminum piece that sticks up on the side here though which is supposed to attach to the handle was actually conflicting with the four slot second graphics card so i busted out the rotary tool this is actually something i've been meaning to do for quite some time just sort of shaved it off bent it off filed it down a little bit so there weren't any sharp edges and then just slap some gaff tape on there so it has a slightly cleaner finish so after all that exposition let's finally get into some testing and my first test were just some baseline tests just running everything at stock with the memory xmp settings as already described and using the steam version of 3dmark and here the 5800x3d scored 27938 in 3dmark port royal whereas r12 900 ks hit 28 438. the amd system was pulling about 1100 watts on average the intel system was around 1150 and here i made a somewhat controversial decision so you guys can let me know in the comments whether or not i was right about this because the amd score was about 500 points less i focused the rest of my testing on the intel test bed running the 12900 ks but i noticed after the fact that the average clock speed on the amd test was 1970 whereas the average clock speed on the intel test was 2004. so i am yet again in a situation where it's totally possible that with further testing we might be able to suss out more details between the two platforms but since my goal was getting higher scores on the 3dmark port royal leaderboard so that is where i focused my efforts and back to the 12900ks testbed my first just a little bit of an overclock uh boosted the power limit up to 107 plus 200 on the memory plus 100 on the gpu we're using evga precision x1 by the way to dial in these overclock settings and that resulted in a score of 29 574 which is just ever so slightly below my top score from video number one on this subject which was 29 579 but it was an improvement nonetheless and of course there's still plenty of headroom available for me so my next steps were to dial in some optimizations some special settings in the nvidia control panel that i covered in my rtx 3090 sli testing from back in the day so i'll link to that video if you guys are interested and at this point i also switched to the standalone version of 3dmark which removes some of the overhead of steam running in the background and that boosts the score up to 29 794. so there was nice to see some quick success and it was definitely faster to get to this point with this setup than it was with the setup from two or three weeks ago but here i'll also point out that i've been running these tests out here in my garage and ambient temperature is a bit of a concern since we want the gpus to run as cool as possible so i was trying to do this testing late at night after the temperature had gone down by 10 or 20 degrees and i also had the system set up on a folding table positioned over on the side of the garage that's near my little standalone ac unit so i could just have that pointed at them running full blast the whole time but my best score on this first night of overclocking was thirty thousand and twenty so i actually broke 30k which seemed like a nice achievement a nice sort of high water mark to get past and according to 3dmark the top gpu or the faster gpu in this test was running it between 2145 and 2175 megahertz so the next night i tried to take advantage of a few more of these levers that you have to try to get more performance out of a card when you're overclocking there's the overclock itself which we are controlling with evga precision x1 software that can allow you to boost up the gpu frequency the memory frequency it can let you add a little bit more voltage you can increase that ever important power limit and you can also ramp up the fan speed because as i've mentioned multiple times the lower the temperature on the cards the higher the frequency they can usually run at stably and that is part of that second lever that you can apply some force to and that is just temperature and you can do that in multiple ways upping fan speed is one of them pointing an air conditioner at your setup is another and then of course there's the more exotic cooling solutions like water cooling which is just more effective in general you can add an ice bath to the water cooling or you can go to the most extreme and you can use ln2 liquid nitrogen or even liquid helium the final lever is these small tweaks you can make to the system itself to try to eke a little bit more performance out so i already mentioned using the standalone version of 3dmark is one of those there are nvidia control panel settings that you can mess with here such as switching to emphasize performance over quality and little tricks that you might not have thought of like swapping the v bios on the card for something that allows you to up the power limit so that was the thing that i did next in my mind i wanted to take the for the win ultra xocv bios and apply it to both cards so they'd be both running at the same frequency out of the gate evga's installer of course prevents you from doing that so i just used the xoc ultra v bios for the ultracard and the xoc black v bios for the black card in both cases the gpu z settings seemed to be exactly the same it didn't change frequency or anything like that you can notice the two different v-bios versions looking at that there but practically speaking the only change is that in evga precision x1 the power slider goes up to 115 instead of 107. i also switched to using two power supplies for these tests and i think it's likely that i would have been just fine running a single 1600 watt power supply because like i said we weren't going up beyond about 1200 watts with the whole system but i kept the gpus via the three-way power adapter dongle here all routed through a single ax1600i power supply and then i used the second 1000 watt power supply to power the rest of the system with that configuration and the 115 power limit the two graphics cards by themselves were pulling around 1025 to 1050 watts but from there i went back to overclocking with precision x1 i went to a plus 110 for the gpu plus 250 for the memory that 115 power limit fans set to 100 and with that configuration we improved the score a little bit to 30 085 just a small bump over the 30 020 from the night before and our gpus are running at between 2160 and 2190 megahertz at least again on the top graphics card i bumped up from there for one more pass to plus 115 on the gpu plus 300 on the memory added a little bit more voltage and there i got a score of 30 264 and that was enough to put me in the top 25 on the port royal leaderboard and and that felt pretty good an improvement from position 100 to sub 50 in video one and then from position 45 or so down to top 25 again some nice improvements there and good to see i was making progress but the thing that was on my mind next was again the fact that i'm using two slightly different graphics cards that are set to run at slightly different frequencies so i uncoupled the gpus in evga precision x1 that allows me to adjust settings for each of them individually rather than just trying to use the same settings for both cards i basically said it so the boost clock on the black card was the same as the boost clock on the ultra card but my score went down from there to just like thirty thousand and five but that's just how things often go when you're overclocking and trying some unique configurations like this sometimes you'll try something and the score goes up sometimes you'll try something and the score goes down from there since again i was running these tests late at night i just was started trying out different things in my documentation of what exactly i was doing got a little bit muddy i did link the gpus back up again and then here i started attacking the memory frequency a bit more because the memory on these cards is supposed to have a lot more headroom than the previous cards especially with better cooling on them so whereas before i was at like plus 300 or so on the memory i bumped it up to plus 868 and there i managed to score of 30 453. the memory effective frequency was 1 421 megahertz for that run but the best scores that i achieved for the evening apart from all the other stuff that i already talked about was using decoupled gpu settings in precision x1 that allowed me to overclock the second black card just a little bit more to get the frequency not quite up to the same level as the ultra card but to give it a little bit more boost to get it in the same ballpark so i was at plus 115 on the top card and plus 145 on the second card the memory was at plus 55 and i will say give or take here because i ran several runs it was getting quite late at night i was getting pretty tired and one of the other things you do to try to eke out a little bit more score is disabling your internet or turning off wi-fi when you run the test which means when the test pops up it says it's not a valid score until you reconnect to the internet and then validate your score so if you look at the 3dmark port royal leaderboard right now you will see that i broke the top 20. in fact i'm in the number 19 position with a score of 30 635 better than 99 of all results of course and here's a quick look at some of the settings like the memory frequency was a 1424 average 1421. average gpu temperature is 55 degrees celsius and again that is because i had the air conditioner pointed at the setup i didn't go so far as to create a cardboard shroud like i've done in the past so that could again maybe lower temperature a little bit more if i went that far but here's what it is for now our average clock frequency was 2160 and it peaked at 22.05 and there's the 1200 ks and a few other stats i also didn't go in and like disable the e cores or disable hyper threading that's another thing that's recommended sometimes but this is all to show that no matter how much time i throw out a project like this i'm always like ooh if i can just a couple more hours to do this and this and this maybe i could bump that score up just a little bit more but practically speaking i'm also making a video for my youtube channel so i do have to cut this off at some point so i can present a video to you guys so that's why i got as far as i did today but that actually isn't my top score 30 635 like i said i was disconnecting from the internet to do these runs and then reconnecting to validate my score and somehow i ran two tests scoring 30 764 and then my best run at 30 820 and apparently i did not hit save to save those results locally on the system and i did not reconnect to the internet to hit validate so it would upload those to my 3dmark account so i did not upload those scores to 3dmark so that they could be proudly shown on my account which i'm slightly disappointed about practically speaking that would have bumped me up i guess maybe two spots here so i'd be in position like 17 instead of position 19 so it's not the end of the world okay hold on i couldn't leave it there i had to do one last night of testing because i just couldn't stand it with that score i wasn't able to upload so i made some additional runs with the 12 900 ks and basically the same settings i already showed you guys scored 30 741 then managed to bump up to 30 930 ran a few more tests after that got around thirty thousand eight hundred thirty thousand nine hundred and it's pretty clear that i'm hitting a ceiling here in terms of performance without further changes to the configuration but the overclock settings i was using was plus 965 on the memory plus 119 on gpu one plus 155 on gpu 2 voltage at plus 40 power target to 115 of course and the fans set to 100 those are the settings i used for my top score of 30 930 and that has bumped me up to position 17 on the 3dmark port royal leaderboard and that's pretty respectable i feel like without the use of ln2 or water cooling even but i felt like i wasn't giving the 5800 x 3d a fair shake unless i tried this configuration on it as well so i transferred everything back over to the x570 test bed same exact overclocking settings with precision x1 and there i scored 30 561 on one run and then 30 665 on the next run so we're still about 300 points behind the 12900 ks setup and yes i did double check and verify that the frequency was good here it was running about 2205 to megahertz but i'm very glad that i took another pass at this because it let me jump up the leaderboard just a few more spots and it let me confirm that the 5800x3d is still just slightly behind the 12900 ks at least in this specific benchmark but it is the end of this video because i got to cut things off at some point i went from position 100 to top 50 top 25. now i am top 20 and i am somewhat proud of that fact although let's be honest most of it was the hardware i was using and a big big thank you to evga for lending me these two 30 90 tis i will be needing to return these cards but evga did say that i could hang on to the x570 dark and i believe i can also keep this evga special envy link sli bridge that they sent as well which is a pretty cool looking one my standard nvidia one has been through a lot and has gotten some scratches and stuff so it was nice to complete the look with the setup with this evga bridge but i'd be very interested to hear your feedback on this project did i take it far enough should i push it farther should i do a part three video i'm probably not gonna do a part three video but who knows perhaps this is just a very nice refresh for me in terms of high-end overclocking so that maybe later this year when more hardware comes out that we've been talking about and anticipating for quite some time we could maybe do it again and we can see how things go there but if you enjoyed this video definitely hit the thumbs up button and let me know i'll post links to the hardware i was using and stuff down in the description below you can also find a link to my store down there at paul's hardware dot net where you can buy shirts mugs pint glasses and other high quality merchandise to help get yourself some high quality merchandise and support my channel at the same time thank you guys so much for watching if you'd like to see more videos like this one in the future then don't forget to subscribe to my channel as well we'll see you all in the next videopeople who build their own gaming pcs are often looking for that perfect balance between the amount of money they have to spend the amount of power they want out of their system what they're using it for if it's gaming or mixed use and then other factors might come into play like power draw temperatures and of course aesthetics this video is about none of that though because this video is about the fastest gaming pc hardware that is currently on the market and at that highest tier of performance a lot of those practical considerations just go out the window starting with budget a couple weeks ago i already tested the fastest gpu configuration that's on the market nvidia geforce rtx 3090 ti graphics cards in two-way sli you guys really enjoyed that video and you asked me for a follow-up so that's what this video is all about but this time i got some extra help excellent today's video is brought to you by kyoccia's family of nvme ssds featuring their latest bix 3d flash memory the bg5 now supports pcie 4.0 and is still available in the incredibly small m.2 2230 form factor so it's a great fit for gaming pcs laptops and compatible handheld gaming devices and for enterprise or hyperscale data center use the cd7 and xd6 feature the new edsff or enterprise and data center standard form factor for ease of integration while the cd7 supports pcie gen 5 for maximum performance when paired with the latest amd epic or intel xeon server hardware for more on keoxia ssds click the sponsor link in the video description so back at the end of march nvidia launched what many of us assume is going to be the last card in the rtx 30 series the geforce rtx 3090ti we're also assuming that that's going to be the fastest card from the current generation that they're coming out with and i happen to end up with two of them i have the msi supreme x card which i tested in my launch review and then i also got this palette game rock version which is very blingy and shiny and reminds me a lot of a g skill trident z royal rgb kit i have a build planned for this palette card as part of my how to build a pc series which i am revamping in 2022 but part of the reason i didn't get that series underway in april is because i was distracted by the fact that i had two of these cards and even though two-way sli configurations are mostly deprecated by nvidia they do not receive the same amount of support that they used to and with the 30 series nvidia limited sli support to only the top card in the stack the rtx 3090 and i already tested the rtx 3090 in two-way sli as part of the rip j rip gn series that i briefly participated in going as far as to do ice water bath cooling and there's a whole video series on that if you're interested so my first thought with the ti now is how easy would it be to beat my own score from back when we attempted that in 2020 and i'm not trying to go after jay or gamer's nexus again this time around or reboot that little rivalry because let's be honest once they moved into the exotic cooling methods such as ln2 pretty much i wasn't really able to continue competing also if we're looking at the benchmarks that i'm going to be running today which is 3dmark port royal i'm not even planning to get too high up the leaderboard because at the top of the leaderboard you have professional overclockers who spent extensive amounts of time with stuff like lm2 to get the scores that they achieved so yes my focus today is on a new personal best and if you checked out the video from a couple weeks back you might have seen that i was down at the very bottom of the leaderboard i was actually in spot 100 but with this configuration right here i was able to beat my 30 90 score and i got up into the top 50. but many of you asked for a follow-up on that video and that's for multiple reasons a lot of that video was about troubleshooting because i'm using two mismatched cards and i was limited to the hardware that i had on hand and while i do have a lot of hardware on hand what i actually needed was a motherboard that had four slot spacing between the two primary pci express slots that the graphics cards slot into prior to now three slot spacing has been just fine but the 39 dtis are pretty big boys they're fairly thick as we like to say and if we're going beyond a three slot cooler and many of these cards do have three plus slot coolers if not full four slot coolers then you need the space to be able to plug those into the motherboard and since i only had three slot spacing motherboards i ended up using these pcie 4.0 riser cables to position the cards a little bit further away from the board and that also helped me get around the sli bridge misalignment problem that i had because the sli fingers on these two boards are actually positioned differently so the bracket on this top card is actually about a centimeter further out this way than the bracket on the lower card so even if i had a board with proper four slot spacing i couldn't plug these two cards in with an sli bridge without using these riser cables because the sli bridge simply wouldn't line up with the fingers but even with these pci express riser cables and this gaff tape and wood block assembly that i put together to try to get a little bit more structural support and my perfectly engineered plastic block framework that i put together to hold these up in the right position what i actually discovered is that this motherboard i was using the asus rog maximus z690 hero which is a 600 motherboard and part of the rog line actually does not support sli it does not have the proper sli keys for it to work so when i finally did get it working on my amd testbed i only had a very limited amount of time to run a few tests and i did get some success like i said i jumped into the top 50 but i didn't have a lot of time to do benchmarking and overclocking so what's going to be different this time around first of all despite my vocal complaints about the z690 hero not supporting sli asus was kind enough to send over the maximus z690 extreme which is a crazy overbuilt board it's super heavy it has rgb scattered throughout it costs 1100 us dollars but it also supports the intel 12th gen core processors and has four slot spacing between the two by 16 pci express slots for graphics cards that meant i could bust out some new cpus and on the intel side we now have the 12900 ks which is a slight upgrade from the 12 900 k intel's charging 800 bucks instead of 600 bucks for it so it's really not the most practical cpu for most people but like i said at the beginning of this video we're not all about practicality here we're about getting a bit of extra performance no matter what the cost and with a 5.5 gigahertz turbo boost this cpu will run faster without any tweaking or tuning from me and that is a good thing i will note that when i first installed the cpu we did have some issues getting booted up but thankfully asus includes usb bios flashback and most of their motherboards these days from my understanding that allows you to update the uefi or the bios and you don't even need a cpu or memory installed so i did that and then the system booted up just fine and recognized our 12 900 ks but another contender also entered the ring since i did my initial video on the 30 90 ti sli and that is the amd ryzen 7 5800 x 3d which is now amd's fastest cpu for gaming and that's all due to the addition of 3d v cache other than that it's just a 5800x which actually runs a slightly lower frequency than the original 5800x so the question is does the 3d v cache help in these synthetic benchmarks that i'm running because these 3dmark synthetic benchmarks are some of the only somewhat practical uses for 3090 ti sli setups these days anyway and for that purpose evga sent over their x570 dark motherboard this is one of evga's first x570 or amd motherboards and they did an amazing job on it it's designed for overclocking so it has some unique features like a rotated cpu socket only two memory slots but they're positioned as close to the cpu socket as possible which is helpful for memory overclocking a host of other features on this board as well evga always does a great job designing motherboards specifically for overclocking and hitting those world records and yes four slot spacing between the two pcie slots which means i should hopefully not have to use those riser cables whether i want to test the intel platform or the amd platform but you're probably thinking paul you just showed us how the sli fingers on these cards do not line up so even with four slot spacing i still couldn't plug both of these cards in and sli them up without using the riser cables and the answer there is yes you're absolutely correct and that is why evga sent other stuff too in their little care package and uh two of the things were graphics cards these are evga for the win rtx 390 ti's and these actually are slightly different cards even though they look very much the same they have the same pcb design so the sli fingers on them will line up but one of these is an ultra card which has a little bit of an overclock and one of these is a black card which runs at a slightly lower frequency it's a difference of like 60 megahertz or something like that but this means that first i have like 6 500 or so worth of graphics cards stacked here in front of me which i'm only slightly nervous about but it also means that i can plug both of those cards into my four slot space motherboards use an sli bridge hopefully get rid of a lot of those complications so i can just focus on getting some benchmarks run and evga just within the past week or so over on their forums has posted has posted xoc or extreme overclocking v-biosis for these cards that upped the power limit from 107 to 115 so that should allow us to overclock that much more and so here are the two test beds for my initial comparison for intel we have the 12 900 ks cpu of course the asus rg maximus z690 extreme motherboard and a g-skill trident z5 rgb 32 gigabyte memory kit that's ddr5 6000 speed and for storage we have a 500 gig samsung 970 evo ssd the ssd shouldn't really affect performance but just wanted to include that for our amd testbed we have the 5800x3d cpu of course the evga x570 dark motherboard a g skill trident z royal rgb 32 gig ddr4 4400 memory kit but i did dial back that speed to 3800 so that the infinity fabric and the memory could be running at the same frequency 1900 megahertz which is kind of where infinity fabric tops out and then the ssd is of course their mp600 pro xt 2 terabyte uh i did have to remove the heatsink that came with it in order to install it on this motherboard but it still worked out on both test beds i have a corsair h150i elite lcd 360 millimeter all-in-one liquid cooler and the power supply is a corsair ax1600i 1600 watt 80 plus platinum power supply which should have more than enough wattage for our needs but i think i will be switching to a two power supply test setup at some point as well of course it still wasn't just a simple plug everything and fire it up situation this test bed over here is a liam lee pct 60 which i've had for quite some time but it has a handle that goes across the top and that can often block my access for installing and removing graphics cards so i usually just don't have it there the aluminum piece that sticks up on the side here though which is supposed to attach to the handle was actually conflicting with the four slot second graphics card so i busted out the rotary tool this is actually something i've been meaning to do for quite some time just sort of shaved it off bent it off filed it down a little bit so there weren't any sharp edges and then just slap some gaff tape on there so it has a slightly cleaner finish so after all that exposition let's finally get into some testing and my first test were just some baseline tests just running everything at stock with the memory xmp settings as already described and using the steam version of 3dmark and here the 5800x3d scored 27938 in 3dmark port royal whereas r12 900 ks hit 28 438. the amd system was pulling about 1100 watts on average the intel system was around 1150 and here i made a somewhat controversial decision so you guys can let me know in the comments whether or not i was right about this because the amd score was about 500 points less i focused the rest of my testing on the intel test bed running the 12900 ks but i noticed after the fact that the average clock speed on the amd test was 1970 whereas the average clock speed on the intel test was 2004. so i am yet again in a situation where it's totally possible that with further testing we might be able to suss out more details between the two platforms but since my goal was getting higher scores on the 3dmark port royal leaderboard so that is where i focused my efforts and back to the 12900ks testbed my first just a little bit of an overclock uh boosted the power limit up to 107 plus 200 on the memory plus 100 on the gpu we're using evga precision x1 by the way to dial in these overclock settings and that resulted in a score of 29 574 which is just ever so slightly below my top score from video number one on this subject which was 29 579 but it was an improvement nonetheless and of course there's still plenty of headroom available for me so my next steps were to dial in some optimizations some special settings in the nvidia control panel that i covered in my rtx 3090 sli testing from back in the day so i'll link to that video if you guys are interested and at this point i also switched to the standalone version of 3dmark which removes some of the overhead of steam running in the background and that boosts the score up to 29 794. so there was nice to see some quick success and it was definitely faster to get to this point with this setup than it was with the setup from two or three weeks ago but here i'll also point out that i've been running these tests out here in my garage and ambient temperature is a bit of a concern since we want the gpus to run as cool as possible so i was trying to do this testing late at night after the temperature had gone down by 10 or 20 degrees and i also had the system set up on a folding table positioned over on the side of the garage that's near my little standalone ac unit so i could just have that pointed at them running full blast the whole time but my best score on this first night of overclocking was thirty thousand and twenty so i actually broke 30k which seemed like a nice achievement a nice sort of high water mark to get past and according to 3dmark the top gpu or the faster gpu in this test was running it between 2145 and 2175 megahertz so the next night i tried to take advantage of a few more of these levers that you have to try to get more performance out of a card when you're overclocking there's the overclock itself which we are controlling with evga precision x1 software that can allow you to boost up the gpu frequency the memory frequency it can let you add a little bit more voltage you can increase that ever important power limit and you can also ramp up the fan speed because as i've mentioned multiple times the lower the temperature on the cards the higher the frequency they can usually run at stably and that is part of that second lever that you can apply some force to and that is just temperature and you can do that in multiple ways upping fan speed is one of them pointing an air conditioner at your setup is another and then of course there's the more exotic cooling solutions like water cooling which is just more effective in general you can add an ice bath to the water cooling or you can go to the most extreme and you can use ln2 liquid nitrogen or even liquid helium the final lever is these small tweaks you can make to the system itself to try to eke a little bit more performance out so i already mentioned using the standalone version of 3dmark is one of those there are nvidia control panel settings that you can mess with here such as switching to emphasize performance over quality and little tricks that you might not have thought of like swapping the v bios on the card for something that allows you to up the power limit so that was the thing that i did next in my mind i wanted to take the for the win ultra xocv bios and apply it to both cards so they'd be both running at the same frequency out of the gate evga's installer of course prevents you from doing that so i just used the xoc ultra v bios for the ultracard and the xoc black v bios for the black card in both cases the gpu z settings seemed to be exactly the same it didn't change frequency or anything like that you can notice the two different v-bios versions looking at that there but practically speaking the only change is that in evga precision x1 the power slider goes up to 115 instead of 107. i also switched to using two power supplies for these tests and i think it's likely that i would have been just fine running a single 1600 watt power supply because like i said we weren't going up beyond about 1200 watts with the whole system but i kept the gpus via the three-way power adapter dongle here all routed through a single ax1600i power supply and then i used the second 1000 watt power supply to power the rest of the system with that configuration and the 115 power limit the two graphics cards by themselves were pulling around 1025 to 1050 watts but from there i went back to overclocking with precision x1 i went to a plus 110 for the gpu plus 250 for the memory that 115 power limit fans set to 100 and with that configuration we improved the score a little bit to 30 085 just a small bump over the 30 020 from the night before and our gpus are running at between 2160 and 2190 megahertz at least again on the top graphics card i bumped up from there for one more pass to plus 115 on the gpu plus 300 on the memory added a little bit more voltage and there i got a score of 30 264 and that was enough to put me in the top 25 on the port royal leaderboard and and that felt pretty good an improvement from position 100 to sub 50 in video one and then from position 45 or so down to top 25 again some nice improvements there and good to see i was making progress but the thing that was on my mind next was again the fact that i'm using two slightly different graphics cards that are set to run at slightly different frequencies so i uncoupled the gpus in evga precision x1 that allows me to adjust settings for each of them individually rather than just trying to use the same settings for both cards i basically said it so the boost clock on the black card was the same as the boost clock on the ultra card but my score went down from there to just like thirty thousand and five but that's just how things often go when you're overclocking and trying some unique configurations like this sometimes you'll try something and the score goes up sometimes you'll try something and the score goes down from there since again i was running these tests late at night i just was started trying out different things in my documentation of what exactly i was doing got a little bit muddy i did link the gpus back up again and then here i started attacking the memory frequency a bit more because the memory on these cards is supposed to have a lot more headroom than the previous cards especially with better cooling on them so whereas before i was at like plus 300 or so on the memory i bumped it up to plus 868 and there i managed to score of 30 453. the memory effective frequency was 1 421 megahertz for that run but the best scores that i achieved for the evening apart from all the other stuff that i already talked about was using decoupled gpu settings in precision x1 that allowed me to overclock the second black card just a little bit more to get the frequency not quite up to the same level as the ultra card but to give it a little bit more boost to get it in the same ballpark so i was at plus 115 on the top card and plus 145 on the second card the memory was at plus 55 and i will say give or take here because i ran several runs it was getting quite late at night i was getting pretty tired and one of the other things you do to try to eke out a little bit more score is disabling your internet or turning off wi-fi when you run the test which means when the test pops up it says it's not a valid score until you reconnect to the internet and then validate your score so if you look at the 3dmark port royal leaderboard right now you will see that i broke the top 20. in fact i'm in the number 19 position with a score of 30 635 better than 99 of all results of course and here's a quick look at some of the settings like the memory frequency was a 1424 average 1421. average gpu temperature is 55 degrees celsius and again that is because i had the air conditioner pointed at the setup i didn't go so far as to create a cardboard shroud like i've done in the past so that could again maybe lower temperature a little bit more if i went that far but here's what it is for now our average clock frequency was 2160 and it peaked at 22.05 and there's the 1200 ks and a few other stats i also didn't go in and like disable the e cores or disable hyper threading that's another thing that's recommended sometimes but this is all to show that no matter how much time i throw out a project like this i'm always like ooh if i can just a couple more hours to do this and this and this maybe i could bump that score up just a little bit more but practically speaking i'm also making a video for my youtube channel so i do have to cut this off at some point so i can present a video to you guys so that's why i got as far as i did today but that actually isn't my top score 30 635 like i said i was disconnecting from the internet to do these runs and then reconnecting to validate my score and somehow i ran two tests scoring 30 764 and then my best run at 30 820 and apparently i did not hit save to save those results locally on the system and i did not reconnect to the internet to hit validate so it would upload those to my 3dmark account so i did not upload those scores to 3dmark so that they could be proudly shown on my account which i'm slightly disappointed about practically speaking that would have bumped me up i guess maybe two spots here so i'd be in position like 17 instead of position 19 so it's not the end of the world okay hold on i couldn't leave it there i had to do one last night of testing because i just couldn't stand it with that score i wasn't able to upload so i made some additional runs with the 12 900 ks and basically the same settings i already showed you guys scored 30 741 then managed to bump up to 30 930 ran a few more tests after that got around thirty thousand eight hundred thirty thousand nine hundred and it's pretty clear that i'm hitting a ceiling here in terms of performance without further changes to the configuration but the overclock settings i was using was plus 965 on the memory plus 119 on gpu one plus 155 on gpu 2 voltage at plus 40 power target to 115 of course and the fans set to 100 those are the settings i used for my top score of 30 930 and that has bumped me up to position 17 on the 3dmark port royal leaderboard and that's pretty respectable i feel like without the use of ln2 or water cooling even but i felt like i wasn't giving the 5800 x 3d a fair shake unless i tried this configuration on it as well so i transferred everything back over to the x570 test bed same exact overclocking settings with precision x1 and there i scored 30 561 on one run and then 30 665 on the next run so we're still about 300 points behind the 12900 ks setup and yes i did double check and verify that the frequency was good here it was running about 2205 to megahertz but i'm very glad that i took another pass at this because it let me jump up the leaderboard just a few more spots and it let me confirm that the 5800x3d is still just slightly behind the 12900 ks at least in this specific benchmark but it is the end of this video because i got to cut things off at some point i went from position 100 to top 50 top 25. now i am top 20 and i am somewhat proud of that fact although let's be honest most of it was the hardware i was using and a big big thank you to evga for lending me these two 30 90 tis i will be needing to return these cards but evga did say that i could hang on to the x570 dark and i believe i can also keep this evga special envy link sli bridge that they sent as well which is a pretty cool looking one my standard nvidia one has been through a lot and has gotten some scratches and stuff so it was nice to complete the look with the setup with this evga bridge but i'd be very interested to hear your feedback on this project did i take it far enough should i push it farther should i do a part three video i'm probably not gonna do a part three video but who knows perhaps this is just a very nice refresh for me in terms of high-end overclocking so that maybe later this year when more hardware comes out that we've been talking about and anticipating for quite some time we could maybe do it again and we can see how things go there but if you enjoyed this video definitely hit the thumbs up button and let me know i'll post links to the hardware i was using and stuff down in the description below you can also find a link to my store down there at paul's hardware dot net where you can buy shirts mugs pint glasses and other high quality merchandise to help get yourself some high quality merchandise and support my channel at the same time thank you guys so much for watching if you'd like to see more videos like this one in the future then don't forget to subscribe to my channel as well we'll see you all in the next video\n"