**Sapphire Rapids Refresh: A Mixed Bag**

The Sapphire Rapids refresh is finally here, and while it's exciting to see new hardware hitting the market, I have to say that I'm a bit underwhelmed by what we've got. The platform itself seems like a great value, but when you start to dig deeper, you realize that there are some significant gaps in terms of features and performance.

One thing that really gets my blood boiling is the lack of clear wins over Thread Ripper counterparts. The 9,000 Series Thread Rippers from AMD have been released just seven months after Sapphire Rapids launched, and they're packing a punch with their AVX 512 acceleration and other optimizations. Meanwhile, Intel's latest workstation CPUs are struggling to keep up, especially when it comes to accelerators. This is a tough spot for Intel to be in, as the company needs to find ways to compete with AMD's aggressive pricing and performance.

So, what can Intel do to turn things around? One potential solution is to focus on creating more affordable options for enthusiasts. The ASRock motherboard I'm working with is an amazing value at $500, especially considering that it includes DDR5 registered ECC memory. This makes it easy for users to build a powerful workstation without breaking the bank.

However, there are still some significant drawbacks to using Sapphire Rapids for workstations. One of the biggest issues is the lack of server-class accelerators on these CPUs. I've lost count of how many times I've been asked about this in the forums, and it's clear that Intel doesn't see a direct overlap between their workstation and server markets.

Another quirk I've discovered with Sapphire Rapids is the "C1e Auto UND demotion" feature in BIOS. When you run benchmarks, this option seems to perform really well, but when you're just using the operating system, it can feel...off. After some digging, I found that this issue may be related to C6 States or C1 promotion and demotion, which is a bit of a technical hurdle for Intel to overcome.

Despite these limitations, there are still some bright spots with Sapphire Rapids. For one thing, the CPU itself performs well in many workloads, especially when compared to older architectures like Thread Ripper Pro 7000. However, I'm not convinced that it's quite enough to take the top spot in the workstation market just yet.

In a more interesting development, I've been thinking about using Docker to test out the Sapphire Rapids platform on a networked environment. The idea is to connect my local machine to a server with accelerators, which would allow me to harness the full potential of these CPUs. This could be an exciting project, and I'm looking forward to exploring it further in a future video.

Finally, I want to thank all of you who've been working on the Comma Project. Your feedback and suggestions have been invaluable in helping me create content that's relevant to your needs. If you haven't checked out the Forum yet, be sure to do so – it's a great place to get involved and share your thoughts on the latest developments in the world of workstations.

**The Future of Workstations**

As I wrap up this article, I'm left with more questions than answers about where Sapphire Rapids is headed. Will Intel be able to catch up with AMD and secure its place in the workstation market? Only time will tell.

For now, I'm excited to see how the platform evolves over the coming months. With the release of new CPUs and motherboards, there's always something new to look forward to. And who knows? Maybe one day we'll see a monster CPU with eight memory channels that takes my Falcon Northwest system to the next level.

Until then, I'm signing off for now. Thanks again for reading, and I'll catch you all in the forums!

"WEBVTTKind: captionsLanguage: enwe're launching new xeons new Sapphire Rapids refresh xeons already and ASRock has a w790 wsr 2.0 which is more affordable than ever Intel and ASRock have partnered together to launch a new more affordable workstation platform Sapphire Rapids refresh and it's still overclockable 26 cores 52 threads here and my Corsair 5000d yeah let's get started all right so first up check it out Intel is launching new Xeon workstation CPUs you know w790 there's two different kinds of w790 motherboard basically look for a horizontal or a vertical slot is the easiest way to tell but there's a lot more differences than just the physical orientation of the slot the vertically oriented slots are for four memory channels it's more of a desktop is workstation whereas the horizontal slot is more of a server class workstation eight memory channels in the horizontal four memory channels for the desktop which can be a little confusing because our w790 WS 2.0 from ASRock actually does have eight memory slots but it's four memory channels two dims per slot if this all sounds very familiar no you're not having bja Vu I did several videos on both the 2000 series and the 3000 Series Zeon CPUs not that long ago including our Falcon Northwest 56 core monster machine learning system which I will be refreshing with the new Zeon 3000 Series CPUs and taking a look at that with RTX 5880 gpus for The Best of Both Worlds between open Veno and everything Intel has but for this this is the more affordable platform because it's for memory panels it's the same socket it's a different board configuration it's an interesting strategy from Intel and this w790 WS from ASRock is the most affordable it has ever been at just under $500 now in order to achieve that if we take a quick look at our board ASRock has had to cut things to the Bone there's only five USB type A ports on the back a single aquantia 10 gbit and then we've got our 2.5 GB Intel Nick as well and then you've got a clear seamos button and that's basically all you've got it does have you know solder options so you could have an alternative version of this motherboard with more USB ports or different layout or whatever but again let's remember this is a workstation I think a lot of people really don't care my system 76 RGB keyboard has a built-in 10 GB USB hub so I can run all my peripherals and dongles and all of that off of that keyboard with a built-in Hub it's got you know type c and type A ports for that now this this CPU this is the 26 core 52 thread Zeon and the performance of this thing screams out of the box up to 250 Watts TDP but it is overclockable and if you remember one of the things that I liked about the original sci-fi Rapids launch was that there were several overclockable skus included in both the four and the eight memory Channel configuration this is four memory channels and this thing can use a lot of power and it can run really amazingly well with an overclock now to start I'm using a notu tower cooler just dual 120 mm Push Pull configuration nothing fancy and we're going to try that in an overclocked and a non- overclocked configuration but this is literally 26p cores it's not e cores and P cores like you would have on a desktop class CPUs this is all performance cores all the time Intel's idea with this kind of a workstation is that this board would be able to support up to say four RTX 6000 gpus and yes azrock didn't skimp on that it has the extra power inputs to be able to support the 75 watts per GPU extra in terms of 12volt power delivery that is required to support that many gpus now four of our five x16 slots are gen 5x16 this platform has a total of about 64 pcie Gen 5 lanes and the other slot is pcie 4.0 x16 there not really a lot that's changed if you're really interested in this board check out my review of the 1.0 version of this board but other than making a little more cost efficient this is the 2.0 version of the board interestingly Intel dropped a little bit of a bombshell on their press pre-briefing for this board and this setup and specifically that is Intel had commissioned a report on what is the breakdown of workstations and how are they being used in terms of operating system like what software is on them and so on so forth and Intel shared a little bit of a bombshell with that Linux workstations for this kind of a setup 15% according to their market research that is enormous now for the machine learning guide that I did we were using the windows sub system for Linux which is pretty competent but when you get into these High Core count CPUs the windows subsystem for Linux still really only properly supports 64 threads and that's owing to the processor groups limitation inherent in Windows which uh is probably unfixable given that that has even tripped up Microsoft with SQL Server what a time to be alive when the computer is so powerful that literally all of the software is going to have to have special stuff and and windows and the operating system and everything else is going to have to have special stuff to handle it meanwhile un L uh yeah it's fine now this system is far from complete I need to add some more cooling into it I probably need to go with a custom Loop or an aiio or something to do a little bit more overclocking I'm going to do a full overclocking guide with this platform a little bit in the future I was sort of happy to discover that running this CPU at about 4 to 500 watts still pretty stable that was true of the previous gen's halfi Rapids but on this platform it really unlocks a whole another level of performance which is good to see I mean I mean yeah you're got a workstation that's using upwards of I don't know 7 8 900 watts from the wall with just a single GPU but it sure is fast Intel's other big advantage in the workstation Market is their software ecosystem Intel has a special version of python that they've put together which will actually perform about twice as good as the vanilla stock python that you'll you'll download at least according to their materials for for what you would be doing in machine learning and everything else and I intend put that claim to the test with this platform and to take that for a little bit of a spin so if you're into machine learning and you do a lot with python and you want to be able to run something on this machine remotely and you're in my audience well reach out on the Forum let's put a project together we've got that going with comma we've got that going with a whole bunch of other software on the Forum and uh I'll be glad to help provide a platform that you can remote into kick the tires or walk me through a script or something to run the test on the platform all right well uh performance you good news bad news kind of a situation going on here sappire Rapids refresh the out of the-box power if we compare to the original Sapphire Rapids launch these CPUs are mysteriously more power efficient they can sleep and wake a little more quickly and I don't mean suspend to RAM and wake up I mean they deal with burster workloads a little bit better probably platform improvements probably bios improvements actually compared to the launch day sci-fi Rapids workstation launch the 2495 for a couple of AI benchmarks the sapphire Rapids refresh is about 40% faster now 40% is actually a result that we need to talk about a little bit because it's really down to accelerators and uncore stuff in Sapphire Rapids refresh it's not really a dramatic change you also get a performance uplift on Sapphire Rapids not refresh through software updates or carefully said launch Day updates because the launch day SciFi Rapids had had some more that were worked out in BIOS updates now if we talk about performance at the wall for basically the same wattage you get a geom Me performance Improvement of about 8% so sci-fi Rapids refresh it's a refresh it's about as much performance uplift as Intel did from 13th to 14th generation now in terms of basic performance the stuff that lives on those peores it's more like a 6 to 8% performance uplift most of the performance uplift seems to come from better management of things under load and a better memory interface speaking of which ddr5 4800 officially supported but I did all of my testing at ddr5 5600 and actually somewhat concerningly in the Bios it looks like DDR 54400 is what it defaults to and that may have to do with the fact that on this platform you only have four memory channels but there's eight memory slots and when you have two memory slots routed even if you're not using the other two that has a negative impact on the the maximum ddr5 memory speed that you're able to use nevertheless ddr5 4800 is what Intel Arc officially says is supported I think that you can count on DDR 54400 with a single dim per Channel a little more realistically or if you're willing to go into overclock territory overclock territory you can get DDR 55600 now I've only got the X variance of these CPUs one thing that I'm super curious about is a workstation based on the non-x variants would you be able to run DDR 55600 and honestly I don't know the answer to that question I mean it wouldn't officially be supported but does it work anyway is what I'm curious about now if you get a workstation from Dell I noticed that the shipping configurations from Dell actually getting Hands-On with the Dell workstation as I was getting this together it literally came in like 4 hours ago uh it was DDR 54400 out of the box but then that may just be a Dell thing on our ASRock w790 r2.0 platform in terms of stability and Burnin testing over the last week it has been a dream so it's very nice to see that ASRock is bringing the stability even though yeah they have sort of cut down the features for this board for uh cost reasons now the other performance I want to talk about is python because over its thread Ripper counterparts it's a 2X this platform 2x's thread Ripper 32 core thread Ripper versus but again software optimizations it's down to Intel and those crazy programmers squeezing out every last ounce of performance and this is where AMD has kind of been sleeping for a while or maybe all their programmers are busy doing other things Intel has the mkl and they have one API and they have all of these other things these are a huge asset Intel has had some struggles of late this physical Hardware is not affected by those struggles because of a lot of reasons that I don't need to get into in this video but this is a solid workstation platform as far as I can tell and the sapphire Rapids machines I've been using continuously Around the Clock since the original Sapphire Rapids workstations launched including my Falcon Northwest 56 core system continuous usage even with an overclock very good stability no regressions no degradation nothing launch day issues with the original sci-fi Rapids launch those have been worked out that is very nice to see historically Intel has more quality insurance and quality control on these platforms python twice as fast is really down to software optimizations that intel was able to do for their platform I've been banging on it and I haven't been able to get those optimizations to run anyway on the threader platform but if we take away those software optimizations this is still about a 1.3 3x per core and if we look at our comma benchmarks for both the long and the short consol benchmarks these systems are posting very good results now consol really likes memory bandwidth and I don't have an eight memory Channel upgrade to this platform yet that's going to be a separate review look for that that's a 3000 Series Xeon workstation processors going to get the X going to upgrade my Falcon Northwest system or build another one or something but eight memory channels that's going to be another video comma posts very very good results here with only 28 cores it is matching a 32 core thread Ripper system now you win some you lose some depending on what sort of scientific compute that you're doing but remember consol on the Windows platform is kind of optimized for the Intel platform you can actually tell consol to use the Intel math libraries on amd's platform and see a lot but not all of the same performance benefit now it's important to realize that with a 32 core thread Ripper as we go through the benchmarks a 32 core thread Ripper is still a 32 core thread Ripper for things like cine bench where it really benefits from having just 32 cores 32 cores is really going to shine the standout here for Intel Xeon especially these refresh workstation CPUs is how much performance per core they're getting out in a workstation context the software optimizations the platform everything coming together on a per core basis that can meet that 32 core in some configuration scenarios is the part that's really impressive could also signal that you know hey amd's going to have to do some more work to get their software in of things together a little bit more optimization it definitely shouldn't be the case that Intel's math Library running on an AMD CPU is more efficient than the default math Library you know what I mean and again this is why AMD needs an army of programmers at the helm doing really amazing work doing the optimizations doing the software passes this is probably something that's true even for the desktop OS as we learned with the ryzen 9,000 Series launch a lot of performance left on the table just because of software launch timing and a couple of other things now that might be Intel that might be Microsoft that might be Intel and Microsoft working together best buds as they have been historically I don't really know but this is actually pretty good now where this loses out to thread Ripper though is in performance per watt this system is using about 50 Watts more than the equivalent 32 core thread Ripper system at the wall now that is a 50 wat reduction approximately over the original Sapphire Rapids and if you do performance per unit compute then the sapphire Rapids refresh is surprisingly more power efficient it's definitely more than a rounding error but not really a lot that I want to dive into because it's a desktop system the difference between you know 50 to 75 Watts at the wall Under full load on a desktop system is basically irrelevant but to only the loudest and screeches of YouTube commenters for a workstation it really doesn't matter maybe for Server when you got a thousand of them and you've got rack density issues okay maybe but these are Rank and file workstation systems for information workers people working in CAD and that sort of thing if you're into machine learning this is also a good platform because you've got all the PCI Lanes you can deploy gpus and I got to say it also goes really well with uh Unholy experiments like our Nvidia a3x now this is starting to get into a whole other video but I wanted to show you this cuz this is such a cool piece of Hardware this isn't like is that 100 GB interfaces on on the Ed yes it is and this is an outof band management interface this is a whole machine on a GPU basically or a GPU that's bolted onto a pcie card that you can shove in a system like this and it's designed for networking so GPU accelerated networking so with just a little bit of glue gluing This Together probably violating some of Cisco's patents somewhere this thing can outperform networking as6 that are not quite 3 years old this might be a market disruptor in the making maybe not this generation of card but in a generation or two not software defined networking I don't really want to call it software defined networking but in order to do a lot of packet processing you don't really want to do that in the CPU because there's just too much overhead with routing the packets to the CPU and back off again and a Cuda like interface for processing Network packets could revolutionize firewalls and SSL processing and a lot of uh let's call it table Stakes Network processing stuff and so the first demo application that Nvidia has going is something like snort or Syra cotta running with a rules engine based on this card and in an Ideal World the Jitter from having to process that the introduced latency from having to process those packets before passing it to The Host operating system uh you want to be as low as possible if you happen to be super into virtualization or other Intel accelerators not just the stuff that's running on the core and you want to run a lot of containers and virtual machines and something on a platform like this then this sort of becomes a useful platform for doing experimentation because youve got this to actually do the network processing and then you've got the pees and everything else on this platform to do everything else that actually maybe should be a different video I'm actually looking for different angles to do that there's not really a lot with this and I love to be able to do more content like that but without also inventing the whole other industry that's really the best aspects of the sapphire Rapids refresh launch other aspect of the sapphire Rapids refresh is well really this is probably closer to what Sapphire Rapids for workstation should have been when it launched several months ago and it's also a bit disappointing that we don't have more of a clear win over thread Ripper counterparts because thread Ripper Pro 7000 launched seven months ago and if AMD is coming in really fast with 9,000 Series thread rippers which have a lot of AVX 512 acceleration and a lot of other acceleration which should help platforms like consol if and only if AMD has their software team together to provide that enablement on the other side of just the hardware then Intel is going to be facing some stiff competition in the workstation Market here's what I think Intel can do to kind of defeat that this motherboard is amazing it's an amazing value for $500 killer job ASRock the 18 core version of this I think it's a 2965 might have been a little better as the 20 core it's too late now 18 core that's like $1,500 New Egg price make it really easy for enthusiasts to buy Intel for $9.99 $999 for the CPU $500 for the motherboard four ddr5 registered ECC memory that as a platform $500 motherboard $999 processor magical doesn't really matter it just needs to be more than 16 cores 20 cores would have been perfect I'll take 22 I don't see Intel offering a 22 core Safi Rapids refresh CPU for $1,000 but hey maybe bottom bline dollars for dollars this 28 core CPU in a lot of workloads matches the 32 core thread dripper beats it when you're looking at accelerators that are enabled on this CPU but the cost parity basically the same speaking of accelerators this CPU does not have server class accelerators and I always get the question why don't the workstation CPUs have the works the uh server the same accelerators a server and Intel just doesn't see the overlap between those two parts one other thing that I'll leave you with which is a bit of a quirk c1e Auto UND demotion Auto demotion Auto unot this option in BIOS for some reason on Sapphire Rapids refresh in a workstation scenario when you run a benchmark it performs really well in a benchmark but if you're just using the operating system it just feels weird and it turns out the the feeling weird part is actually something to do with C6 States or C1 promotion and demotion I asked Intel about this but you know they didn't really get back to me in time for making this video as to what this might be but toggle this option on in BIOS and you'll have a better interactive user experience and doesn't really make much of a difference in benchmarks but you'll feel it if you use the machine and this is true on Sapphire Rapids is Rel Sapphire Rapids refresh just depends on the boards bios configuration defaults whatever I'm working on another video kind of sort of if you want to help out with this hit me up in the Forum level one Tex um sort of working on a video where we use Docker to work on the workstation and test a bunch of stuff on the workstation but the qat accelerated stuff actually uses Docker to connect to another machine on your network which is an actual sci-fi rapid server that has those accelerators and so you can use the accelerators that are present on your workstation and the accelerators that are present in the server in order to get things done so that might be interesting that might be something fun but that's also going to be another video for another day and that's pretty much it for the Sci-Fi Rapids refresh and again this is just one CPU I've only got the one I am working on getting the monster CPU to upgrade with eight memory channels in my Falcon Northwest system which I think is still going to continue to be the ultimate machine learning workstation I'm what this level one I'm signing out you can find me in the level one forms let me know if I miss anything our comma project huge thanks to the folks on the Forum the comma project is basically driven by members like you that is is helping me it's like this is what we need to test I have to do this day in and day out and I suffer a lot cool hit me up in the Forum let's run some test all right I'm signing out and I'll see you therewe're launching new xeons new Sapphire Rapids refresh xeons already and ASRock has a w790 wsr 2.0 which is more affordable than ever Intel and ASRock have partnered together to launch a new more affordable workstation platform Sapphire Rapids refresh and it's still overclockable 26 cores 52 threads here and my Corsair 5000d yeah let's get started all right so first up check it out Intel is launching new Xeon workstation CPUs you know w790 there's two different kinds of w790 motherboard basically look for a horizontal or a vertical slot is the easiest way to tell but there's a lot more differences than just the physical orientation of the slot the vertically oriented slots are for four memory channels it's more of a desktop is workstation whereas the horizontal slot is more of a server class workstation eight memory channels in the horizontal four memory channels for the desktop which can be a little confusing because our w790 WS 2.0 from ASRock actually does have eight memory slots but it's four memory channels two dims per slot if this all sounds very familiar no you're not having bja Vu I did several videos on both the 2000 series and the 3000 Series Zeon CPUs not that long ago including our Falcon Northwest 56 core monster machine learning system which I will be refreshing with the new Zeon 3000 Series CPUs and taking a look at that with RTX 5880 gpus for The Best of Both Worlds between open Veno and everything Intel has but for this this is the more affordable platform because it's for memory panels it's the same socket it's a different board configuration it's an interesting strategy from Intel and this w790 WS from ASRock is the most affordable it has ever been at just under $500 now in order to achieve that if we take a quick look at our board ASRock has had to cut things to the Bone there's only five USB type A ports on the back a single aquantia 10 gbit and then we've got our 2.5 GB Intel Nick as well and then you've got a clear seamos button and that's basically all you've got it does have you know solder options so you could have an alternative version of this motherboard with more USB ports or different layout or whatever but again let's remember this is a workstation I think a lot of people really don't care my system 76 RGB keyboard has a built-in 10 GB USB hub so I can run all my peripherals and dongles and all of that off of that keyboard with a built-in Hub it's got you know type c and type A ports for that now this this CPU this is the 26 core 52 thread Zeon and the performance of this thing screams out of the box up to 250 Watts TDP but it is overclockable and if you remember one of the things that I liked about the original sci-fi Rapids launch was that there were several overclockable skus included in both the four and the eight memory Channel configuration this is four memory channels and this thing can use a lot of power and it can run really amazingly well with an overclock now to start I'm using a notu tower cooler just dual 120 mm Push Pull configuration nothing fancy and we're going to try that in an overclocked and a non- overclocked configuration but this is literally 26p cores it's not e cores and P cores like you would have on a desktop class CPUs this is all performance cores all the time Intel's idea with this kind of a workstation is that this board would be able to support up to say four RTX 6000 gpus and yes azrock didn't skimp on that it has the extra power inputs to be able to support the 75 watts per GPU extra in terms of 12volt power delivery that is required to support that many gpus now four of our five x16 slots are gen 5x16 this platform has a total of about 64 pcie Gen 5 lanes and the other slot is pcie 4.0 x16 there not really a lot that's changed if you're really interested in this board check out my review of the 1.0 version of this board but other than making a little more cost efficient this is the 2.0 version of the board interestingly Intel dropped a little bit of a bombshell on their press pre-briefing for this board and this setup and specifically that is Intel had commissioned a report on what is the breakdown of workstations and how are they being used in terms of operating system like what software is on them and so on so forth and Intel shared a little bit of a bombshell with that Linux workstations for this kind of a setup 15% according to their market research that is enormous now for the machine learning guide that I did we were using the windows sub system for Linux which is pretty competent but when you get into these High Core count CPUs the windows subsystem for Linux still really only properly supports 64 threads and that's owing to the processor groups limitation inherent in Windows which uh is probably unfixable given that that has even tripped up Microsoft with SQL Server what a time to be alive when the computer is so powerful that literally all of the software is going to have to have special stuff and and windows and the operating system and everything else is going to have to have special stuff to handle it meanwhile un L uh yeah it's fine now this system is far from complete I need to add some more cooling into it I probably need to go with a custom Loop or an aiio or something to do a little bit more overclocking I'm going to do a full overclocking guide with this platform a little bit in the future I was sort of happy to discover that running this CPU at about 4 to 500 watts still pretty stable that was true of the previous gen's halfi Rapids but on this platform it really unlocks a whole another level of performance which is good to see I mean I mean yeah you're got a workstation that's using upwards of I don't know 7 8 900 watts from the wall with just a single GPU but it sure is fast Intel's other big advantage in the workstation Market is their software ecosystem Intel has a special version of python that they've put together which will actually perform about twice as good as the vanilla stock python that you'll you'll download at least according to their materials for for what you would be doing in machine learning and everything else and I intend put that claim to the test with this platform and to take that for a little bit of a spin so if you're into machine learning and you do a lot with python and you want to be able to run something on this machine remotely and you're in my audience well reach out on the Forum let's put a project together we've got that going with comma we've got that going with a whole bunch of other software on the Forum and uh I'll be glad to help provide a platform that you can remote into kick the tires or walk me through a script or something to run the test on the platform all right well uh performance you good news bad news kind of a situation going on here sappire Rapids refresh the out of the-box power if we compare to the original Sapphire Rapids launch these CPUs are mysteriously more power efficient they can sleep and wake a little more quickly and I don't mean suspend to RAM and wake up I mean they deal with burster workloads a little bit better probably platform improvements probably bios improvements actually compared to the launch day sci-fi Rapids workstation launch the 2495 for a couple of AI benchmarks the sapphire Rapids refresh is about 40% faster now 40% is actually a result that we need to talk about a little bit because it's really down to accelerators and uncore stuff in Sapphire Rapids refresh it's not really a dramatic change you also get a performance uplift on Sapphire Rapids not refresh through software updates or carefully said launch Day updates because the launch day SciFi Rapids had had some more that were worked out in BIOS updates now if we talk about performance at the wall for basically the same wattage you get a geom Me performance Improvement of about 8% so sci-fi Rapids refresh it's a refresh it's about as much performance uplift as Intel did from 13th to 14th generation now in terms of basic performance the stuff that lives on those peores it's more like a 6 to 8% performance uplift most of the performance uplift seems to come from better management of things under load and a better memory interface speaking of which ddr5 4800 officially supported but I did all of my testing at ddr5 5600 and actually somewhat concerningly in the Bios it looks like DDR 54400 is what it defaults to and that may have to do with the fact that on this platform you only have four memory channels but there's eight memory slots and when you have two memory slots routed even if you're not using the other two that has a negative impact on the the maximum ddr5 memory speed that you're able to use nevertheless ddr5 4800 is what Intel Arc officially says is supported I think that you can count on DDR 54400 with a single dim per Channel a little more realistically or if you're willing to go into overclock territory overclock territory you can get DDR 55600 now I've only got the X variance of these CPUs one thing that I'm super curious about is a workstation based on the non-x variants would you be able to run DDR 55600 and honestly I don't know the answer to that question I mean it wouldn't officially be supported but does it work anyway is what I'm curious about now if you get a workstation from Dell I noticed that the shipping configurations from Dell actually getting Hands-On with the Dell workstation as I was getting this together it literally came in like 4 hours ago uh it was DDR 54400 out of the box but then that may just be a Dell thing on our ASRock w790 r2.0 platform in terms of stability and Burnin testing over the last week it has been a dream so it's very nice to see that ASRock is bringing the stability even though yeah they have sort of cut down the features for this board for uh cost reasons now the other performance I want to talk about is python because over its thread Ripper counterparts it's a 2X this platform 2x's thread Ripper 32 core thread Ripper versus but again software optimizations it's down to Intel and those crazy programmers squeezing out every last ounce of performance and this is where AMD has kind of been sleeping for a while or maybe all their programmers are busy doing other things Intel has the mkl and they have one API and they have all of these other things these are a huge asset Intel has had some struggles of late this physical Hardware is not affected by those struggles because of a lot of reasons that I don't need to get into in this video but this is a solid workstation platform as far as I can tell and the sapphire Rapids machines I've been using continuously Around the Clock since the original Sapphire Rapids workstations launched including my Falcon Northwest 56 core system continuous usage even with an overclock very good stability no regressions no degradation nothing launch day issues with the original sci-fi Rapids launch those have been worked out that is very nice to see historically Intel has more quality insurance and quality control on these platforms python twice as fast is really down to software optimizations that intel was able to do for their platform I've been banging on it and I haven't been able to get those optimizations to run anyway on the threader platform but if we take away those software optimizations this is still about a 1.3 3x per core and if we look at our comma benchmarks for both the long and the short consol benchmarks these systems are posting very good results now consol really likes memory bandwidth and I don't have an eight memory Channel upgrade to this platform yet that's going to be a separate review look for that that's a 3000 Series Xeon workstation processors going to get the X going to upgrade my Falcon Northwest system or build another one or something but eight memory channels that's going to be another video comma posts very very good results here with only 28 cores it is matching a 32 core thread Ripper system now you win some you lose some depending on what sort of scientific compute that you're doing but remember consol on the Windows platform is kind of optimized for the Intel platform you can actually tell consol to use the Intel math libraries on amd's platform and see a lot but not all of the same performance benefit now it's important to realize that with a 32 core thread Ripper as we go through the benchmarks a 32 core thread Ripper is still a 32 core thread Ripper for things like cine bench where it really benefits from having just 32 cores 32 cores is really going to shine the standout here for Intel Xeon especially these refresh workstation CPUs is how much performance per core they're getting out in a workstation context the software optimizations the platform everything coming together on a per core basis that can meet that 32 core in some configuration scenarios is the part that's really impressive could also signal that you know hey amd's going to have to do some more work to get their software in of things together a little bit more optimization it definitely shouldn't be the case that Intel's math Library running on an AMD CPU is more efficient than the default math Library you know what I mean and again this is why AMD needs an army of programmers at the helm doing really amazing work doing the optimizations doing the software passes this is probably something that's true even for the desktop OS as we learned with the ryzen 9,000 Series launch a lot of performance left on the table just because of software launch timing and a couple of other things now that might be Intel that might be Microsoft that might be Intel and Microsoft working together best buds as they have been historically I don't really know but this is actually pretty good now where this loses out to thread Ripper though is in performance per watt this system is using about 50 Watts more than the equivalent 32 core thread Ripper system at the wall now that is a 50 wat reduction approximately over the original Sapphire Rapids and if you do performance per unit compute then the sapphire Rapids refresh is surprisingly more power efficient it's definitely more than a rounding error but not really a lot that I want to dive into because it's a desktop system the difference between you know 50 to 75 Watts at the wall Under full load on a desktop system is basically irrelevant but to only the loudest and screeches of YouTube commenters for a workstation it really doesn't matter maybe for Server when you got a thousand of them and you've got rack density issues okay maybe but these are Rank and file workstation systems for information workers people working in CAD and that sort of thing if you're into machine learning this is also a good platform because you've got all the PCI Lanes you can deploy gpus and I got to say it also goes really well with uh Unholy experiments like our Nvidia a3x now this is starting to get into a whole other video but I wanted to show you this cuz this is such a cool piece of Hardware this isn't like is that 100 GB interfaces on on the Ed yes it is and this is an outof band management interface this is a whole machine on a GPU basically or a GPU that's bolted onto a pcie card that you can shove in a system like this and it's designed for networking so GPU accelerated networking so with just a little bit of glue gluing This Together probably violating some of Cisco's patents somewhere this thing can outperform networking as6 that are not quite 3 years old this might be a market disruptor in the making maybe not this generation of card but in a generation or two not software defined networking I don't really want to call it software defined networking but in order to do a lot of packet processing you don't really want to do that in the CPU because there's just too much overhead with routing the packets to the CPU and back off again and a Cuda like interface for processing Network packets could revolutionize firewalls and SSL processing and a lot of uh let's call it table Stakes Network processing stuff and so the first demo application that Nvidia has going is something like snort or Syra cotta running with a rules engine based on this card and in an Ideal World the Jitter from having to process that the introduced latency from having to process those packets before passing it to The Host operating system uh you want to be as low as possible if you happen to be super into virtualization or other Intel accelerators not just the stuff that's running on the core and you want to run a lot of containers and virtual machines and something on a platform like this then this sort of becomes a useful platform for doing experimentation because youve got this to actually do the network processing and then you've got the pees and everything else on this platform to do everything else that actually maybe should be a different video I'm actually looking for different angles to do that there's not really a lot with this and I love to be able to do more content like that but without also inventing the whole other industry that's really the best aspects of the sapphire Rapids refresh launch other aspect of the sapphire Rapids refresh is well really this is probably closer to what Sapphire Rapids for workstation should have been when it launched several months ago and it's also a bit disappointing that we don't have more of a clear win over thread Ripper counterparts because thread Ripper Pro 7000 launched seven months ago and if AMD is coming in really fast with 9,000 Series thread rippers which have a lot of AVX 512 acceleration and a lot of other acceleration which should help platforms like consol if and only if AMD has their software team together to provide that enablement on the other side of just the hardware then Intel is going to be facing some stiff competition in the workstation Market here's what I think Intel can do to kind of defeat that this motherboard is amazing it's an amazing value for $500 killer job ASRock the 18 core version of this I think it's a 2965 might have been a little better as the 20 core it's too late now 18 core that's like $1,500 New Egg price make it really easy for enthusiasts to buy Intel for $9.99 $999 for the CPU $500 for the motherboard four ddr5 registered ECC memory that as a platform $500 motherboard $999 processor magical doesn't really matter it just needs to be more than 16 cores 20 cores would have been perfect I'll take 22 I don't see Intel offering a 22 core Safi Rapids refresh CPU for $1,000 but hey maybe bottom bline dollars for dollars this 28 core CPU in a lot of workloads matches the 32 core thread dripper beats it when you're looking at accelerators that are enabled on this CPU but the cost parity basically the same speaking of accelerators this CPU does not have server class accelerators and I always get the question why don't the workstation CPUs have the works the uh server the same accelerators a server and Intel just doesn't see the overlap between those two parts one other thing that I'll leave you with which is a bit of a quirk c1e Auto UND demotion Auto demotion Auto unot this option in BIOS for some reason on Sapphire Rapids refresh in a workstation scenario when you run a benchmark it performs really well in a benchmark but if you're just using the operating system it just feels weird and it turns out the the feeling weird part is actually something to do with C6 States or C1 promotion and demotion I asked Intel about this but you know they didn't really get back to me in time for making this video as to what this might be but toggle this option on in BIOS and you'll have a better interactive user experience and doesn't really make much of a difference in benchmarks but you'll feel it if you use the machine and this is true on Sapphire Rapids is Rel Sapphire Rapids refresh just depends on the boards bios configuration defaults whatever I'm working on another video kind of sort of if you want to help out with this hit me up in the Forum level one Tex um sort of working on a video where we use Docker to work on the workstation and test a bunch of stuff on the workstation but the qat accelerated stuff actually uses Docker to connect to another machine on your network which is an actual sci-fi rapid server that has those accelerators and so you can use the accelerators that are present on your workstation and the accelerators that are present in the server in order to get things done so that might be interesting that might be something fun but that's also going to be another video for another day and that's pretty much it for the Sci-Fi Rapids refresh and again this is just one CPU I've only got the one I am working on getting the monster CPU to upgrade with eight memory channels in my Falcon Northwest system which I think is still going to continue to be the ultimate machine learning workstation I'm what this level one I'm signing out you can find me in the level one forms let me know if I miss anything our comma project huge thanks to the folks on the Forum the comma project is basically driven by members like you that is is helping me it's like this is what we need to test I have to do this day in and day out and I suffer a lot cool hit me up in the Forum let's run some test all right I'm signing out and I'll see you there\n"