ASUS Pro WS W790E-SAGE SE + Intel Xeon Sapphire Rapids SPR-SP

[JosefHrib]

Yes i was doing that, and negatively amazed about my relatively poor performance. I got only 59K CBR23 so far. But in general its still a very powerful cpu. but over 10% difference still seems quite a lot.
Yes i know about the silicon lottery and benchmark beautification. Still i would trust the benchmarks i saw here on the forum from users. I do not really want to go BLCK, if i have time i would play with undervolting, and thermal settings.

Bios 0805, is working, but memory overclocks still only work manually. XMP not working get some unknown f0 error with almost any setting apart from 4800MHz.

With 4x32GB Fury pro 5600 manual, timings i get 56K in CBR23 Aida6 Memory reas speed is now 114GB/s about 2x more as with 2 sticks. and now about half of the 235GB/s JosefHrib got. Which was interesting because, i thought that theoretically for 4 channels 4 RDIMMS of 1Rx4 should be better than 8 RDIMMS, because besides overclocking according to officiall specifications with 2Dimms per channel speed reduces to 4400, instead of 4800 (at least that is how i understood it).
But with 8 sticks and 4 channels JosefHrib still has double the bandwidth. So probably i misunderstood something. Maybe the difference is also 1Rx8 vs 1Rx4? I am very curious if someone with an 8 channel board, and 8 dimms could run AIDA64 and see the bandwidth? Would that then be double of the 235GB/s?

EDIT: JosefHrib i went trough your screenshots again, did you notice this?
View attachment 34100
It says Octa channel with yours! Maybe the mainboards only specify 4-channel because the cpu's they officially support are up to 4 Channels but with these they might be actually 8?

My board W790E-SAGE is 8 channel and I using 8 modules. Best case scenario. If you have 4 channel with 4 slots so best case scenario are 4 modules.
Now I retested with AIDA64 7.00.6700 results are very similar.

 

[sam55todd]

Since it's still Sapphire Rapids then mine 8461V showing these memory results.



Write and copy are a bit lower of course (around/below 200GB/s)

P.S. screenshot from @JosefHrib above strangely shows worse speeds (except for L1/L2/L3 caches which are IMV a bit more important) despite having better latencies and RAM clock.

 

[RolloZ170]

It says Octa channel with yours! Maybe the mainboards only specify 4-channel because the cpu's they officially support are up to 4 Channels but with these they might be actually 8?

SPR-SP 8 channels
SPR-WS W?-34xx = 8 channels
SPR-WS W?-24xx = 4 channels

 

[Kizune]

Here's my BIOS screenshot with the very mild o/c, i haven't touched the XMP profiles, server Hynix memory 5600 o/c just by setting target speed in "DRAM Frequency" pulldown menu. Thing is applying XMP profile changes voltage settings and any change of voltage causes my system to stop posting until i reset CMOS. Aida64 and CB results are not from a clean run - there is a lot of stuff running in background including like 10 windows of chrome.

 

[DHamov]

My board W790E-SAGE is 8 channel

Thanks for sharing your old and new screenshots.
Auch! i did not notice that doing my research that the Asus W790-SAGE is 8-channel, and i somehow assumed that all workstation boards ws790 chipsets were 4-channel, because so many are. So that is one important difference to notice for others who will build a system. Another thing that could be disturbing about the w790-ACE, and what i missed to nice when deciding is that: the PCIE slots, start 2 slots under the IO shield. I wanted the possiblility to move from a single to a dual rtx 3090 setup with 1 slot in between. So with a 3 slot gpu in the bottom slot, you need a case with 9 slot support, because the top 2 slots are not there on this board. During planning, silly me started counting immediately under the IO shield and did a 5+2=7 so and thought that 8 slots in my case are sufficient, but no. My Lesson: count from the bottom up.

Since it's still Sapphire Rapids then mine 8461V showing these memory results.

Thanks for sharing. Very good performance. Honestly, seeing that now i regret a bit, that i did not the 8461V vs the QYFS and took use that $200 difference of the 8-channel mainboard. But who knows how lucky i would have been with that Silicon Lottery ticket. Kizunes QYFS seems to work very well. That about 20% higher memory speed of 5600 over 4800 in my mind was an argument for the Asus board, but your memory speeds with stock 4800 are quite amazing.

Thing is applying XMP profile changes voltage settings and any change of voltage causes my system to stop posting until i reset CMOS.

I think i have similar issues, xmp is only working at 4800 speeds when the voltages are 1.1 instead of 1.25 or so. But very good results, i mean we basically have the same hardware (W790-ACE+QYFS) although you have other 5600 Memory, and i have Fury pro 5600, but difference of 8k from my 59k to your 67k on CBR23 is quite a lot. Mine is not taking 102 BCLK.

So compared to all of this mine seems a bit slow, but i did some load testing, and at least it's stable. I think i can live with it.


Does any one know, if one would buy 5600Mhz memory for another serverboard, or workstation board that has no overclocking and a QYFS, or 8461V, can one just set the memory speet to 5600? or is that overclocking, and because standard memory speed is restricted to 4800?

 

[RolloZ170]

Does any one know, if one would buy 5600Mhz memory for another serverboard, or workstation board that has no overclocking and a QYFS, or 8461V, can one just set the memory speet to 5600? or is that overclocking, and because standard memory speed is restricted to 4800?

the current Kingston Renegade Pro DDR5-5600/6000 are jedec 4800. means without XMP they run at 4800.
if you buy Jedec DDR5-5600 you can set to 5600, Emerald Rapids i.e. supports 5600.

 

[DHamov]

the current Kingston Renegade Pro DDR5-5600/6000 are jedec 4800. means without XMP they run at 4800.
if you buy Jedec DDR5-5600 you can set to 5600, Emerald Rapids i.e. supports 5600.
QYFS/8461V memory controller: overclocking supported

That is great news, so do i understand it correctly that in this case, the important thing is the Jedec frequency, and that the CPU controller supports overclocking, but not essential that the motherboard explicitly supports memory overclocking

 

[RolloZ170]

and that the CPU controller supports overclocking, but not essential that the motherboard explicitly supports memory overclocking?

if your motherboard supports 5th gen scalables it supports DDR5-5600 at 1.1V = no overclocking.
make sapphire rapids to work at 5600 is overclocking the processor.
Kingston Renegade Pro XMP DDR5-5600 is overlcoking CPU&Mobo because the voltage rise to 1.25V
• XMP Profile #1: DDR5-5600 CL36-38-38 @1.25V

 

[DHamov]

if your motherboard supports 5th gen scalables it supports DDR5-5600 at 1.1V = no overclocking.
make sapphire rapids to work at 5600 is overclocking the processor.
Kingston Renegade Pro XMP DDR5-5600 is overlcoking CPU&Mobo because the voltage rise to 1.25V
• XMP Profile #1: DDR5-5600 CL36-38-38 @1.25V

Sorry to make in completely clear to me and maybe also others:
I understand that 5th gen supports DDR5-5600 without overclocking.
But..
For example would GIGABYTE MS73 or MS03-CE0 with 5600 micron memory run it at 5600 at 1.1v with a QYXX (4th gen)?
So basically as a question if the board supports 5th gen cpus, but even we have a 4th gen cpu, can the ram still run at 5600 without an overclocking motherboard, because , jedec is 5600 and 1.1v so we are not overclocking the motherboard just the cpu?

 

[DHamov]

yes if the CPU is not locked to 4800. (5600 or overclocking supported)

Is that ever the case? i just looked trough some intel specification for example

Intel® Xeon® Platinum 8480+ Processor (105M Cache, 2.00 GHz) - Product Specifications | Intel

Intel® Xeon® Platinum 8480+ Processor (105M Cache, 2.00 GHz) quick reference with specifications, features, and technologies.

www.intel.com

and almost always there is a max memory speed, and i did not find anything about overclocking.

 

[DHamov]

Sorry i still do not really understand it.
Maybe my question was also asked in the wrong way, so now i modified it, to a more specific question

For example would GIGABYTE MS73 or MS03-CE0 with 5600 micron memory run it at 5600 at 1.1v with a QYFS/8461V (4th gen) cpu?
Does the answere remain the same as before?

yes if the CPU is not locked to 4800. (5600 or overclocking supported)

How to find or know if QYFS or 8461V or any other xeon scalable supports memory overclocking? For the official intel specs they write 4800MT or 4400MT and i could not find anything about overclocking in the cpu specs. It seems that all CPU's have locked memory speed.

What confuses me is the distinction you seem to make between memory overclocking support on the cpu and or motherboard.
I thought that memory overclocking was just a motherboard property. (one of the reasons in favor of asus W790), but you also seem to speak about:

QYFS/8461V memory controller: overclocking supported

How do you know this? from the official intel specification? or how to find this information?

My interest in this is that: if there is really a distinction between motherboard memory overclocking support, and cpu memory overclocking support, that maybe one can use faster jedec memory at standart voltages on a motherboard that does not support overclocking (but still supports that memory speed because of 5th gen, but with a 4th gen cpu that does support overclocking. And hopefully it would run at the faster speed? But maybe this is just also wishfull thinking.

As i understand it now for my Asus W790-ACE, micron jedec 5600 1.1v memory would probably work with QYFS.
But what i also would like to know is if in for example on MS03-CEO or MS73 (non-overclocking boards) the same memory would also run at 5600 speed with QYFS.
Which i still do not understand. Probably i am missing an important point.

 

[DHamov]

MS03-CEO or MS73 with 4th/5th gen BIOS(R0x) supports 5600 memory, no overclocking.
if CPU supports 5600 or mem. overclocking 5600 is working.

Thanks its clear now, that it works.
But now from HWiNFO64 reports we can know that overclocking works for that cpu, any other pointers how one could know that before owning the cpu? (Except asking the seller to make this report) I tried intel official specs and www.cpu-world.com, the HWinFO reports of other users do not seem to be uploaded and easily searchable. Or any more general hypothesis possible based on your observations? did you come across any recent 4th gen ES Xeons which did not support memory frequency overclocking?

 

[RolloZ170]

But now from HWiNFO64 reports we can know that overclocking works for that cpu, any other pointers how one could know that before owning the cpu? (Except asking the seller to make this report)

asking here and hope

https://forums.servethehome.com/index.php?threads/es-xeon-discussion.5031/

I tried intel official specs and www.cpu-world.com,

if they write 4800 then this is max. except unlocked X/K models
cpu-world is collection from intel ark and user reports.

did you come across any recent 4th gen ES Xeons which were not unlocked?

sure. i.e. Q075 silver 4416 ES

Memory Frequency Support:

DDR-2133

and more.

 

[scouzi]

Anyone here actually running on MS03-CEO? I'm still debating waiting for my back-ordered MS03-CEO (without knowing when) vs Asus W790-ACE which I can get in a few days time from Amazon. Both around the same price.

 

[JosefHrib]

P.S. screenshot from @JosefHrib above strangely shows worse speeds (except for L1/L2/L3 caches which are IMV a bit more important) despite having better latencies and RAM clock.

The test was performed on production computer where many applications are running in the background, etc., the usage ram was about 30GB. I didn't make any adjustments just for retesting. Therefore, the results are not quite the best. If I restarted the computer and turned off all background applications etc., so test results would be probably “maybe” little better.

Anyone here actually running on MS03-CEO? I'm still debating waiting for my back-ordered MS03-CEO (without knowing when) vs Asus W790-ACE which I can get in a few days time from Amazon. Both around the same price.

And CPU? If D0 so I prefer ASUS. But with GIGABYTE you can run EMR too.

 

[scouzi]

And CPU? If D0 so I prefer ASUS. But with GIGABYTE you can run EMR too.

It is a QYFS D0. EMR is nice for the future but SPR should still last me a while! Asus ACE does have nice consumer features and a $1k credit of Adobe Cloud. I suppose I could try to sell the Adobe coupons at a steep discount.

$1k CAD it seems

asus offers adobe creative cloud | ASUS Canada

www.asus.com

 

[DHamov]

Yesterday i got a used rtx 3090, not for gaming, but for to speed up tensorflow and pytorch code for my work. So i plugged it in the W790-ACE+QYFS. In order to test relative functionality an performance against other RTX 3090, i tried 3dmark time spy benchmark. Its running some game like rendered animation. However i only get half the score of what i am supposed to get (around 9000, instead of 18000, the gpu seems not to get to full load). In some other test i did the card seems to run very well (better than my old used ex mining A2000). Writing with the owner he told that the benchmark is not suitable for HEDT cpu's. To which this QYFS is some what similar. Also then i saw this video about an old 81xx xeon, with similar issues frame rates all over the place, from 50-200. Honestly i do not game, so it is irrelevant for me. But i want to have some good test as a performance reference for my setup and new used 3090, to know if everything is working properly. Because also the return window day for the W790 is closing.

Can any one of you confirm this type behavior of these newer ES Xeons for gaming, time spy benchmark or in general? Or is it still something else on my setup. On the ai-benchmark i got 34000 better than the V100, but its a bit of an old benchmark, and no reference scores for the 3090.
Maybe its a general problem, not ES related, but if one has an ES cpu and a problem that seems difficult to solve, it is hard not to see it as a potential ES issue.

 

[JosefHrib]

Yesterday i got a used rtx 3090, not for gaming, but for to speed up tensorflow and pytorch code for my work. So i plugged it in the W790-ACE+QYFS. In order to test relative functionality an performance against other RTX 3090, i tried 3dmark time spy benchmark. Its running some game like rendered animation. However i only get half the score of what i am supposed to get (around 9000, instead of 18000, the gpu seems not to get to full load). In some other test i did the card seems to run very well (better than my old used ex mining A2000). Writing with the owner he told that the benchmark is not suitable for HEDT cpu's. To which this QYFS is some what similar. Also then i saw this video about an old 81xx xeon, with similar issues frame rates all over the place, from 50-200. Honestly i do not game, so it is irrelevant for me. But i want to have some good test as a performance reference for my setup and new used 3090, to know if everything is working properly. Because also the return window day for the W790 is closing.

Can any one of you confirm this type behavior of these newer ES Xeons for gaming, time spy benchmark or in general? Or is it still something else on my setup. On the ai-benchmark i got 34000 better than the V100, but its a bit of an old benchmark, and no reference scores for the 3090.
Maybe its a general problem, not ES related, but if one has an ES cpu and a problem that seems difficult to solve, it is hard not to see it as a potential ES issue.

Non RTX3090Ti, normal RTX3090 vs RTX3080Ti is very similar in performance with RTX3090 and RTX3090Ti, main difference is memory size.

And because before I bought RTX4090 I owned RTX3080Ti so I can post here some results from 3DMark.
But with my older GPUI have results only from Test: 3DMark Speed Way
(only one thing 3DMark bad named my processor, 3DMark showing Xeon W-3175X but this is bug)

 

[bayleyw]

That seems about right - the 3495x is a similar die configuration and posts very bad Time Spy scores:

I wonder what's going on, the 3475X is XCC as well but scores almost double...except when stock, where it apparently performs like a Core i3?!

 

[JosefHrib]

That seems about right - the 3495x is a similar die configuration and posts very bad Time Spy scores:

I wonder what's going on, the 3475X is XCC as well but scores almost double...except when stock, where it apparently performs like a Core i3?!

Few days back I try 3DMark Time Spy Extreme, but only one time, with many applications under load and without fresh start (machine was poweron many days). My Time Spy Extreme with RTX4090 and QYFS:
Graphics Score 10998
CPU Score 14060
I not playing many games (only few, Cyberpunk 2077, Flight Simulator,..) and under these games run RTX4090 same as in other gaming results what have for example techpowerup etc. So maybe is only something bad with benchmark apps? I don't know.