ES Xeon Discussion

[RolloZ170]

and Intel has not fixed it from D0 to E0,but I saw lower idle power in E3 8461V

E3 8461V is same with high.perf.profile.

even Xeon W9-3495x testers reported same "idle" power.
i have 3408U QS, 30 Watts ilde, mouseclick reaction is not same than usual but you get used to it.

 

[MillionMiles]

mouseclick reaction is not same than usual but you get used to it.

XCC have bigger lagging than MCC and IceLake Xeon

 

[DHamov]

150w is not unacceptable, E0 is good,CPU is use for running not sleeping

I agree, but i think that principle it should be possible for the OS to manage this better. Keep 1-2 cpu's in fast state, and set the affinity of main GUI related processes, to those cpu. For 8 p-cores on lga 1700 it does not pay but for 40+ it does pay of. Manually its a headage, but they can do it.

Does the Supermicro bios allow setting temporary power limits, upward from TDP?
In HWInfo the cpu section if it does not crash are all AVX512 functionalities enabled?
And probably useless in practice but just out of curiosity, is memory overclocking of the cpu enabled according to HWInfo or not? Even if it were the case now there would not be any supporting mainboards, but maybe
RolloZ170 or some one else will hack a bios to get it working .

 

[MillionMiles]

Does the Supermicro bios allow setting temporary power limits, upward from TDP

No way, And CPU power less than 300w(TDP) in AVX2

are all AVX512 functionalities enabled?

is memory overclocking of the cpu enabled

IA and MESH are locked, I'm not sure if the imc can overclocking,but looks good

@RolloZ170 or some one else will hack a bios to get it working

I can set 5600MHZ in BIOS, but it doesn't take effect
Because of the ROT, this is almost impossible on the SuperMicro board

 

[sam55todd]

IMV this "idle power" priority doesn't seem to be consistent with a very nature of high-core Server grade Xeon CPU concept on a first place.
For low idle states there are desktop cpus and "efficiency" cores.

 

[RolloZ170]

XCC have bigger lagging than MCC and IceLake Xeon

8461v QS is not much different in lagging ( if i can call it so ) i mean it is a fraction of a second,

 

[RolloZ170]

In HWInfo the cpu section if it does not crash are all AVX512 functionalities enabled?

settings: safety, check: "safe scan of potentialy problematic PCI devies"

 

[RolloZ170]

I can set 5600MHZ in BIOS, but it doesn't take effect
Because of the ROT, this is almost impossible on the SuperMicro board

works on all W790 with steping D0 ( if supported, ASUS W790-ACE & SAGE SE, ASRock W790 WS)
but you can set UEFI variables with a tool(AMISCE,RU.efi) for hidden settings.
on my gigabyte MS33-AR0 the EFI code doen't allow more than 4800 with SPR processor, even XMP settings are there but no code for using them in the BIOS. and gigabyte MW83 W790 board does't support XMP too.

 

[RolloZ170]

Solved the issue of system response slow, but the power returned to 150w

so "out of bands" mode instead of "native" for the "hardware p-states"
keeps the mesh/cache/uncore high active but the cores are energy saving.
looks for me this is not issue with cores.

 

[bigwomp]

well i have not tested this model but ALL D0 ES2 i have tested are working on ASUS W790 ( QYFQ,QYFP,QY06,QY0C,QYFS... )
btw:
ASRock W790 WS board keeps D0 microcode as well, maybe someone has the chance to check if D0 ES2 runs on it too, would not wondering me.

I assume QYFU D0 ES also works on ASUS W790 ACE? otherwise I'm thinking of buying xeon 8468 ES QYFQ for the ASUS W790 ACE since it seems to be the popular option.

 

[RolloZ170]

I assume QYFU D0 ES also works on ASUS W790 ACE?

yes.

otherwise I'm thinking of buying xeon 8468 ES QYFQ for the ASUS W790 ACE since it seems to be the popular option.

QYFU is better option or QYFS/QYFR if you can.
QYFP is better than QYFQ if you don't need QAT accel. ( not tested if usable so far )

 

[bigwomp]

yes.

QYFU is better option or QYFS/QYFR if you can.
QYFP is better than QYFQ if you don't need QAT accel. ( not tested if usable so far )

excellent, now i just need to find RAM to order for it, im considering the Kingston Renegade Pro KF548R36RB-32 RDIMM with XMP profile, hopefully those work

 

[RolloZ170]

excellent, now i just need to find RAM to order for it, im considering the Kingston Renegade Pro KF548R36RB-32 RDIMM with XMP profile, hopefully those work

have not tested KF548R36RB 4800 with low latency. but with a modern motherboard like this the JEDEC PnP Profile should be no problem.

 

[MillionMiles]

Q0KS look the same as the 8452Y on Intel's website, with only one default enabled DSA found. 4xDSA and 4xIAA seem to be locked by Intel On Demand?

 

[DHamov]

Q0KS look the same as the 8452Y

Does it also have the Intel Speed Select Technology options and can you activate it in the bios?
Intel on demand ..., i wonder if they are greedy enough to activate it on ES if you pay for it? But i guess not.

 

[MillionMiles]

Does it also have the Intel Speed Select Technology options and can you activate it in the bios

i wonder if they are greedy enough to activate it on ES if you pay for it? But i guess not.

Maybe they can be activate,but not from Intel,and there are bugs in the accelerator before E3 stepping

 

[RolloZ170]

Q0KS look the same as the 8452Y on Intel's website, with only one default enabled DSA found. 4xDSA and 4xIAA seem to be locked by Intel On Demand?

#231,#236,#241,#246 = 4 ???

 

[sam55todd]

Maybe they can be activate,but not from Intel,and there are bugs in the accelerator before E3 stepping

Non-prod units (mostly ES but often even QS) are a lottery, I'm getting a lot of those (QAT, not DSA as per original post) with 8461V-QS despite latest drivers and all updates, although it might not be CPU but something else too.

 

[scouzi]

Anyone testing with y-cruncher?

Here's my 1b score with a pseudo 8480+ (QYFS) 11.3 secs

Start Time: Fri Feb 16 20:43:21 2024

Working Memory... 4.66 GiB (locked, spread: ?)
Twiddle Tables... 9.99 MiB (locked, spread: ?)

Begin Computation:

Series CommonP2B3... 70,513,681 terms (Expansion Factor = 2.699)
Time: 9.098 seconds ( 0.152 minutes )
Large Division...
Time: 0.622 seconds ( 0.010 minutes )
InvSqrt(10005)...
Time: 0.363 seconds ( 0.006 minutes )
Large Multiply...
Time: 0.312 seconds ( 0.005 minutes )

Pi: 10.396 seconds ( 0.173 minutes )

Base Converting:
Time: 0.966 seconds ( 0.016 minutes )
Writing Decimal Digits:
Time: 0.269 seconds ( 0.004 minutes )

Start Time: Fri Feb 16 20:43:21 2024
End Time: Fri Feb 16 20:43:35 2024

Total Computation Time: 11.362 seconds ( 0.189 minutes )
Start-to-End Wall Time: 13.332 seconds ( 0.222 minutes )

CPU Utilization: 8816.55 % + 33.88 % kernel overhead
Multi-core Efficiency: 78.72 % + 0.30 % kernel overhead

Last Decimal Digits: Pi
6434543524 2766553567 4357021939 6394581990 5483278746 : 999,999,950
7139868209 3196353628 2046127557 1517139511 5275045519 : 1,000,000,000

Spot Check: Good through 1,000,000,000

Version: 0.8.3.9533 (Linux/18-CNL ~ Shinoa)
Processor(s): Genuine Intel(R) CPU 0000%@
Topology: 112 threads / 56 cores / 1 socket / 1 NUMA node
Usable Memory: 269,875,261,440 ( 251 GiB)
CPU Base Frequency: 1,900,037,664 Hz

Validation File: Pi - 20240216-204335.txt

 

[RolloZ170]

Non-prod units (mostly ES but often even QS) are a lottery

ES = "0000%@"
QS = "...Platinum 8461v"
QS is same silcon than prod.unit w. same stepping, just ES bit set.
QS are thrown out to driver programmer, ES bit set to avoid them to sell them as prod.units, ES bit voids warranty.