Well if you like to try the E3 version of 8461v - just DM me where to send it and you will have one for free. I would love to hear back performance comparison with D0 variant.
ES Xeon Discussion
- Thread starter britinpdx
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
Ok, with the help from @Kizune have provisionally booked Xeon Gen4 8461v from US in the end, amazing thing about this model (despite disabled UPIs, so no Dual-CPU setups {as was also pointed by @RolloZ170 straight away}) is the fact that it has all accelerators for testing purposes (although not sure how to design programs for those yet and highly doubt existing databases {my specialty} have relevant support implemented, the only thing I could find is QAT features by Microsoft SQL Server and nothing about offloading workloads to DSA/IAA, but may be it's just my limited quick googling skills).
p.s.
@RolloZ170 - yes, with recent Intel' corporate fine-tuning and binning policies we may see all kind of non-retail off-roadmap gizmos offloaded to ebay soon.
p.s.
@RolloZ170 - yes, with recent Intel' corporate fine-tuning and binning policies we may see all kind of non-retail off-roadmap gizmos offloaded to ebay soon.
Last edited:
Hi RolloZ170,
I got two QUYR (step 4) recently, and I want to buy a MS X12DPI (1.1c bios) for them. Since you only mentioned 1.1b bios work, will 1.1c also work? Thanks!
Thanks! Do you know what happens with the PCIe slots if it is a SAGE version of the mobo? I am asking because SAGE has 7 PCIe slots ( x16/x16/x16/x16/x16/x8/x16) and that's clearly more than what single non-WS SPR Xeons have (WS versions trade non-enabled UPI links for more PCIe lanes) ?
@edit - nevermind, just saw the latest post. Still, if somebody has tested this it would be great to know.
like with Xeon W2400
CPU M2s missing, only slots 1,3,5,7 (but probably one more, but unsusable for 2Slot width GPUs)
check the schematic in the manual Appendix page83/84
Preliminary feedback on Xeon Scalable Gen4 8461V CPU (all accelerators but no UPI) QS version (Q16Z) with E3 stepping (48-cores / 96-Threads)
it has resulted in my first successful LGA4677 Sapphire Rapids platform boot
(previous attempt from different supplier with different model of early ES version having D0 stepping has failed/no boot {ff post code, no any signs of life}).
It does work on SuperMicro X13DEI
sold by @Kizune (ebay)
don't have much time to test it all right now but despite lack of TPM on my MB (add-on module takes some time to be delivered from China to UK) - will run more detailed diagnostics via Windows 11 (internet says there are ways to install without TPM) over the weekend.
I was starting to suspect other components (MB/RAM/PSU) but getting this QS CPU finally made all the worries to go away.
it has resulted in my first successful LGA4677 Sapphire Rapids platform boot
(previous attempt from different supplier with different model of early ES version having D0 stepping has failed/no boot {ff post code, no any signs of life}).
It does work on SuperMicro X13DEI
sold by @Kizune (ebay)
don't have much time to test it all right now but despite lack of TPM on my MB (add-on module takes some time to be delivered from China to UK) - will run more detailed diagnostics via Windows 11 (internet says there are ways to install without TPM) over the weekend.
I was starting to suspect other components (MB/RAM/PSU) but getting this QS CPU finally made all the worries to go away.
yes i know. 8272cl runs not on T7920, you have to do a VRM reprogram.
Given it another sensor capture run during CPU-Z benchmarks,
AVX keeps triggering PL throttling, so benchmarks for all-core keeps showing nothing (there are ways obviously to extend PL durations to overcome this issue but won't push for it for now).
and for comparison my current Prod i7-11700
As you can see - single-core performance is roughly the same (which is expected given SR running on "Golden Cove" hence not that much of a generational gap plus frequency limitations on Xeon vs Consumer CPUs).
AVX keeps triggering PL throttling, so benchmarks for all-core keeps showing nothing (there are ways obviously to extend PL durations to overcome this issue but won't push for it for now).
and for comparison my current Prod i7-11700
As you can see - single-core performance is roughly the same (which is expected given SR running on "Golden Cove" hence not that much of a generational gap plus frequency limitations on Xeon vs Consumer CPUs).
WoW: only 55 watts IDLE package power, the lowest value i have ever seen at SPR.
the physical redesign of the DIE tiles indeed was worth it.
There were some concerns about power draw by accelerators, but seems like it's not an issue (although I'm not sure about if invocation logic cuts it automatically).
There's another potential flaw - these programs might not be adopted for SR and new PCH/drivers, therefore reporting could be wrong.
in my experience existence of the accelerators do NOT increase IDLE package power.
you used latest HWinfo verision, the reporting is correct within the tolerances.
you can check/compare with wattage at the wall.
Is there a motherboard which supports dual 84xx QS samples and also S3 sleep?
I'd like to build a new workstation at some point, and need S3 support, lots of high clocked cores + RAM, and lots of pci-e (multiple x16 GPUs, x16 100GbE, multiple x8 and x4 NVme)
I've had great experience with QS Xeons of old - I still have 3 dual QGN7 workstation systems (EP2C612 WS, X10DAi) with S3 support and plenty of pci-e lanes (80)/slots(7,6), which have served me really well since ~2016, but are now getting very long in the tooth in terms of performance.
On the other hand, my H11DSi-NT + ZS1406E2VJUG5 system performs well for e.g. building code, but has been disappointing both for lacking S3 and lack of exposed pci-e (58/66 lanes: 5 slots, only 2 x16, and an x2 m.2, plus two disabled x4 oculink; despite the platform having ample lanes available!), limiting its usefulness for lots of primary daily driver workstation use-cases that the dual-2011 Xeon systems excel at.
It may be that Genoa-X is the thing to wait for, but it seems like QS Xeons could probably compete on performance with my Rome system, if there is a good workstation combination (but I also noticed that some QS are missing UPI links needed for dual socket)?
I'd like to build a new workstation at some point, and need S3 support, lots of high clocked cores + RAM, and lots of pci-e (multiple x16 GPUs, x16 100GbE, multiple x8 and x4 NVme)
I've had great experience with QS Xeons of old - I still have 3 dual QGN7 workstation systems (EP2C612 WS, X10DAi) with S3 support and plenty of pci-e lanes (80)/slots(7,6), which have served me really well since ~2016, but are now getting very long in the tooth in terms of performance.
On the other hand, my H11DSi-NT + ZS1406E2VJUG5 system performs well for e.g. building code, but has been disappointing both for lacking S3 and lack of exposed pci-e (58/66 lanes: 5 slots, only 2 x16, and an x2 m.2, plus two disabled x4 oculink; despite the platform having ample lanes available!), limiting its usefulness for lots of primary daily driver workstation use-cases that the dual-2011 Xeon systems excel at.
It may be that Genoa-X is the thing to wait for, but it seems like QS Xeons could probably compete on performance with my Rome system, if there is a good workstation combination (but I also noticed that some QS are missing UPI links needed for dual socket)?
Last edited:
Probably there are some options to push it even lower (BIOS or Operating System power configuration).
@RolloZ170 as per your request - turbo states for Sapphire Rapids / LGA4677 Xeon 8461V QS Q16Z sold by @Kizune on ebay
As you've predicted some time ago - there are already off-roadmap Prod units starting appearing on a market, for instance this affordable 8469 52-core from China for only $1.8K (I've kept an eye on this supplier for a while now since they do offer quite good options sometimes). The reason I didn't go with their listing this time - 8469 doesn't have any index after model name suggesting there aren't all accelerators included (most likely only one DSA is present), hence testing horizon is reduced.
Not sure if it's question of configuration (I'm not an expert on Power Sleep States) , default state on X13DEI looks like this (without any configuration of BIOS or in Windows, only fresh install with drivers):
At the same time some devices (don't have anything plugged into PCIe unfortunately) do have some S3 capabilities and mappings:
They do have Workstation version too - X13DAI , that might have a bit more broad/flexible choices.
For comparison my i7-11700 Dell Workstation/Precision 3450 system shows this (S3 Power State is supported):
As for Dual-Socket EATX/EATX+ motherboards for LGA4677 platform - haven't seen any MB with all PCIe slots (or lanes) implemented from any known brand yet (for instance previous CPU generation SuperMicro had X12DPG extended board, but for X13 they haven't even bothered to channel excess/remaining lanes via NVMe connectors which would fit on MB just nicely, not like PCIe requiring lots of space).
Theoretically Chipset/PCH has some PCIe lanes for slower-speed devices too, but those are hardly implemented/wired these days either.
Last edited:
no effect.
have disabled all cores except 8 (of 48 cores) and just 10watts lower.
lowered DDR5 from 4800 to 3600Mt = 5 watts lower.
don't realy know, must have something todo with emib/mesh.
intel's next gen Emerald Rapids do not use 4 tiles anymore, only 2 identical DIEs like the Sapphire Rapids MCC SKUs.