Intel's ATX v3.0 PSU Standard Has More Power for GPUs
5 Mar 2022 0:03:25 UTC
Topic 227136
(moderation:
A new specification (v3.0) for ATX Power Supply Units (PSU), soon to be implemented.
https://www.tomshardware.com/news/intel-atx-v3-psu-standard
What are you using for high end GPUs with respect to the PSU?
Are you having any particular trouble with power drop outs?
Thoughts??
George
Proud member of the Old Farts Association
Mostly 1200/1300/1600W units
)
Mostly 1200/1300/1600W units with one sole 1000W unit
The new spec is just designed for the next generation of high power consumption gpus that both AMD and Nvidia will be releasing later this year or next.
I had one system running 7
)
I had one system running 7 GPUs on a single 1600W Platinum EVGA unit. it ran fine, but I've decommissioned that system now.
but my 2 multi-gpu systems now run three (3) PSUs each. one ~1000-1200W unit running the motherboard and maybe 1 GPU. then 2x 1200W server PSU with a breakout board and cabling to power the other GPUs. this distributed load across the PSUs so no single one is stressed that much, but for me the better benefit is having enough PCIe power cables when running a lot of GPUs. pretty much no consumer PSU can provide 14x 8-pin PCIe connectors without some kind of splitter.
my single GPU desktop runs a corsair 1000W platinum unit, more than enough.
if you start having issues with dropouts, try to get a read on the PSU voltages to see if one of them is sagging under load, which can be a sign of PSU wear.
_________________________________________________________________________
I think power consumption by
)
I think power consumption by modern GPUs has pretty nasty high transients when running some of the Einstein applications. So a PSU which is really rigorous in enforcing maximum output current limits in the very short term that are close to matching their steady-state capability will fail to run the application when the time-average power consumption is quite far below the PSU spec limit.
The new Intel spec may require some design change in the actual power delivery (find the few places where the extra-high transient is a real problem), and different design changes in the "enforcement" circuitry so that the momentary transient limit is far above steady state. I think this could help quite a bit in reducing the perceived need to use a PSU with a spec far above the actual average consumption.
I also think that for some Einstein users the new emphasis on power conversion efficiency at the very low end may be helpful. Assuming this triggers actual design changes which don't cost too much, people who turn off Einstein GPU processing when the weather is hot, but leave their "daily driver" PC running may actually save a bit of power.
I'm afraid I generally regard the proliferation of connector plugs as an annoyance.
Caution: while I am a former design engineer, my knowledge of power supply design issues is negligible.
If you read the article
)
If you read the article linked in the first message, you will discover that they are indeed designing the power supplies to have less sensitivity to instantaneous power spikes that will be common with the new cards.
So less chance of the supply to trip overcurrent limits.