transient power consumption spikes

You let other people be guinea pigs and buy hardware combinations confirmed to be working by other people on the internet. It's not hard. That's some of the best advice anyone would ever get with building pcs on the internet, yet none of these streamer/vod scumbags will ever just say it.
This would be good advice for people getting ready to build a new a new system. But I would want to hold manufacturers (PSU / GPU / CPU / whatever) to a higher standard. Ideally if the GPU says you need 500W ATX PSU, you should be able to just get any 500W ATX PSU and it should work reliably. This is not how it is, and it may never be like that, but I think we should always aim for standards and interface specifications that can be depended on.

Apprently ATX 3.0 allows spikes of 3x power for 100 us.
https://www.pcworld.com/article/631...upplies-their-first-overhaul-in-20-years.html

Despite this, I feel like this is better fixed on the GPU. If the GPU needs 3x power draw for 100us it should be up to the GPU to put more capacitors on their board. Small high power transient spikes are usually better served by bringing capacitors as close to the destination of the power as possible.
 
I always buy a PSU equal to 2x what my systems would draw at full load (both CPU & GPU combined). As most PSUs rated efficiency are in that 50-60% range anyways right?

My 3080ti + 3900X system is drawing 583W according to my UPS so having a 1200W Platinum PSU makes sense to me :)
 
As most PSUs rated efficiency are in that 50-60% range anyways right?
I mean... yes, but not in a meaningful way 80+ Gold for example is rated for 90% efficiency at 50% but you still get at least 87% efficiency at 100% load. The efficiency difference between 50 and say 70-80% load which still gives you plenty of headroom is probably closer to 1%
 
This would be good advice for people getting ready to build a new a new system. But I would want to hold manufacturers (PSU / GPU / CPU / whatever) to a higher standard. Ideally if the GPU says you need 500W ATX PSU, you should be able to just get any 500W ATX PSU and it should work reliably. This is not how it is, and it may never be like that, but I think we should always aim for standards and interface specifications that can be depended on.

Apprently ATX 3.0 allows spikes of 3x power for 100 us.
https://www.pcworld.com/article/631...upplies-their-first-overhaul-in-20-years.html

Despite this, I feel like this is better fixed on the GPU. If the GPU needs 3x power draw for 100us it should be up to the GPU to put more capacitors on their board. Small high power transient spikes are usually better served by bringing capacitors as close to the destination of the power as possible.
Something tells me that GPU manufacturers that already skimp on the components will be less than enthusiastic about increasing the bill of materials even a few cents to compensate for this when they can deflect blame to the PSU.
 
This is something that's missing from the ATX spec. If there was a standard curve of % overcurrent vs. min/max time to trip, then there would be no need for finger pointing, because either the GPU or PSU would be out of spec if OCP tripped in normal operation.
Sorry to bump a thread by quoting myself, but it seems that Intel were way ahead of me, and had already added this to the ATX12VO spec in March 2022. It allows up to 3x nominal power as long as it's no more than 100 microseconds long.
pci_express_excusion_limits.png
 
It seems pretty obvious that video card manufacturers need to step up and put enough capacitors on cards to stop this. 3x transients are insane and no power supply vendor should have to account for that like it's normal. Is it a 400w video card or is it a 1200w video card? The requirement on the video card box needs to be stated appropriately. If they want to get cheaper on the video card build, then they need to take the pain that comes with requiring a 1000w power supply written on the box.
 
Back
Top