Power supplies do NOT affect performance! ... right?

[Real_Smoky]

I would like to confirm something I already think I know.

Q: "I have low performance in X, my specs are these"
A: "Weak PSU"
_______

I see this kind of interaction in Facebook PC groups almost every day and I wonder why when it simply is not true and makes no sense.

A low wattage PSU will NOT lower FPS/increase render times/cause stutter. At least not normally. Maybe in some very, very rare cases.

A low wattage PSU will either die or it will turn itself of causing the system to shut down. A system can not slow down because of the PSU. The system draws however much it wants and the PSU tries to provide. All of this might change with the new PSU and GPU connector standard with sense pins, but currently, it is not a thing.

Source: experience, logic, and this article.
https://xtrium.com/does-power-supply-affect-fps/....

I am ready to be proven wrong if someone posts a video of the exact same system performing differently by just switching the PSU.

Thanks! 

[narrdarr]

Uhm yes/no.  Even using the pigtail/daisychain can have these affects.  it really comes down to how well that power is delivered. an under powered psu cant deliver the correct amount of power consistently this may happen for an extended amount of time before the safeguard kicks in or the unit fails or kills another part of the system.

[mariushm]

It can affect performance by a few percent, usually very small amount.

 

The smoother the voltage going into a video card is (and the faster the power supply reacts to sudden demands and maintains smoothness), the more precise the voltage converters on the video card can be and allow the video card to boost to a bit higher frequencies when needed.

A cheap power supply may output 12v with +/- 0.01v fluctuations to video card when in Windows, but when in games when video card consumes 200w, the power supply may output 12v but due to losses in the wires the video card may get only 11.8v and also, as the consumption varies, that voltage may fluctuate by +/- 0.1v, so in turn the voltage converter on the video card needs to work harder to adjust the output to the demands of the gpu chip (you have 11.8v ...12.2v in, you may have 1v ... 1.5v out, changing from ms to ms)

[Chiyawa]

Well, that pretty much depends on the system. Yes, if your PSU can't supply enough power, it could impact performance. With the new ATX 3.0 standard, the PSU can 'limit' the GPU performance by indicating that it can only provide X amount of Watt to it.

 

But yeah, currently, PSU shouldn't cause your system to perform poorly unless it was overloaded beyond what it is capable of. The issue pose by PSU is very negligible because the VRM usually can correct fluctuation of the voltage.

[matangk]

I"m unsure what exactly the internal schematics of a modern GPU are, but I'd say a PSU (except if as stated can communicate power draw limits to motherboard) is unlikely to make your graphics just 'slower'. The actual supply voltage of the processing core of the GPU chip is sub 1V. Usually on-chip regulators produce this voltage from say 1.2V-1.8V. Other peripherals (communication controllers etc) require probably anywhere from 1V to 3.3V, so every power consumer on the GPU card is regulated.

 

A voltage 'regulator' is so named because it outputs a known voltage even if the input varies within a permissible range which may be quite large, usually much larger than +/-1V. That of course, if the changes are slow enough for the regulation circuitry to track them. We sometimes call the rate at which a regulator tracks changes its 'band-width'.

 

So, what happens when a card tries to pull more power than what the supply can deliver? If voltage is fixed, it means that it pulls more current. p=V*I  (power=voltage times current). But there is a limit to current draw, and at that limit, the increase in current will result decreased voltage and regulators will stop functioning correctly. Now imagine a chip with billions of transistors that needs minimum 0.8V to function but is fed 0.75. You will see far worse than slow downs.

 

Another scenario is if you have a crap PSU, and power draws change so fast, that they are beyond the power supply ability to correct them. One thing that we do is to add electronic filters (usually capacitors) to provide these quick charge draws without a big change of supply voltage. But if the power draws are both Lage AND fast, they may mess up regulation momentarily or even trip the PSU's protection. Those again are also not nice little 'slow downs'.

 

The only scenario I can think of where supply can actually make the system slower is if on the PCIE interface to the mother board there is a back-and-forth verification or error correction. Then, poor supply might cause data corruption that will constantly need retransmission and result a slowdown. But then again, if that is the case it will probably manifest in other various parts of your system as well.

 

So no, I'd say no. Those answers you read on FB are incorrect,

 

but then again I might be wrong and will be happy to learn from the experts.

 

 

 

 

 

[NobleGamer]

14 hours ago, Real_Smoky said:

A low wattage PSU will NOT lower FPS/increase render times/cause stutter. At least not normally. Maybe in some very, very rare cases.

A low wattage PSU will either die or it will turn itself of causing the system to shut down.

You are normally true about the last part. You know what can also be true?

 

The first reply and matangk both provided scenarios in which PSUs can in fact result in the issues you describe without ever exceeding their rated wattage capacity.

 

This actually happened to me with my RTX 3080 EVGA FTW3 and 5600X using a 5 year old 750W PSU: Antec HCG-750. I had to do daisy chaining for 2 of 3 pcie power connectors, as the PSU only had 2 cables with 4 pcie connectors total.

 

I had frame drops and instability with Shadow of the Tomb Raider benchmarks when switching between scenes. This was despite total power draw not exceeding ~500W.

 

Perfect example of what you claim very very rarely happens. I don't think its that rare for people to keep a power supply for 5 years, for people to daisy chain connectors, nor rare for people to get a GPU that needs 2-3 connectors.

[matangk]

31 minutes ago, NobleGamer said:

I had frame drops and instability with Shadow of the Tomb Raider benchmarks when switching between scenes. This was despite total power draw not exceeding ~500W.

Excellent case in point, but I'm not sure that's what OP would call a 'slow down'.

 

What you describe sounds like sporadic voltage drops when GPU is trying to pull a lot of current. From what I've read those RTX 30xx cards can pull up to x2(!) Their rated power momentarily. Some PSUs were known to have trouble with that. W/O being a PSU expert I guess it's either for lack of sufficient filtering capacitance or for tripping over current protection.

 

I wonder. Were your issues happening without the daisy chain as well? The connection point in cables depends on the cable quality, there can also be loose contacts, long term oxidation etc. What happens is that connector then adds series resistance to cable, and large current draw results a 'voltage drop' b/c voltage=current*resistance (V=IR). If you've ever had a dryer outlet heating or melting that's exactly the same.

 

Just for a reference, TDP of 3080 is 320W so at max power it draws 320W/12V=26.6 Amps. If your supply line has a resistance of just 0.2 Ohm, the card itself will receive only 12-26.6*0.2=6.7V which is really bad, and on a spike 12-2*26.6*0.2=1.36V which is guaranteed to result a crash.

[NobleGamer]

These are some of the specs I found about the PSU: +3.3V@25A, +5V@25A, +12V1@40A, +12V2@40A, +12V3@40A, +12V4@40A, -12V@0.5A, +5VSB@3.0A

 

I don't think there's anything with the amp limits that's relevant here since you showed that it would only draw 26A, and that would be divided by two 12V cables.

 

1 hour ago, matangk said:

What you describe sounds like sporadic voltage drops when GPU is trying to pull a lot of current. From what I've read those RTX 30xx cards can pull up to x2(!) Their rated power momentarily. Some PSUs were known to have trouble with that.

I do think that this could've been a thing since it would've been within months of RTX 3080 launch before there were any driver side fixes.

 

1 hour ago, matangk said:

I guess it's either for lack of sufficient filtering capacitance or for tripping over current protection.

I think it could be lack of filtering capacitance. I don't know of OCP applies.

 

1 hour ago, matangk said:

Were your issues happening without the daisy chain as well?

With the 750W PSU, I had no option NOT to daisy chain because I was limited to 2 pcie cables with 2 connectors each, and the 3080 EVGA FTW3 required 3 pcie connectors (yes that's more than the FE's 2 connectors).

 

I did not have the same issue with a RM850X from Corsair.

EDIT: My RM850X has 3 separate pcie cables with 2 connectors each. Newer models only come with 2 pcie cables, but more EPS cables for some reason. I can't confirm that I've tried pcie daisy chaining with the RM850X, as I only recall ever using separate cables.

Edited October 21, 2022 by NobleGamer
RM850X correction

[Motifator]

"Lack of filtering capacitance", you'd think you know better than the engineers over Seasonic at making PSUs, ESPECIALLY at putting capacitors with proper ratings, which is one of the most basic parts of making a PSU? It's called transient load spike performance, look up on it. 

Also, that Seasonic you spoke about uses some of the top notch Hitachi and NCC Japanese formula caps. In 5 years time, it will NOT degrade to a point you will have instabilities under any regular conditional usage. Maybe if you mine on it with multiple cards all day, but not under everyday tasking, nope. It's the kind of PSU that lasts 10 years.

There are things people don't speak about when it comes to these stuff, like delivering dirty, ungrounded, un-protected power from the plug to the PSU. This can murder a PSU in much shorter periods, or reduce its functionality. This, I'd say might be your situation. 

As for PSUs of suffice power makes changes on setups, that is a myth. As long as both units are well capable, a better PSU does not high up your everyday overclock.

[matangk]

 

1 hour ago, Motifator said:

"Lack of filtering capacitance", you'd think you know better than the engineers over Seasonic at making PSUs, ESPECIALLY at putting capacitors with proper ratings, which is one of the most basic parts of making a PSU? It's called transient load spike performance, look up on it. 

Like I said, I am not a Power Electronics guy, but I do know a thing or two about circuits. That's why I wrote 'lack of capacitance' and not 'lack of capacitors'. I do agree (and hope) that it's less likely to be the cause of transient related misbehavior on a new unit.

 

(Funny that you mention Seasonic as their older Focus platform was most known to not handle 30xx transients well. It's definitely not their fault though that Nvidia released a product with such atrocious behavior. Did they ever say how they solved it?)

 

It's a good point that poor wall outlet power quality can kill circuits and caps, Nichicon, Nippon Chemicon, and Rubicon included. So, following your statement, do we agree that electrical damage may have harmed capacitors in a 5 Y/O unit? Regardless of what they put there on first place.

 

[Motifator]

33 minutes ago, matangk said:

 

Like I said, I am not a Power Electronics guy, but I do know a thing or two about circuits. That's why I wrote 'lack of capacitance' and not 'lack of capacitors'. I do agree (and hope) that it's less likely to be the cause of transient related misbehavior on a new unit.

 

(Funny that you mention Seasonic as their older Focus platform was most known to not handle 30xx transients well. It's definitely not their fault though that Nvidia released a product with such atrocious behavior. Did they ever say how they solved it?)

 

It's a good point that poor wall outlet power quality can kill circuits and caps, Nichicon, Nippon Chemicon, and Rubicon included. So, following your statement, do we agree that electrical damage may have harmed capacitors in a 5 Y/O unit? Regardless of what they put there on first place.

 

 

I mean yeah sure, and by that yes I mean lowered capacitance, not lack of capacitors. But lack of capacitors would lead to the same thing.

Dirty power can kill a PSU in 1-2 years, let alone 5.

[matangk]

BTW just to complete the picture,

Seems like @NobleGamer may have had a specific unit (HCG) with transient sensitivity issues according to latest PSU tier list:

https://cultists.network/140/psu-tier-list/

[NobleGamer]

14 hours ago, matangk said:

BTW just to complete the picture,

Seems like @NobleGamer may have had a specific unit (HCG) with transient sensitivity issues according to latest PSU tier list:

https://cultists.network/140/psu-tier-list/

I agree that the issue could have been caused by transient sensitivity, in that the power ramped up between SotR scenes (regardless of whether power draw from GPU was ~TDP or more like TDP x2).  IIRC, I ran HWINFO in the background and don't recall a sum of wattage numbers that went much further than TDP.

 

It is not clear to me from your tier list whether there is evidence of any known issues specific to the HCG-750? Specificially, it is the 750 Bronze, and not an "M" suffix.

 

It's in Tier C under "speculative" meaning that there were no direct proper reviews found of the PSU to inform the tier assignment. I also looked at the spreadsheet it links to, and I didn't see any notes regarding known blatant issues.

[matangk]

Ah, sorry, I automatically assumed you meant the HCG Gold, which has a specific reference on the tier list:

 

For the lower tier units - there are more knowledgeable members than me that will be able to tell you about the internals of this unit, and if it can handle the demands of an RTX 3080.

 

In addition, I wouldn't count on internal monitoring of your PC to capture current spikes (if anyone knows better i'll stand corrected.) I would imagine that it samples power draw fairly slowly. What usually happens then - the reading is averaged by either the sampling circuitry or algorithm.

 

It's actually surprisingly difficult (and expensive) to capture fast current spikes.

 

Edited October 23, 2022 by matangk

[Motifator]

It is, you need a precision power meter likely for it. Software does not report fast enough.