PSU wattage numbers are nonsense

[Chen G]

I'm not sure how most people feel about this now days but I've always felt a lot of people tend to buy too big of a PSU for their build so I've been careful not to do that myself.

However, I'd like to suggest anyone looking to just go big, because PSUs are cheap, the wattage numbers are nonsense, and if you don't get enough you get a lot of headaches later on.

 

So a few years ago I got started with my current ultra silent build, I went for one of those Seasonic fanless "600W" PSUs. I thought that would be enough because I was running a ~130w CPU and ~250w GPU. even with some CPU overclocking let's say 250w, I'd still only max out at 500w so "600W" continuous power should theoretically be enough, right? 

 

It wasn't enough

 

the PC would just randomly reboot when running some games. 

so I had to swap over to a Corsair HX850i, now I'm at "850w" continuous power, and it's been fine. And with the PSU reports I could see my power usage never reached 500w.

Last year, I picked up an RTX3090 to replace the 2080ti I had, now I'm at 350w, 100w more than before. But still, the 3950x CPU I have is pretty efficient, I'm not even overcloking it really. So still theoretically, I should still be well below 850w. In fact the PSU tells me I'm never reaching 600w.

 

Guess what, it wasn't enough

 

Well it is enough for most games, in fact I could run 3Dmark with both overclocking CPU and GPU. But it's not enough for Cyberpunk 2077 2.0

I started getting those same old reboots again. I thought my clocking had become unstable or something, but no, I 've eliminated all those variables. The only thing that will prevent the reboots is 90% GPU power limit, with the same clocks.

 

So even though the reported power usage from my PSU is ~500w while running the game, I in fact need, more than 850w continuous power PSU to keep it stable. I'm not sure why they even bother with the PSU wattage numbers, it's basically meaningless. Next time I'm just getting a 1000w.

 

and yea, I could totally see how if you're running a 13900k and 4090, 1200w+ is NOT overkill.

[RollinLower]

3 minutes ago, Chen G said:

So even though the reported power usage from my PSU is ~500w while running the game, I in fact need, more than 850w continuous power PSU to keep it stable. I'm not sure why they even bother with the PSU wattage numbers, it's basically meaningless. Next time I'm just getting a 1000w.

isn't this due to transient spikes? the RTX3xxx series are notorious for those.

[tkitch]

no, it's not the PSUs that're the problem.

 

It's the GPUS.  When a GPU Using 250W suddenly spikes to 400 or 500W of draw?  That can put a HUGE strain on things.  

 

My 7 year old 750W CPU is running an Overclocked i9 10850k and an AMD RX 9700 XT just fine.

[Tetras]

35 minutes ago, Chen G said:

So a few years ago I got started with my current ultra silent build, I went for one of those Seasonic fanless "600W" PSUs. I thought that would be enough because I was running a ~130w CPU and ~250w GPU. even with some CPU overclocking let's say 250w, I'd still only max out at 500w so "600W" continuous power should theoretically be enough, right? 

Funnily enough, fanless/SFF PSUs are often built way overspec, due to the size/thermal constraints, but Seasonic had big problems when the high transient GPUs hit the market and had to replace/revise their Focus Plus line, so I wonder if that was part of your issue there.

 

36 minutes ago, Chen G said:

Well it is enough for most games, in fact I could run 3Dmark with both overclocking CPU and GPU. But it's not enough for Cyberpunk 2077 2.0

I started getting those same old reboots again. I thought my clocking had become unstable or something, but no, I 've eliminated all those variables. The only thing that will prevent the reboots is 90% GPU power limit, with the same clocks.

 

So even though the reported power usage from my PSU is ~500w while running the game, I in fact need, more than 850w continuous power PSU to keep it stable. I'm not sure why they even bother with the PSU wattage numbers, it's basically meaningless. Next time I'm just getting a 1000w.

ATX 3.0/PCIE5 PSUs are supposed to address this problem, so you shouldn't (in theory, at least) have to buy a PSU that is way overspec. If their marketing for these "power excursions" actually works out in reality, I don't know.

[Chen G]

2 hours ago, tkitch said:

no, it's not the PSUs that're the problem.

 

It's the GPUS.  When a GPU Using 250W suddenly spikes to 400 or 500W of draw?  That can put a HUGE strain on things.  

 

My 7 year old 750W CPU is running an Overclocked i9 10850k and an AMD RX 9700 XT just fine.

I would agree, but regardless of who's fault it is, the end result is my 850w PSU cannot properly power my system drawing ~500w average.

 

Older CPU and GPUs basically run all out all the time, they don't have power limits, their TDP basically is their maximum power. Where as modern chips can theoretically run much more but are limited to their TDP number, so I guess the power limiters don't react instantly so sometimes you get power spikes.

 

Again theoretically, if they can put some truly gigantic capacitors in the PSU then you'd be able to run much closer to your actual power requirements, but since there's no spec for any of that, you only get one power wattage number, so from now on I'm defiantly going more to the save side.

[--SID--]

1 hour ago, Chen G said:

the end result is my 850w PSU cannot properly power my system drawing ~500w average

Is the PSU set at 12v single rail or 12v multi rail mode?

 

Do you use extensions on the original cables between the PSU and GPU or replacement cables?

[Beerzerker]

2 hours ago, Chen G said:

I would agree, but regardless of who's fault it is, the end result is my 850w PSU cannot properly power my system drawing ~500w average.

 

Older CPU and GPUs basically run all out all the time, they don't have power limits, their TDP basically is their maximum power. Where as modern chips can theoretically run much more but are limited to their TDP number, so I guess the power limiters don't react instantly so sometimes you get power spikes.
 

There is alot of variance in accuracy of wattage capacity between PSU makes and models so that's one factor you have to include.
That's one reason why a PSU made by "Whizz-BANG!" (No-name brand) for example is just that and will eventually, if not immediately live up to it's name but a good, quality unit will be more accurate per stated wattage rating and do what it's supposed to.
Never skimp - Always get a good quality unit.
 

I've always spec'ed mine with extra wattage capacity for that reason and a couple of other reasons too.
 

It's NOT good to have a unit running at 100% capacity all the time because that has the PSU unit working hard all the time, which will lead to increased internal thermals because heat is the enemy of anything electronic and will shorten it's lifespan.
Another reason why is if you starve the system for power if it's calling for it, that in turn makes everything in the system start running hotter too.

Try placing a teeny-tiny battery into a vehicle that needs a larger battery or try starting one with a really weak battery - It's fry and die time when that happens as evidenced by the vehicle's starter getting really hot, really fast.

Once power is introduced into any power circuit, it has to go somewhere and if it's not dissapated as work done by the device, it dissapates as heat and that's why.

The power circuitry itself will not store this energy if it's not designed to and if not, that's what happens.

It's also due to wattage overhead you'll need at times to handle spikes in load/demand for power too, if the unit is already running "At Capacity" and THEN a spike occurs that's not a good thing, even though they are engineered to absorb spikes.

The total wattage of what the system will (What you'd expect) have to run should be added up and then you'd want to figure in about a 20% increase above that total value for the entire system to detemine the actual PSU wattage you'd need in a unit.

2 hours ago, Chen G said:

 

Again theoretically, if they can put some truly gigantic capacitors in the PSU then you'd be able to run much closer to your actual power requirements, but since there's no spec for any of that, you only get one power wattage number, so from now on I'm defiantly going more to the save side.

When you talk about "Safe" I'm assuming you're thinking the PSU per wattage rating will cram excess power into the system - PSU's DO NOT work like that.

It only supplies the power the system needs at the time you'd measure power draw, based on load demand at that time.
That's why for example, I'm running a 1600W unit (Rosewill Hercules) right now in a system I use daily that could easily be handled by a 800W unit....
I'm typing this on that system so that means the system is working and not affected by the excess wattage capacity, in fact I've been running this one for several years this way and 0 issues to even mention about it.

PSU's don't just cram their rated wattage into the system, it simply provides what power the system is wanting at any given moment, which can vary wildly sometimes (Power spikes) if you were to be measuring power draw all the time - You'd see it to know it.

Wattage ratings only means the unit is (Supposedly) capable of that wattage output if it's needed and if it's not, it won't supply that amount of wattage, only what the system wants to run.
You can go too small and cause what I've described above AND make components in the system itself get hotter too but you really can't go "Too large" and have it to blow the guts out of the system when you press the power button - If that ever happens you already have a bigger problem than the wattage rating of your PSU to start with.

Again, PSU do not cram power into the system, it's simply supplies what the system needs up to it's rated capacity.

So - Add up the entire expected wattage draw from the system and be aware everything in it will draw power such as the board/chipsets, RAM and all else, not just the CPU and GPU alone.
Figure all that up, then spec it for about 20% over that value (Or more if you want) and go with it.
Even just 10% over is better than having right at the limit all the time.
I hope this clear up some things for you.

EDIT:
To the one that thinks this is funny.....
I've worked with electrical systems, power panels, electric drive motors in an industrial setting and so on for over 30 years...
And I did this too.....
https://hwbot.org/submission/4740676_bones_superpi___1m_fx_8320_9sec_141ms
Way to show your own ignorance, which is laughable in itself.
Have fun with that.

[Chen G]

2 hours ago, --SID-- said:

Is the PSU set at 12v single rail or 12v multi rail mode?

 

Do you use extensions on the original cables between the PSU and GPU or replacement cables?

it is single rail, there's no extension, the custom silver cables go straight to the PSU

1 hour ago, Beerzerker said:

When you talk about "Safe" I'm assuming you're thinking the PSU per wattage rating will cram excess power into the system - PSU's DO NOT work like that.

It only supplies the power the system needs at the time you'd measure power draw, based on load demand at that time.

yea I know, obviously.

What I think is happening is the PSUs have over current protection that detect a over current, even just for 5ms (from power spike) and it shuts down the system.

I just think a bigger PSU will have their protection set at higher current so it won't trip and reboot.

1 hour ago, Beerzerker said:

So - Add up the entire expected wattage draw from the system and be aware everything in it will draw power such as the board/chipsets, RAM and all else, not just the CPU and GPU alone.
Figure all that up, then spec it for about 20% over that value (Or more if you want) and go with it.
Even just 10% over is better than having right at the limit all the time.
I hope this clear up some things for you.

my original spec if just counting the CPU and GPU, was 120+250=370w, way under the 600w PSU rating. Even adding in RAM and chipset I'm pretty sure I'm under 450w, which is also confirmed by PSU report I got after switching PSU. That's over 25% headroom, and it wasn't enough.

 

with the 3090 upgrade, I added 100w to the GPU and maybe 10 on SSD. So let's say 560w theoretical, again confirmed from PSU report, that's about as high as it goes, way under 850w PSU, this time with 34% headroom, it wasn't enough.

 

so I'd say add 50% headroom for spikes.

 

[Beerzerker]

2 hours ago, Chen G said:

it is single rail, there's no extension, the custom silver cables go straight to the PSU

yea I know, obviously.

What I think is happening is the PSUs have over current protection that detect a over current, even just for 5ms (from power spike) and it shuts down the system.

I just think a bigger PSU will have their protection set at higher current so it won't trip and reboot.

my original spec if just counting the CPU and GPU, was 120+250=370w, way under the 600w PSU rating. Even adding in RAM and chipset I'm pretty sure I'm under 450w, which is also confirmed by PSU report I got after switching PSU. That's over 25% headroom, and it wasn't enough.

 

with the 3090 upgrade, I added 100w to the GPU and maybe 10 on SSD. So let's say 560w theoretical, again confirmed from PSU report, that's about as high as it goes, way under 850w PSU, this time with 34% headroom, it wasn't enough.

 

so I'd say add 50% headroom for spikes.

 

Looks like you've got it in the ballpark, that's good to hear.
That was the main thing I had about it, as long as all that has been taken into account with what you get/have you should be fine.

[Blue4130]

3 hours ago, Chen G said:

What I think is happening is the PSUs have over current protection that detect a over current, even just for 5ms (from power spike) and it shuts down the system.

I just think a bigger PSU will have their protection set at higher current so it won't trip and reboot.

Yes, it's doing exactly what it is supposed to do. The problem is the the GPU spiking up to 550w acusing the psu to trip.

[Beerzerker]

16 hours ago, Blue4130 said:

Yes, it's doing exactly what it is supposed to do. The problem is the the GPU spiking up to 550w acusing the psu to trip.

Could be overdrawing the PSU on the rail the GPU uses, making it trip even though the unit's wattage is supposted to be enough for the entire system.
That could mean what's being used to power the GPU is a weak rail so it's just not holding up, causing the trip.
That's one reason why I've always preferred a single rail PSU, or at least one with a few rails as possible.

[Blue4130]

3 hours ago, Beerzerker said:

Could be overdrawing the PSU on the rail the GPU uses, making it trip even though the unit's wattage is supposted to be enough for the entire system.
That could mean what's being used to power the GPU is a weak rail so it's just not holding up, causing the trip.
That's one reason why I've always preferred a single rail PSU, or at least one with a few rails as possible.

Sure, but the psu is still doing what it should be doing. It's not a fault on the PSU. Read the spcs on the PSU, it will tell you what each rail is rated for.

[Beerzerker]

I agree - You're probrably right about that.
I do know each rail is supposed to handle a given amount of power (Rails are power rated per rail on the unit's label) before tripping and you're rightly pointed out the GPU is spiking the rail, causing the trip..... Just as it's designed to do.

Yes the PSU IS doing what it's supposed to be doing. 
Maybe the OP could look into another PSU unit with a better/stronger rail on that part of it for their GPU or maybe even a different GPU that doesn't have as much power demand instead, a different PSU is the cheaper overall option of course and what I'd look into first. 
It's up to the OP on what to do here.

[freeagent]

My 5900X and EVGA 3070TI cannot be used with my Prime GX-750. In game I will get reboots, so a light to medium load. And again when running F@H. System would draw 650w and shut off. I would have to trip the PSU switch to get the system going again. My EVGA PSU, also a 750 can drive that load for weeks at a time with no problem. That GX-750 is running that 3070Ti and a clocked 5600X with no troubles.

[NobleGamer]

2 hours ago, freeagent said:

My 5900X and EVGA 3070TI cannot be used with my Prime GX-750. In game I will get reboots, so a light to medium load. And again when running F@H. System would draw 650w and shut off. I would have to trip the PSU switch to get the system going again. My EVGA PSU, also a 750 can drive that load for weeks at a time with no problem. That GX-750 is running that 3070Ti and a clocked 5600X with no troubles.

I think this example means that OPs issue comes down to how PSUs handle transient spikes or something technical & brief that HWInfo won't show it easily.

Heck I even have a kill-a-watt meter at the outlet and I don't know that it could pickup GPU transient spikes.

Sadly transient spikes seem like a black box unless you have specialized lab or electrician equipment.

 

On 10/3/2023 at 11:45 AM, Chen G said:

so I had to swap over to a Corsair HX850i, now I'm at "850w" continuous power, and it's been fine. And with the PSU reports I could see my power usage never reached 500w.

Last year, I picked up an RTX3090 to replace the 2080ti I had, now I'm at 350w, 100w more than before. But still, the 3950x CPU I have is pretty efficient, I'm not even overcloking it really. So still theoretically, I should still be well below 850w. In fact the PSU tells me I'm never reaching 600w.

 

Guess what, it wasn't enough

Weird, I have a 3 pcie plug EVGA RTX 3080 and 5600X with ~520W total power draw and an RM850x with zero issues.  Running it on an old 750W PSU caused hitching when loading new scenes in a SoTR benchmark.  Are the spikes for the 3090 that much worse?  Worst case scenario is that it can transient spike 525w (450 from 3x pcie connectors, 75 from pcie slot) which is only 175W more than 350W - Doesn't seen like a lot more to burden your presumably higher quality Corsair HX model.

[freeagent]

1 hour ago, NobleGamer said:

Heck I even have a kill-a-watt meter at the outlet and I don't know that it could pickup GPU transient spikes

It shouldn't see it, happens too fast. I did mention it would happen to me in game too. For what its worth, my 3070Ti can do 325w ( I want to say 340) with OC @ F@H. With my 5600X and 58X3D it is fine, but my tuned 5900X with that GPU/PSU combo is a bad idea. Like I said, my evga PSU can drive the same load and play the same games.

[AI_Must_Di3]

@Chen G Great post! I have a EVGA 3090 FTW3 Ultra V2 with a EVGA x299 Dark/10900X setup and when i bought the 3090 i had to tear down my rig, pull out the prior psu and put in the 1500 watt unit i had, but never used. The transients are fleeting, but with a 3090 they are high pulls and with the sensitivity of PC's anymore and the ever so tight tolerances were up against they will trip protections. With the 1500 unit i havent had any of those problems and my rig runs great...unless im tinkering with it of course.

Also, the 1500 watt PSU is a Coolermaster Silent Pro M2 V1 that was brand new in a box for years before i used it, it was supposed to go to a sli build that i didnt do, but its a silver rated unit with the old ATX spec and it runs this build like a champ, not a hiccup to be had. So that shows that you dont need a titanium rated, every bell and whistle, newest atx spec to have a solid running rig. As we all know, PSU's run at the 50-70% level of their total wattage will give you the best efficiency returns, cause the last thing you should be doing is running your psu to its peak. IMO, the high power transient spikes issue has really put a exclamation mark on that last sentence.

[Bagzie]

On 10/3/2023 at 4:45 PM, Chen G said:

I'm not sure how most people feel about this now days but I've always felt a lot of people tend to buy too big of a PSU for their build so I've been careful not to do that myself.

However, I'd like to suggest anyone looking to just go big, because PSUs are cheap, the wattage numbers are nonsense, and if you don't get enough you get a lot of headaches later on.

 

So a few years ago I got started with my current ultra silent build, I went for one of those Seasonic fanless "600W" PSUs. I thought that would be enough because I was running a ~130w CPU and ~250w GPU. even with some CPU overclocking let's say 250w, I'd still only max out at 500w so "600W" continuous power should theoretically be enough, right? 

 

It wasn't enough

 

the PC would just randomly reboot when running some games. 

so I had to swap over to a Corsair HX850i, now I'm at "850w" continuous power, and it's been fine. And with the PSU reports I could see my power usage never reached 500w.

Last year, I picked up an RTX3090 to replace the 2080ti I had, now I'm at 350w, 100w more than before. But still, the 3950x CPU I have is pretty efficient, I'm not even overcloking it really. So still theoretically, I should still be well below 850w. In fact the PSU tells me I'm never reaching 600w.

 

Guess what, it wasn't enough

 

Well it is enough for most games, in fact I could run 3Dmark with both overclocking CPU and GPU. But it's not enough for Cyberpunk 2077 2.0

I started getting those same old reboots again. I thought my clocking had become unstable or something, but no, I 've eliminated all those variables. The only thing that will prevent the reboots is 90% GPU power limit, with the same clocks.

 

So even though the reported power usage from my PSU is ~500w while running the game, I in fact need, more than 850w continuous power PSU to keep it stable. I'm not sure why they even bother with the PSU wattage numbers, it's basically meaningless. Next time I'm just getting a 1000w.

 

and yea, I could totally see how if you're running a 13900k and 4090, 1200w+ is NOT overkill.

Wattage numbers are irrelevant but next time you are going for 1000w.

 

Wait how did that conclusion come about?