Hardware & Technical Rig makers ... Graphics Card - PSU question

Caspar

Caspar
Hi - could do with some friendly advice about gaming rig building please :)

I threw a rig together using a mix of old kit and new. It's basically this:

AMD R9 280 (Sapphire dual X)
AMD FX 6350
Asus Mobo
8 GB RAM (1 DIMM)

All shoved in to an old case with a 480 WATT PSU.

Now it's worked fine since I've had it up and running and I can play ED at Ultra settings 1080p on my monitor - but I've recently bought an Oculus Rift DK2 which really hammers the box. The problem I'm having is that even on the low settings using the Rift, the FPS drop below 75ps, loses vsync and then the experience gets very choppy (if you use the rift, you'll understand that FPS is everything).

It's probably just that my card isn't up to it - but I did notice something and I wanted an opinion from someone who knows better about these things than me.

So I get good FPS when the card is below 70% GPU utilisation but anything after that and I start to notice frame drops. Is that normal for these cards (i.e. the performance drops off in the final third) or could something else be happening? I've also noticed that I'm not seeing the GPU climb above 90% (see image below). It's left me wondering - could the issue be that the card cannot draw enough power?

Another way of asking the questions is - if the GPU is underpowered, will the card blue screen / produce artifacts etc - or would the GPU just under perform?

I'd really appreciate a view point from someone who knows something about this. I'll grab a new PSU like a shot if there's a real chance it would make a difference. :)

View attachment 21241
 

adoredtv

adoredtv
It's not your PSU that's causing it. Your components demand power which your PSU can either provide or it can't. If it can't provide the power it'll shut down at best or catch fire if it's a really low quality one.

A more likely cause is the CPU just isn't powerful enough to keep the GPU fed, though you appear to be getting pretty good utilisation out of it. It's kinda complicated but the AMD CPUs can be limited on one core if the game requires a single strong thread (most games do currently).

Finally, an R9 280 isn't really capable of running the rift smoothly at 75 fps - no single card is, not even a GTX 980.
 

huwthomas

huwthomas
Not sure if this is relevant, but the AMD Catalyst drivers let you boost or throttle how much power the card will use as part of the performance/clock configuration.

Also I wonder if it self-throttles if it gets too hot?
 

Caspar

Caspar
adoredtv said:
It's not your PSU that's causing it. Your components demand power which your PSU can either provide or it can't. If it can't provide the power it'll shut down at best or catch fire if it's a really low quality one.

A more likely cause is the CPU just isn't powerful enough to keep the GPU fed, though you appear to be getting pretty good utilisation out of it. It's kinda complicated but the AMD CPUs can be limited on one core if the game requires a single strong thread (most games do currently).

Finally, an R9 280 isn't really capable of running the rift smoothly at 75 fps - no single card is, not even a GTX 980.
Click to expand...

Thanks for the points but I'll disagree with the last one. It depends on the level of detail that you define in the options but I think plenty of rigs with that tier of card (GTX 980) are able to have a smooth ED rift experience.

I agree with your view on PSU. That's how I'd expect it to work. I'm just not quite understanding why I'm not touching the last 10% GPU performance and why I'm seeing a dip in performance from about 70% utilisation upward. I don't think this card is short of memory bandwidth but it seems to hit an exponential wall in the final third of the GPU. If it's CPU support of the GPU that's lacking then it's sad to see it happening at 60% util (6 x 40% remaining).
 
Last edited:

adoredtv

adoredtv
Caspar said:
Thanks for the points but I'll disagree with the last one. It depends on the level of detail that you define in the options but I think plenty of rigs with that tier of card are able to have a smooth ED rift experience.
Click to expand...

Yes clearly if you drop settings you can, and in ED you can get a decent experience but if you check the VR forum you'll see that few are entirely satisfied with the performance.

I'm just not quite understanding why I'm not touching the last 10% GPU performance and why I'm seeing a dip in performance from about 70% utilisation upward. I don't think this card is short of memory bandwidth but it seems to hit an exponential wall in the final third of the GPU. If it's CPU support of the GPU that's lacking then it's sad to see it happening at 60% util (6 x 40% remaining).
Click to expand...

Like I said, if the CPU is being held back due to its single-thread performance then that is where the likely bottleneck is.

If you take your CPU with it's 70% utilisation, that's a nice and easy ~4 cores used out of 6 (which is really quite good for a game using DirectX). However if the game code demands a single strong thread/core (the vast majority of current DirectX games do), then your CPU will start to choke on that and that will bring down the maximum performance of your graphics card.

ED performs pretty well on AMD CPUs though.
 

norcus

N
Caspar said:
Thanks for the points but I'll disagree with the last one. It depends on the level of detail that you define in the options but I think plenty of rigs with that tier of card (GTX 980) are able to have a smooth ED rift experience.

I agree with your view on PSU. That's how I'd expect it to work. I'm just not quite understanding why I'm not touching the last 10% GPU performance and why I'm seeing a dip in performance from about 70% utilisation upward. I don't think this card is short of memory bandwidth but it seems to hit an exponential wall in the final third of the GPU. If it's CPU support of the GPU that's lacking then it's sad to see it happening at 60% util (6 x 40% remaining).
Click to expand...
Run some stress tests and see if you can't get 100% utilization. Just to rule out the PSU. I concur with you both that if the PSU did not deliver enough power you would experience more severe symptoms. However, 480W is cutting it close so you might be better off with a new PSU in any case :)
 

Caspar

Caspar
norcus said:
Run some stress tests and see if you can't get 100% utilization. Just to rule out the PSU. I concur with you both that if the PSU did not deliver enough power you would experience more severe symptoms. However, 480W is cutting it close so you might be better off with a new PSU in any case :)
Click to expand...

Yes. I'm right on the edge it seems. The book requirement for the card is 500W min. I'm guessing that rated values and actual are not always the same and that actual delivery probably comes in under the rated value - especially when the PSU is a decade old. So I've ordered a 650 Watt supply to give a bit of head room.

The more I read about the card though, the worse it sounds. I bought the thing in a hurry thinking that dual-x thing meant that it was better than 280x. But no, it was a marketting term for the dual fans made up by Sapphire. The 280 is just a rebranded older Radeon. I suspect it will get replaced sometime soon.
 

adoredtv

adoredtv
It's a good card and overclocks like a beast but it'll never run the Rift on high settings except for older games.
 

Caspar

Caspar
->
->
->
Using HWMonitor - PSU shows that the 12v rail drops to as low as 11.076v

PSU still not an issue?

I understand that components can throttle / scale in low power situations rather than just shut off.
 
Last edited:

adoredtv

adoredtv
CPUs and GPUs throttle down mostly due to heat which is an artifact of them running at max performance, and therefore power - but they won't throttle down if a PSU can't deliver the power. The instant a PSU can't deliver the power asked of it, you'll get some kind of PSU-related issue (normally shutting down if it's any kind of halfway-decent PSU).

Monitoring voltages with software is unreliable at best tbh - I don't want to say that you don't need a new PSU because you might - however you will not see one jot of performance improvement if you buy it - but you might get some peace of mind.

You can run OCCT PSU test and it'll stress both CPU and GPU to the max or as close as it can possibly get. If you see near 100% GPU (you should get over 120 fps by my reckoning) and CPU then you'll know that your PSU can handle the load.

If your PSU doesn't shut down or start acting weirdly then it's highly unlikely that it has an issue. Just be aware that if your PSU *does* have an issue that the potential is there for damage to occur. If you're careful then you should be ok as most reasonable-quality PSUs have some kind of protection built in.
 
Last edited:

adoredtv

adoredtv
norcus said:
Intel CPUs will also throttle back when they reach TDP (rated max power) I think.
Click to expand...

A typical i5/i7 desktop CPU can easily reach 150W with a large overclock, but it won't throttle if you stick a good cooler on it. It depends on how good the cooling is really, which is why laptops throttle much more easily.

There's some good info on Intel's turbo boost that gives some more indication of how this works - http://www.intel.com/content/www/us...ology/turbo-boost/turbo-boost-technology.html
 
Last edited:

adoredtv

adoredtv
Caspar said:
There were a couple of interesting quotes from Nvidia on pasted in this thread as well:
http://www.tomshardware.co.uk/forum/246614-28-question-power-supply-affect-performance

Plus a couple of anecdotes that suggest power has been a factor for some.

On the basis that it's over a decade old, i'm swapping it out anyway. If by some miracle it does resolve anything, you'll be the first to know.
Click to expand...

That's a classic case of the poster misinterpreting what Nvidia says.

"The Geforce 8800GTS requires a minimum 400W or greater system power supply with a 12V current rating of 26A. While most Geforce 8800GTX users power supply meets or exceeds the minimum wattage requirements, the power supply may fail the minimum current necessary to achieve consistant framerates during heavy use of the graphics card."

It should be read as "During heavy gaming where the graphics card is delivering consistent (high) framerates, the power supply needs to be able to provide enough continuous current to the card or it will fail".

Anyway, toms is pretty bad for tech advice. This is a good article that explains a lot about PSUs - http://www.enthusiastpc.net/articles/00002/4.aspx

He even mentions this in it -

When components are drawing too much power

While we're on the subject of overcurrent protection, let's put one persistent myth to rest. Every so often I read comments on bad graphics performance during gaming that talk about the power supply as a possible cause of performance degradation. This is a false assumption! If a videocard draws too much power from a power supply there are only these possible outcomes:


  • The overcurrent protection of the power supply kicks in, causing the power supply to cut power to your system and your computer turns off instantly. This is what is supposed to happen.
  • The overcurrent protection does not kick in but the power supply is not able to maintain a stable 12V output. This will cause a hard crash of your entire system that will present as a freeze and your screen will show either a garbled image or your screen will go black. This is not what is supposed to happen but if you are using a cheap power supply it is entirely possible that it will happen nonetheless. This can possibly cause damage across your entire system!
  • The overcurrent protection does not kick, the voltage stays stable and your power supply and/or the wires running between the power supply and the graphics card start running very hot. This is a dangerous situation (FIRE HAZARD!) and can cause the power supply to ultimately fail and potentially take part or all of your system with it!

These are the only possible outcomes of an underpowered PSU.
Click to expand...

It's good that you're getting a new PSU anyway - that one you have probably doesn't owe you anything at 10 years old. I'm assuming you had to hook up extra adapters for it, ie an extra 6-pin or two?
 

Briarée

Briarée
norcus said:
Intel CPUs will also throttle back when they reach TDP (rated max power) I think.
Click to expand...
this can be disabled tho, and shouldn't be required unless you O/C. (some mobo also allow to increase slightly TDP limit)

also it's not really a throttle, tdp act on turbo, trying to maintain MAX turbo while staiying within TDP, thermal throttle will reduce freq to "idle" freq. (Threshold is 70-75°C on most recent intel) and ultimately, if temp is too high (above 95°C i think), will shut down.

turbo and 1st throttle can be disable, afaik the 2nd safety (shut down) cannot.

the thermal part works pretty much the same on gpu, only with different threshold, most are able to run normally up to +-90°c, throttle around 100°C shut down above.

of course threshold are gpu / cpu specific, and it is, of course, advise to never reach the first one
 
Last edited:

norcus

N
It is interesting to see that now the CPU is the heat bottleneck not the GPU. At least on my system. GPU and CPU are on the same waterloop and the CPU (i7 4790k@4.4GHz) will shoot up to 80C and increase to around 85C during stresstests, but the GPU (780 OC 1150MHz) never goes above 60C. Granted, ambient temp is probably slightly higher in the CPU area it is still a big difference.

I wonder if the shrinking process now faces a new challenge. Maybe heat dissipation is lower since the density of transistors are so high. Devil's Canyon are 22nm, but the GPUs are still at 28nm. Skylake will start at 14nm?
 

adoredtv

adoredtv
The transistor density is actually much higher on GPUs (7 billion transistors in your 780 for example and even though it's a very large chip at ~550mm2 it still has a much higher density than most Intel CPUs). Clocks are also much lower of course, a typical GPU around 1GHz while CPUs are nearer 3.5/4GHz.

Not entirely sure what's going on with Intels processes but they haven't really gone anywhere since Sandy Bridge on 32nm. Wish I still had my 2500K actually. It's likely a combination of moving closer to mobile and further from ultimate performance (a process tweaked more towards lower power than higher clock speed means that the further you try to push the clocks the more the power/heat runs away from you), the smaller transistors, FinFETs probably making it worse too then they had that TIM fiasco with Ivy Bridge as well on top. Haswell again is far better at low power but really quite mediocre at the performance level. I've long since given up caring about Intel CPUs (now have a 3570 non-K after swapping my 2500K with my girlfriend), just hope AMD can get back in the game with Zen and maybe force Intel to shape up a bit.
 
Last edited:

norcus

N
adoredtv said:
The transistor density is actually much higher on GPUs (7 billion transistors in your 780 for example and even though it's a very large chip at ~550mm2 it still has a much higher density than most Intel CPUs). Clocks are also much lower of course, a typical GPU around 1GHz while CPUs are nearer 3.5/4GHz.

Not entirely sure what's going on with Intels processes but they haven't really gone anywhere since Sandy Bridge on 32nm. Wish I still had my 2500K actually. It's likely a combination of moving closer to mobile and further from ultimate performance (a process tweaked more towards lower power than higher clock speed means that the further you try to push the clocks the more the power/heat runs away from you), the smaller transistors, FinFETs probably making it worse too then they had that TIM fiasco with Ivy Bridge as well on top. Haswell again is far better at low power but really quite mediocre at the performance level. I've long since given up caring about Intel CPUs (now have a 3570 non-K after swapping my 2500K with my girlfriend), just hope AMD can get back in the game with Zen and maybe force Intel to shape up a bit.
Click to expand...
Aha

I did a little research and made the following calculations:
ChipTDP - WattDie Size - mm²Transistor Count - billionsTransistor density -bt*/mm²Watt/mm²Watt/bilion transistors
Geforce GTX7802505507.10.01290.4535.21
Core-i7 4790k881771.40.00790.4962.86
* billion transistors

Transistor density of the GPU is infact 50% higher than the CPU, but each transistor uses a lot less power. Still the Watt/mm² is not that different. Maybe GPU manufacturers simply use better quality heatsink and TIM.
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom