A new ultimate mode of testing for GPUs (for w1zzard)

[birdie]

Hi,

I love TPU reviews on GPUs but I've been thinking recently that there's one particular testing mode which is worth reporting on (I've never seen other reviewers doing it).

Let me describe it:

Take a not very recent game (WoW: Mists of Pandaria), which has a high frame rate even on mid level GPUs (> 100FPS @FullHD resolution), cap the game to 60 frames per second (e.g. using VSync), then check how much energy 1) the GPU consumes 2) the whole PC (barring the display) consumes.

This test will reveal how power efficient different GPUs are.

The rationale behind this request is that many people often overlook one very important consideration: when you replace a slower GPU with a faster GPU (more power hungry), not only your new GPU consumes more energy, your whole system starts to consume more energy because the CPU has to calculate more data (more FPS = more physics, AI, etc.)

Thank you.

 

[VulkanBros]

Newer GPU´s tend too be less power-hungry - so I dont get the point??

 

[Mathragh]

In an ideal world, this idea would be the way to go.

However, because of differences in the GPUs, Drivers and even the games themselves, efficiency will always differ between games, even when locked at 60 fps.

 

[Mussels]

he already does similar with the average and maximum loads.


i've done several PC upgrades and got lower wattages in the same games, even when performance was similar.

 

[de.das.dude]

yes, he already does efficiency tests which are more of a better guesstimate.

if you are personally concerned about how much your GPU is consuming, try lowering its clocks more and more unlil your frame rates drop below 60fps.

tbh as others said, gpus are more and more efficient, for eg my previous GTS450 OC2 edition would consume 160W, current 7790 consumes 85W max and is twice as powerful, plus only 6W when idling or watching movies and stuff.

 

[GorbazTheDragon]

yes, he already does efficiency tests which are more of a better guesstimate.

if you are personally concerned about how much your GPU is consuming, try lowering its clocks more and more unlil your frame rates drop below 60fps.

tbh as others said, gpus are more and more efficient, for eg my previous GTS450 OC2 edition would consume 160W, current 7790 consumes 85W max and is twice as powerful, plus only 6W when idling or watching movies and stuff.

The whole point is to compare the efficiencies at realistic loads on less demanding games, not at 200+ FPS...

 

[OneMoar]

it doesn't matter the efficient at lower load will always be more then the efficiency at full load
I don't think you people understand that

 

[AsRock]

Hi,

I love TPU reviews on GPUs but I've been thinking recently that there's one particular testing mode which is worth reporting on (I've never seen other reviewers doing it).

Let me describe it:

Take a not very recent game (WoW: Mists of Pandaria), which has a high frame rate even on mid level GPUs (> 100FPS @FullHD resolution), cap the game to 60 frames per second (e.g. using VSync), then check how much energy 1) the GPU consumes 2) the whole PC (barring the display) consumes.

This test will reveal how power efficient different GPUs are.

The rationale behind this request is that many people often overlook one very important consideration: when you replace a slower GPU with a faster GPU (more power hungry), not only your new GPU consumes more energy, your whole system starts to consume more energy because the CPU has to calculate more data (more FPS = more physics, AI, etc.)

Thank you.

I was thinking of posting some thing like this as i nearly always use vsync but thought people on here would be to negative about the idea.

As for power usages and what w1z posts are totally pointless to me as i use vsync, there fore how much power does a 290\X take while vsync is on compared to a 970\980.

Why i use vsync ?, well it saves about 100w+ at least in most games with no real difference to my self.

 

[GorbazTheDragon]

it doesn't matter the efficient at lower load will always be more then the efficiency at full load
I don't think you people understand that

I don't think FPS scales linearly with power usage... I very much doubt it.

The whole point of a review is to see how a product will perform under normal usage, and for all we know a 290 might draw 40% of its max power at 60% load and a 970 might draw 70% at the same loading...

 

[OneMoar]

it doesn't matter what the curve is
if a GPU uses 95W at FULL LOAD then it will always use less then 95W at less then full load
that`s what you people don't seem to get
I am running a game at 125FPS and the GPU is at 50% load then I will still be drawing less then 95W
conversely I am running crysis 3 at insane setting at am getting 40FPS and the GPU is at 98% load then I will still be drawing about 95W

frame rate has nothing todo with power consumption as far as the GPU is concerned if anything it affects CPU power consumption

so if I know how much power the GPU uses at MAXIMUM load then I can always be sure It will use that much or less efficiency is just a number at a given load
a GPU is not gonna be 70% efficient at 50% load and 30% efficient at 100% load thats not how it works and since we know what the card draws at idle it doesn't frigging matter

 

[Mussels]

The whole point is to compare the efficiencies at realistic loads on less demanding games, not at 200+ FPS...

you cant reliably test that, except in a synthetic benchmark whose results will not be comparable to any game out there. I game with Vsync on and still see a 50W variation in my power consumption in the same game, and upto 100W difference depending on the game itself.

I get what you're asking for, but the reality of how GPU's work simply doesnt match up in a way that makes it worth testing. No game sits still at a certain load, except 100% - so even if you're capped at 60FPS, the power consumption will vary constantly as whats on screen changes.

 

[OneMoar]

you cant reliably test that, except in a synthetic benchmark whose results will not be comparable to any game out there. I game with Vsync on and still see a 50W variation in my power consumption in the same game, and upto 100W difference depending on the game itself.

I get what you're asking for, but the reality of how GPU's work simply doesnt match up in a way that makes it worth testing. No game sits still at a certain load, except 100% - so even if you're capped at 60FPS, the power consumption will vary constantly as whats on screen changes.

pretty much this witch is why I said we only "care" about what the efficiency is at 0% and 100%

 

[Mussels]

pretty much this witch is why I said we only "care" about what the efficiency is at 0% and 100%

Yep, i agree with you. its just a complicated thing to explain, as there is a lot of misunderstandings about how performance, FPS, heat and power consumption all tie in together. Since its poorly documented online you either rely on educated guesses, or buy a power meter and spend hours testing it all out.

 

[The Von Matrices]

I think the fundamental problem with the OP's argument is that the vast majority of people don't buy faster GPUs to run the same games at the same settings and consume less power; they buy faster GPUs to play games at higher quality settings.

If the GPU is running at low load and/or producing a framerate greater than the monitor's refresh rate, then it's time to turn up the quality settings, not tweak the power saving options.

A more realistic power use scenario compared to what the OP has posted would be to test all games with quality settings adjusted for each GPU so that all the cards produce around the same framerate, and then compare power consumption. Of course this also introduces a lot of subjectivity as to what quality settings are important not to mention an immense amount of time required to tweak the games, so I can't blame reviewers for not doing it.

 

[FordGT90Concept]

Newer GPU´s tend too be less power-hungry - so I dont get the point??

Not really. Would anyone dispute that the R9 290X of today is the HD 5870 of 2009? The former consumes 300w and the latter consumes 188w. I think the reason why graphic card consumption is going up is because the cards have to further distinguish themselves from integrated graphics that are not growing in power draw but are increasing in performance. The only way to get more performance out of chips on the same process is to increase the transistor count which translates to more power consumption.

That said, I don't see much point in benchmarking cards on WoW at 60 fps.

 

[Mussels]

Not really. Would anyone dispute that the R9 290X of today is the HD 5870 of 2009? The former consumes 300w and the latter consumes 188w. I think the reason why graphic card consumption is going up is because the cards have to further distinguish themselves from integrated graphics that are not growing in power draw but are increasing in performance. The only way to get more performance out of chips on the same process is to increase the transistor count which translates to more power consumption.

That said, I don't see much point in benchmarking cards on WoW at 60 fps.

i went from a 5870 to a 7970, got more than double the performance at lower wattage. you CAN upgrade and get a more power efficient card - but the max/average wattage numbers matter more there, than any 60FPS number.

 

[birdie]

Most people don't get what I'm asking for.

Imagine you have the same system with two wildly different GPUs, say, 290X and GTX 770.

By capping the framerate, you'll be able to see how power efficient these GPUs are in regard to the energy they require to give you the same framerate.

Many of us play with VSync enabled, so this request makes a perfect sense.

It may turn out that 290X is more power efficient in this regard than GTX 980 (jokes aside).

 

[Mussels]

Most people don't get what I'm asking for.

Imagine you have the same system with two wildly different GPUs, say, 290X and GTX 770.

By capping the framerate, you'll be able to see how power efficient these GPUs are in regard to the energy they require to give you the same framerate.

Many of us play with VSync enabled, so this request makes a perfect sense.

It may turn out that 290X is more power efficient in this regard than GTX 980 (jokes aside).

And we understand that - but its impossible to measure. the power required to run 60FPS will change every second in that game, since different things are happening.

 

[birdie]

And we understand that - but its impossible to measure. the power required to run 60FPS will change every second in that game, since different things are happening.

Are you joking??

How difficult is to run a new level of a $SAMPLE_GAME and do nothing, i.e. just run the new level and do not touch the keyboard or mouse? Then measure power consumption.

Any GPU will produce 100% identical output unless there's something happening in the game (yet I don't think it's going to be any difficult to find a more or less static scene).

 

[RCoon]

Are you joking??

How difficult is to run a new level of a $SAMPLE_GAME and do nothing, i.e. just run the new level and do not touch the keyboard or mouse? Then measure power consumption.

Any GPU will produce 100% identical output unless there's something happening in the game (yet I don't think it's going to be any difficult to find a more or less static scene).

So your grand plan is to measure GPU power consumption while VSync is turned on, while doing absolutely nothing except standing still?

That will literally provide us all with entirely useless information with regards to gaming with VSync on, because it won't technically be gaming.

 

[birdie]

So your grand plan is to measure GPU power consumption while VSync is turned on, while doing absolutely nothing except standing still?

That will literally provide us all with entirely useless information with regards to gaming with VSync on, because it won't technically be gaming.

What makes you think so? Do you really understand how GPUs are assessed?

Games benchmarking is basically measuring a performance of a given sequence of identical frames. Otherwise it's not benchmarking but a wild guesstimating.

 

[RCoon]

What makes you think so? Do you really understand how GPUs are assessed?

Games benchmarking is basically measuring a performance of a given sequence of identical frames. Otherwise it's not benchmarking but a wild guesstimating.

Ah, you mean benchmarking levels, not standing still in a game level.

Either way, measuring power consumption with VSync on will provide with such a wide range of results per second on a single benchmark it will not provide an accurate reading regardless of averages.

 

[birdie]

Ah, you mean benchmarking levels, not standing still in a game level.

No, I still want to measure power consumption for one "static" scene of any given game (with framerate capped). That will give 100% reproducible results.

"Static" is quoted because in modern games many parts of the scene usually have some slight movement (e.g. leaves on trees swaying, in game characters nodding or otherwise moving).

 

[RCoon]

No, I still want to measure power consumption for one static scene of any given game (with framerate capped). That will give 100% reproducible results.

Yeah but those 100% reproducable results will not mean anything to anyone, because nobody plays a game standing still for 60 seconds. You'd be measuring something that doesn't need to be measured. It is entirely irrelevant information to gamers, because they will never being doing it.

 

[birdie]

God, are you talking to yourself?

I'm asking to measure GPU's power consumption for a static in game scene to assess its computational efficiency. What you do in games is only your concern. This topic is not about you or your ideas.

You'd be measuring something that doesn't need to be measured.

Do you have any rationale behind this statement?

Different GPUs have very similar graphs for their performance (i.e. ups and lows):

So by measuring power consumption for a single frame, you can easily assess how power efficient a given GPU is.