Blogs Battle of the GPUs: Is power efficiency the new must-have?

I think there's a lot more to this topic.

Maxwell was designed to be very scalable architecture, usable for mobile and all the way up. Nvidia decided to do this because they have substantial laptop and mobile interests, and it makes sense at this point for them to have a single fully fledged architecture which can be used across all those platforms, AMD on the other hand have no such interests. Given that their mobile Adreno division was sold off to Qualcomm (in 2009 for 65millionUSD), they don't have the same incentive to produce such a scalable architecture. They will far more likely aim to produce a high performance design capable of out-doing Maxwell, but bare in mind, these architecture decisions were probably made 3-5 years ago, it's not like AMD at this point can say "Ahh look guys those Nvidia fellas have done a power efficient one, maybe we should do that with ours too". The decisions are made and the chips are well on their way to being released. They will already know whether they are going to be able to compete on power efficiency and my guess will be they can't. That's not a bad thing, they've got a different target for the architecture, and as long as it does compete with Maxwell sufficiently, they can always price it to an appropriate point in the market.

Another point, People are very fast to jump on the "GTX 480 is the work of satan" bandwagon. I still have 2 in my desktop, I know that puts me "behind the times" by modern standards, but in applications which scale well I comfortably get GTX 680 levels of performance. Gaming at 1920x1200 this is plenty of performance, and to get any meaningful upgrade would mean considering likely a GTX 970, or a pair of/equivalent, which would be largely wasted at this resolution.
As for temps, they're really not that bad. Yes they're power hungry, yes they run a little hotter than some (but not all) GPUs, but they're perfectly manageable. It's not like I'm trying to hold my desktop against my face while using it, I would rather it used all it's available TDP to give me as much performance as my hardware is capable of.

The only factor which sadly limits this setup is the lack of Vram for games like Watchdogs and other games written and then loosely ported back from the modern consoles. It was obvious several years ago when the rumors began to circulate about the consoles having 8GB of ram shared across the GPU and CPU that ports would emerge which had been lazily carried out and would need a larger amount of Vram than was present on the average cards of the day. For some time now when friends have asked me for a spec which is "future proof" I have advised they purchased cards with at least 4GB of Vram (the typical amount developers allocate to the GPU on the consoles). As such friends with, for example, a pair of GTX 680 4GB cards can still utilise all of the shading horsepower without becoming bottlenecked by the lack of available on card storage.

As for 4k, I have tried several screens and found the vast majority to be lacklustre. The Dell 24" IPS screen was nice but the resolution was wasted with such a high ppi, the 28" is a non IPS, TN panel (and therefore vomit inducing) and the Asus 31.5" is nice but still priced well out of the reach of most consumers. I would consider that 4k is still an unwise investment at the current time, It requires a shift in the consumer for what they consider to be acceptable graphically. Even with very powerful setups it requires compromise, which is not something that people who can afford this technology are used to experiencing. GPU's have not yet scaled to be sufficiently large to handle the resolution and frankly without the finalised display connectors being readily available on the market at the moment, and while we wait to see the adoption of FreeSync, it would be nigh on foolish to purchase an expensive high end panel at the current time, simply opt for a quality IPS 1440p panel instead.

 

For me power efficiency is irrelevant. As Gareth demonstrates the efficiency has negligible impact on a Gamer from a cost perspective. I'd rather have a space heater that munches through graphics than something mildly better than the previous gen running mildly cooler than the previous gen.

(Insert Clarkson Powah image here. )

 

I agree with you on that point. I use watercooling though so noise isn't really an issue for me.

 

My 1200w hcp PSU isn't even breaking a sweat with my 2 770's, still won't when I get the third... it probably heats the basement up a little. for a mini itx rig I would definitely want a powerful, low wattage, quite card. But it will be a generation or two before I worry about that.

 

Going SFF, low power gives a massive bonus: SFX PSUs are expensive, and there's only so much heatsink you can cram into a small case without screaming-banshee fans.
I can run a Sandy-bridge Xeon and a 970 comfortably on a 300W supply (and I've probably got some room to overclock the 970). If I wanted to go for a 290x (comparable performance), I'd need to purchase a new PSU, my choice being the Silverstone 450w unit (infamous for being very loud), or the Silverstone 600w unit (pretty quiet, but £100).
By not needing a new PSU, power savings directly translate to cost saving before I've even plugged the thing in. Quieter too.

If I was still going for a 'big rig' things would be different, with a larger PSU and a LOT more airflow available, and the option of SLI.

 

All these comments seems to say the same thing, that efficiency is indeed a great benefit but it is not a justifiable reason for an 'upgrade' for this reason alone.

 

Value for money comes first for me. My gpu doesn't spend enough time at 100% for a more power-hungry card to affect my electricity bills.

 

Well that depends on your circumstances, doesn't it? Not sure why you're so keen to draw a definitive conclusion on this subject: like so many things with PC hardware, it's horses for courses and they're isn't a one size fits all answer.

Edit: in response to SchizoFrog rather than hyperion

 

Couldn't give a monkeys about heat.
All I read when I bought my 970 on launch day was the performance figures.
If it runs cooler and more efficiently, bonus, but it has never and will never bother me.
If it gets warm, open window/turn fan on.

 

Also this.

 

Yeah sorry, as mentioned earlier I am tired (and still haven't slept). My brain gets 'locked-on' from time to time.

 

And if you are in the market for GTX 970/80, I would hazard a guess that not one person ever considers the efficiency/heat.
I think only if AMD had an equivalent performing card for similar money, different story.
But they don't.
Yet.

All I look for is frame rates.
My test is if a game I absolutely love starts lagging or is getting close to lag with all the main eye candy on, time for an upgrade.

 

You realise that 970s can pull close to 200W right? That's going by BTs article, if you go by Toms and look at what theirs pulled during their stress test, it's closer to 240W. Even if you ignore Toms as an outlier, most places I've seen have them still pulling around 180W-200W. More than their quoted 145W by Nvidia. You could blame that on the lack of a reference option and partners going off and doing their own thing, but even the reference 980 pulls more than it's stated 165W.

 

Peak draw != average load. And unless your PSU is particularly awful (and in the case of the 970/980, your workload is entirely Compute rather than gaming), average load is what you need to worry about.

 

Their gaming test averages range from 160W with TPU running an Asus 970, to 182W again on TPU running a Palit 970. Toms Gigabyte is at 177W.

While I agree no game is going to run like a torture test, they do show what we assume to be the absolute max the cards can pull. Now on TPU, their peak Furmark draw is actually lower than their peak gaming draw. On Toms torture test, the reference 980 actually stays around 180W, the Gigabyte 970 and 980 average 240W and 280W.

Now it's all well and good if a game isn't as power intensive as the one they used in their examples, lets face it using less power isn't really an issue. What happens though if a game uses more than their examples?

 

165W TDP + GPU boost %
Palit 110% boost
Asus: weird throttling (bad HS contact?)
Gigabyte: 107% boost (107% seems to be the voltage lock, so this may be a particularly bad chip)

My 970 is boosting to 110% with no stability issues (and the boost limit reason is almost always TDP rather than VREL, which is a good sign), meaning it should be drawing on average 182W at peak non-compute load, leaving a comfortable 118W for the rest of the system. I won't be modifying the BIOS for higher boosts until I have a wattmeter though.

 

http://www.tomshardware.co.uk/core-i7-4790k-devils-canyon-overclock-performance,review-32968-9.html

that's toms showing a stock 4790k @ 123watts - so chip alone above the rest of your system

 

It does to me... I replaced my HD5870 that was dying and replaced it for a MSI GTX750 Ti, it has the same performance (+/-) and it's good enough to play at 1080p.

 

Isn't that 123 watts total system draw though, averaged through thousands of data point (excluding idle), peak power draw of the whole system was around the 330-340 mark and that's including a Titan GPU that Toms own data shows draws up to 250w.