Polaris Reportedly Doubles Performance Per Watt

[ivan134]

26 minutes ago, SuperShires said:

But that makes no sense, some people like less heat emitted, me being one of them also less noise because of said heat reduction. I dont get your arguement of "if you want a lower wattage GPU go buy a console or laptop"

What I'm saying is that we should be wanting increased performance with this lower power consumption advantage, not same performance with lower consumption side grades. A 390 and 970 can barely do 60 fps in ROTTR at 1080p with the settings people would prefer to use and obviously games are going to get more demanding in the future as graphics tech improves. We should be wanting the next $300 - $350 cards to perform like a Fury.

[Real_PhillBert]

GPU manufacturers always seem to make really grand claims then fall short on delivery.

 

I'll just wait for the launch to see what it actually does instead of arguing based on speculation.

[SuperShires]

3 minutes ago, ivan134 said:

What I'm saying is that we should be wanting increased performance with this lower power consumption, not same performance with lower consumption side grades. A 390 and 970 can barely do 60 fps in ROTTR at 1080p with the settings people would prefer to use and obviously games are going to get more demanding in the future as graphics tech improves. We should be wanting the next $300 - $350 cards to perform like a Fury.

Yeah from that angle I agree, my point was some people like efficiency so it wouldnt be all bad if they did the same performance less power, but that wont be the case I dont think haha. Well hopefully they will be on par with the Fury.

[Watermelon Guy]

1 hour ago, MEC-777 said:

They also said the Fury X would be "an overclocker's dream"...

They were right about the effiency gains of the fury x. 1.5 times over the r9 290x.

16 minutes ago, patrickjp93 said:

You forget Nvidia is getting a double node shrink too. I doubt we're going to see Thermi Part 3.

Thermi was about power consumption, a smaller process node means each core is more efficient, but if nvidia shoves a lot of cores in it will still use a lot of power.

[patrickjp93]

2 minutes ago, Watermelon Guy said:

They were right about the effiency gains of the fury x. 1.5 times over the r9 290x.

Thermi was about power consumption, a smaller process node means each core is more efficient, but if nvidia shoves a lot of cores in it will still use a lot of power.

But Nvidia has also probably learned from its mistakes with Fermi. Furthermore, I see it being much more like Kepler given we already know peak double precision is 4 teraflops. I'm guessing that will put single precision at 12 Teraflops. Volta goes to 7 DP, so if anything, that will be like Fermi.

[ivan134]

4 hours ago, patrickjp93 said:

You forget Nvidia is getting a double node shrink too. I doubt we're going to see Thermi Part 3.

No, I didn't forget. Yea, Fermi was unfortunate to be made on a 40 nm node, otherwise the architecture was pretty good. It was just ahead of its time.

 

What I'm talking about is the people who believe this myth that Nvidia has figured out something about efficiency that AMD hasn't. Hawaii used roughly the same amount of power as high end Kepler and you know how Maxwell's efficiency was achieved since you mentioned Fermi. AMD even took the same approach with GCN 1.2, but still have more compute units than Maxwell and didnt throw out their hardware schedulers.

 

Pascal and Polaris will be roughly equal in power consumption since Nvidia themselves have already given us confirmation they're not skimping on double precision. The only thing that was worrisome about that article was that they said they were having problems bringing consumption under control iirc and I'm paraphrasing right. Nvidia has some good engineers though and they should be able to pull through.

[-BirdiE-]

How does power consumption (generally) scale with transistor size? I thought it was pretty linear (meaning this claim would make sense).

 

I really don't know though.

[iamdarkyoshi]

Just now, -BirdiE- said:

How does power consumption (generally) scale with transistor size? I thought it was pretty linear (meaning this claim would make sense).

 

I really don't know though.

HBM seems to be what has taken the power consumption down quite a bit, and optimizations in architecture.

 

Take a look at these CPUs:

http://ark.intel.com/products/33910/Intel-Core2-Duo-Processor-E8400-6M-Cache-3_00-GHz-1333-MHz-FSB

http://ark.intel.com/products/42801/Intel-Pentium-Processor-E5700-2M-Cache-3_00-GHz-800-MHz-FSB

 

Both of them have a 65W TDP, 3ghz clock speed, two cores, and the same Lithography size (45nm)

 

But the core2duo manages a pretty decent increase in performance over the pentium on passmark

http://www.cpubenchmark.net/cpu.php?cpu=Intel+Core2+Duo+E8400+%40+3.00GHz&id=955

http://www.cpubenchmark.net/cpu.php?cpu=Intel+Pentium+E5700+%40+3.00GHz&id=1101

 

Intel improved how the CPUs chug, and nearly doubled the transistor count on the core2duo, even though the TDP and Lithography size was the same.

[patrickjp93]

6 minutes ago, -BirdiE- said:

How does power consumption (generally) scale with transistor size? I thought it was pretty linear (meaning this claim would make sense).

 

I really don't know though.

It stopped being linear at 90nm. Dennard Scaling was the name for it, but it stopped being true a long time ago.

[-BirdiE-]

3 minutes ago, patrickjp93 said:

It stopped being linear at 90nm. Dennard Scaling was the name for it, but it stopped being true a long time ago.

Thanks!

[Nielips]

Looks like so poeple don't understand how performance per watt is worked out.....

[DocSwag]

I personally think efficiency is a bad argument unless it is because your power supply can't power it and you would have to buy a new PSU too.

You'd be much better off buying a few LED light bulbs and replacing some bulbs with those!

Of course if you already have LED bulbs then....

[MEC-777]

I used to say I didn't care about power consumption and only mainly about performance. That was when I ran a single R9 290. Then I ran a pair of 290's in CF and oh my word, the heat! I had to water cool them to keep temps in control (and in a case with very good air flow). I've since switched to a single 980 (which delivers about 75-90% of the performance of the CF 290's) and oh my word, the silence! lol. Runs low-demanding game while remaining totally passive and temps are very well controlled with tons of cooling overhead available. Performance still takes a high priority but I now care a lot more about power efficiency than I did before and I think a lot of other people do as well. 

 

It boils down to personal preference, but there is something to be said about the benefits of power efficiency.

[W-L]

4 hours ago, N3v3r3nding_Newb said:

According to the website "mobipicker," AMD's head, Srinivasan M., said that Polaris will double performance per watt.

http://www.mobipicker.com/amds-polaris-to-double-performance-per-watt-india-head-srinivasan-m/

Please edit your post to follow the Posting Guidelines for the Tech News and Reviews Section.

 

 

[Dabombinable]

But can it compute?

[Coaxialgamer]

Great. Noiw let's all wake up and stop believing rumors. On a side note , i would be nice, but they did say the fury x would be an "overclocker's dream" , not to mention all the BS surrounding the fx launch.

[Master Disaster]

So the nuclear GPUs now go thermonuclear. 2x PPW and 2x the heat output.

[Watermelon Guy]

5 hours ago, MEC-777 said:

They also said the Fury X would be "an overclocker's dream"...

They were right about the effiency gains of the fury x. 1.5 times over the r9 290x.

4 hours ago, patrickjp93 said:

You forget Nvidia is getting a double node shrink too. I doubt we're going to see Thermi Part 3.

Thermi was about power consumption, a smaller process node means each core is more efficient, but if nvidia shoves a lot of cores in it will still use a lot of power.

[Deli]

6 hours ago, Delicieuxz said:

Does lower wattage also mean quieter fan cooling?

 

I hope Polaris is a home-run for ATI, by way of dominating in performance over Nvidia. Graphics hardware has been stagnating, and increasing by meager amounts, for far too many years.

I won't say that. From 980Ti to 980 is like 25-30% faster. Just think how much faster 980Ti to, say a 580. Comparing Skylake to Haswell....yeah, I mean if Intel gives us 25-30% jump of IPC every generation.

[Delicieuxz]

25 minutes ago, Deli said:

I won't say that. From 980Ti to 980 is like 25-30% faster. Just think how much faster 980Ti to, say a 580. Comparing Skylake to Haswell....yeah, I mean if Intel gives us 25-30% jump of IPC every generation.

 

A 580 to 980Ti is like 5 years to achieve just 2.2 times the performance. That's not all that great. Graphics technology progression really slowed down with the 360 / PS3 console generation, as everything was targeted for those machines - and the major slowdown in graphics tech progression hasn't stopped from then until now. I was hoping Pascal / Polaris would finally be the end of it, but Nvidia seem to be really dragging their feet on doing anything meaningful for graphics power. The new normal is is all about milking consumers with minimal progression, while the consequence is that graphics could have been more than twice what they are at this point, if the 360 / PS3 slowdown hadn't occurred.

[Deli]

1 hour ago, Delicieuxz said:

 

A 580 to 980Ti is like 5 years to achieve just 2.2 times the performance. That's not all that great. Graphics technology progression really slowed down with the 360 / PS3 console generation, as everything was targeted for those machines - and the major slowdown in graphics tech progression hasn't stopped from then until now. I was hoping Pascal / Polaris would finally be the end of it, but Nvidia seem to be really dragging their feet on doing anything meaningful for graphics power. The new normal is is all about milking consumers with minimal progression, while the consequence is that graphics could have been more than twice what they are at this point, if the 360 / PS3 slowdown hadn't occurred.

You sound like Nvidia and AMD owe us if they do not double their GPU performance every year.

 

How about there is only 30% IPC improvement from Sandy to Skylake. Boycott Intel?

[patrickjp93]

3 hours ago, DocSwag said:

I personally think efficiency is a bad argument unless it is because your power supply can't power it and you would have to buy a new PSU too.

You'd be much better off buying a few LED light bulbs and replacing some bulbs with those!

Of course if you already have LED bulbs then....

Tell that to server and supercomputer architects. Sorry but modern dGPUs are not designed around gaming.

[patrickjp93]

6 minutes ago, Deli said:

You sound like Nvidia and AMD owe us if they do not double their GPU performance every year.

 

How about there is only 30% IPC improvement from Sandy to Skylake. Boycott Intel?

There's a doubling of performance from Sandy Bridge to Skylake. Blame consumer software developers for not keeping up with newer instruction sets.

[Maxxtraxx]

6 hours ago, ivan134 said:

No thanks. I'd rather have more performance for the same wattage. We're talking about having one less light bulb in your house worth of savings.

Agree on the performance on the same wattage... but i've fully switched over to LED bulbs sooo.... at 7 watts per bulb (lets say 150 watts reduction(1/2 of 300watts) that puts me at 21 light bulbs of more power usage)

[Maxxtraxx]

53 minutes ago, Delicieuxz said:

 

A 580 to 980Ti is like 5 years to achieve just 2.2 times the performance. That's not all that great. Graphics technology progression really slowed down with the 360 / PS3 console generation, as everything was targeted for those machines - and the major slowdown in graphics tech progression hasn't stopped from then until now. I was hoping Pascal / Polaris would finally be the end of it, but Nvidia seem to be really dragging their feet on doing anything meaningful for graphics power. The new normal is is all about milking consumers with minimal progression, while the consequence is that graphics could have been more than twice what they are at this point, if the 360 / PS3 slowdown hadn't occurred.

well, a 780Ti is roughly equal to a gtx970... which when put into SLI is a bit more powerful than a 980Ti... so from 780ti to 980ti is maybe 1.7x the performance? not quite double but thats darn impressive on the same manufacturing node!