[WCC] AMD Rolling Out New Polaris GPU Revisions With 50% Better Perf/Watt

[Concillian]

Are people confused over what 50% means? A 100% improvement is half the power consumption at the same performance. 50% improvement is 2/3rds the power consumption at the same performance (1/1.5 ==> 2/3). This is not outside the realm of physical possibility, as this would actually bring the RX480 to be in line with that of the 1060.

[H] had the stock speed 480 and 1070FE at essentially equivalent power consumption with the 1070 offering quite significantly more performance, I mean we all know that the 1070 is way faster. Tom's has the 480 consuming 164 watts in Metro Last Light Redux while the 1060 is consuming 119 watts in the same game, and they have roughly similar overall performance (1060 slightly better in this particular case.)

Increasing the perf/watt of the rx480 strictly by offering the same performance at lower power consumption would mean the 480 performs the same at 110w instead of 164w. Hey, guess what, if you figure 50% is the corner "best case scenario we use for marketing that can be reproduced in one condition in one game if our lawyers make us" and real world in an average game is more like 35-40ish%, then you're suddenly right in line with the 1060 power consumption. From the perspective of someone who is an engineer in a not unrelated industry, that seems completely rationally possible. Especially if the RX480 was rushed out before it was really ready for prime time due to competitive pressures. I've certainly been involved in similar product releases, and these are often followed up with later revisions that offer significant improvement in yield and cost.

So I'm not sure why people think a 50% perf per watt improvement is unrealistic. Yes, it's very good to be skeptical of this kind of leaked info. Temper enthusiasm and such, but still it's not like AMD would be bending physics here. It's a large change, but it wouldn't be like they were breaking new ground here. This improvement wouldn't make them miles better than nVidia, or even better at all. It would merely bring them IN LINE with nVidia perf / watt.

 

[ConsoleLover]

From the latest Vega information it seems as they are able to run a GPU on almost twice the shaders on 175W, basically double the performance of the RX 480 for 175W, if they can do that, than surely with more mature process and better binning they can reduce power draw. If you look at the youtube video for the new gpu from evga, its running at 90W when at reference (oc evga) speed and up to 120W(max) when overclocked additionally.

 

[Headfoot]

Are people confused over what 50% means? A 100% improvement is half the power consumption at the same performance. 50% improvement is 2/3rds the power consumption at the same performance (1/1.5 ==> 2/3). This is not outside the realm of physical possibility, as this would actually bring the RX480 to be in line with that of the 1060.

[H] had the stock speed 480 and 1070FE at essentially equivalent power consumption with the 1070 offering quite significantly more performance, I mean we all know that the 1070 is way faster. Tom's has the 480 consuming 164 watts in Metro Last Light Redux while the 1060 is consuming 119 watts in the same game, and they have roughly similar overall performance (1060 slightly better in this particular case.)

Increasing the perf/watt of the rx480 strictly by offering the same performance at lower power consumption would mean the 480 performs the same at 110w instead of 164w. Hey, guess what, if you figure 50% is the corner "best case scenario we use for marketing that can be reproduced in one condition in one game if our lawyers make us" and real world in an average game is more like 35-40ish%, then you're suddenly right in line with the 1060 power consumption. From the perspective of someone who is an engineer in a not unrelated industry, that seems completely rationally possible. Especially if the RX480 was rushed out before it was really ready for prime time due to competitive pressures. I've certainly been involved in similar product releases, and these are often followed up with later revisions that offer significant improvement in yield and cost.

So I'm not sure why people think a 50% perf per watt improvement is unrealistic. Yes, it's very good to be skeptical of this kind of leaked info. Temper enthusiasm and such, but still it's not like AMD would be bending physics here. It's a large change, but it wouldn't be like they were breaking new ground here. This improvement wouldn't make them miles better than nVidia, or even better at all. It would merely bring them IN LINE with nVidia perf / watt.

People are skeptical because of the source, which is a known bad source which makes stuff up all the time. This exact same information coming from somebody trustworthy would be something else entirely.

 

[SpaceBeer]

Bigger chips/cards always have better perf/watt ratio than smaller ones (GM104 vs GM106, GP104 vs GP106, Fiji vs Tonga...). So the only way to compare Pascal and Polaris is to compare P10 and GP106. And their perf/price ratio depends on test or workloads. It's not the same when you compare them in gaming or mining or some GPGPU task (single or double precision). Of course, vast majority is interested only in gmaing test, since these cards are usually used for gaming. And P10 has maybe same perf/watt ratio as GTX 970. But there are some cases in which P10 is ~2x more efficeint than GTX 970
https://techaltar.com/amd-rx-480-gpu-review/2/

And little bit on topic - it's hard to believe these rumors. But the fact is that 2x larger Vega chip with maybe 8 GB of HBM2 instead of GDDR5 could have 2x better performance with ~230W TDP. It is also impossible to et P10 with 100W TDP and same performance with little bit modified GF 14nm process