AMD Polaris 10 Samples work at 1.27 GHz

[Azix]

if they have improved async compute to the point that graphics and compute can run completely in parallel, not sharing resources, then they would have a huge advantage the greater % of compute is used in games. It would be like having double the GPU.

That would require widening whatever bottleneck was there with async before. Wonder if thats possible.

 

[xthetenth]

This strategy is still very risky since NV is going to launch GP106 sooner or later, before Vega drops. NV will close the low end and mainstream Pascal gap far quicker than AMD will close the $350+ gap with Vega. That's one big risk for AMD.

Being the only show in town for a wide fraction of the market benefits them a lot more than NV, so it may very well be worth it.

 

[tviceman]

Under DX12, small Vega will decimate the GTX 1080.
Big Vega will beat big Pascal. Polaris will cover the rest of the market.

AMD has just as much a problem with its fans and their out-of-control expectations as they do with launch issues/impressions. Then when the product is released, disappointment always sets in followed by floods of excuses and how future releases will be better because XYZ.

 

[thesmokingman]

AMD has just as much a problem with its fans and their out-of-control expectations as they do with launch issues/impressions. Then when the product is released, disappointment always sets in followed by floods of excuses and how future releases will be better because XYZ.

AMD is far from perfect but they're not the only ones who get it wrong.

 

[4K_shmoorK]

AMD has just as much a problem with its fans and their out-of-control expectations as they do with launch issues/impressions. Then when the product is released, disappointment always sets in followed by floods of excuses and how future releases will be better because XYZ.

Could be said about fans from both sides, all things considered. Seems there's a growing emotional component among red & green fans these days.

That's something I'll never understand.

 

[boozzer]

I am not getting into the hype. waiting on benchmarks.

 

[dark zero]

I feel that 1060 won't reach even 1.5 GHz trying to be as strong as a theorical 970Ti.

AMD has a chances with Polaris 11.

Also... on Mobile I expecting a rebrand of all the GTX 1040M to below and the rest will come on Fall 2016.

On AMD side is a massive failure what they are doing... just they released Polaris 11 at R7 450M and Polaris 10 on R9 490M.... and no rebrands....

Also.. the DP and QP from Pascal is a massive dissapoint. And if Tesla can't deliver the improvement on that... is likely the end of GPGPU.

 

[sirmo]

AMD has just as much a problem with its fans and their out-of-control expectations as they do with launch issues/impressions. Then when the product is released, disappointment always sets in followed by floods of excuses and how future releases will be better because XYZ.

Weren't they right though? GCN cards have aged better than Kepler.

 

[tviceman]

AMD is far from perfect but they're not the only ones who get it wrong.

In many cases it is faster than 980 SLI. I'm not going to waste my time looking up reviews and averaging out scores, but I'd expect at 4k 1080 is faster on average vs. 980 SLI.

 

[thesmokingman]

In many cases it is faster than 980 SLI. I'm not going to waste my time looking up reviews and averaging out scores, but I'd expect at 4k 1080 is faster on average vs. 980 SLI.

lol

 

[MrTeal]

Well, not perfect due to different benchmark games and drivers, but there's this.

980 SLI is 67% faster than a single 980 in the SLI review, and GTX 1080 is pretty much exactly the same amount faster than a 980. Now, the SLI review included a lot of quite old games and only two didn't have good SLI scaling. However 980 SLI was also 14% faster than 970 SLI, which is included on the new review. You'd expect SLI scaling between 970 SLI and 980 SLI to be pretty similar, so that's probably a better comparison. The 1080 is 35% faster than 970 SLI, so in that mix of games you'd expect GTX 1080 to be ~15-20% faster than 980 SLI.

 

[thesmokingman]

Well, not perfect due to different benchmark games and drivers, but there's this.

980 SLI is 67% faster than a single 980 in the SLI review, and GTX 1080 is pretty much exactly the same amount faster than a 980. Now, the SLI review included a lot of quite old games and only two didn't have good SLI scaling. However 980 SLI was also 14% faster than 970 SLI, which is included on the new review. You'd expect SLI scaling between 970 SLI and 980 SLI to be pretty similar, so that's probably a better comparison. The 1080 is 35% faster than 970 SLI, so in that mix of games you'd expect GTX 1080 to be ~15-20% faster than 980 SLI.

You guys are doing their marketing for them. They prefaced their slide by saying only in VR is it 2X faster.

http://www.geforce.com/hardware/10series/geforce-gtx-1080

 

[Phynaz]

Under DX12, small Vega will decimate the GTX 1080.
Big Vega will beat big Pascal.

Source?

 

[flopper]

You guys are doing their marketing for them. They prefaced their slide by saying only in VR is it 2X faster.

http://www.geforce.com/hardware/10series/geforce-gtx-1080

Had a blind dog and his noose was torn, he couldnt find his way around the kitchen.
we put him down.

Not so easy reading now and then.

 

[Arachnotronic]

Source?

404 Not Found.

 

[Armsdealer]

For what it's worth, the owner of wccftech, khalid moammer, went into the comments section on his recent Polaris / computex article, and stated with a high degree of conviction that ref board core clock would be between 1300-1400 Mhz.

The reason he has so much conviction is because he claimed he has "seen them"

If he's right, it should be a very strong contender against 1070, and it wouldn't be unjustified to use g5x as AMD has hinted for some parts.

 

[MrTeal]

You guys are doing their marketing for them. They prefaced their slide by saying only in VR is it 2X faster.

http://www.geforce.com/hardware/10series/geforce-gtx-1080

What marketing? GTX1080 isn't 2x faster than a 980, but it is faster than 980SLI. That's less to do with the "irresponsible amount of power" inherent in Pascal, than the poor state of multi-GPU. In that first SLI review from Sept 2014, 970 SLI was 69% faster than a single card at 4k. In this latest 1080 review, 970 SLI is only 48% faster than a single card at the same resolution.

 

[Erithan13]

At least it paints a better picture than the leaks a while back suggesting 800-1000mhz clocks (maybe that's for the lower end/mobile variants). Depending on how conservatively clocked the cards are 1400-1500mhz overclocks may be in reach which would definitely help narrow the gap between the top Polaris 10 and the 1070. Also makes Polaris look like a decent step up from 390 level performance rather than a sidegrade.

 

[thesmokingman]

What marketing? GTX1080 isn't 2x faster than a 980, but it is faster than 980SLI. That's less to do with the "irresponsible amount of power" inherent in Pascal, than the poor state of multi-GPU. In that first SLI review from Sept 2014, 970 SLI was 69% faster than a single card at 4k. In this latest 1080 review, 970 SLI is only 48% faster than a single card at the same resolution.

I think you're missing the point. It's not outright faster in all situations and you'd really have to look hard because it's not out there yet. That's why its really more marketing speak which lends to propping up of the the legend. It's liek the 2x faster charts, in fine print in one specific instance. But ppl will take away what, zomg its 2x faster, zomg its faster than sli in all cases.

 

[IEC]

404 Not Found.

While he may not have a source, it's pretty obvious a 450mm2-600mm2 FinFET chip with a (slight) density advantage over a 300mm2-320mm2 FinFET chip should perform far better.

However, none of the posts dealing with this topic are "on-topic" as this thread is about Polaris. Not Vega.

 

[Armsdealer]

At least it paints a better picture than the leaks a while back suggesting 800-1000mhz clocks (maybe that's for the lower end/mobile variants). Depending on how conservatively clocked the cards are 1400-1500mhz overclocks may be in reach which would definitely help narrow the gap between the top Polaris 10 and the 1070. Also makes Polaris look like a decent step up from 390 level performance rather than a sidegrade.

Narrow the gap? How many compute units do you think Polaris 10 has? If it's 40, and you can reach 1500 mhz on boost, you're looking at a 7.7 TFLOP part vs 1070 at 6.1 (boost).

While I appreciate that it doesn't necessarily equate 1:1 to fps, I can assure you at that discrepancy, p10 will be faster than 1070.

The relevant question now is how many compute units does the uncut die have.

 

[airfathaaaaa]

AMD is far from perfect but they're not the only ones who get it wrong.

there is wrong and there is missinformation

https://www.youtube.com/watch?v=UKkFqG77-x4
glorious as they say "dx12 example of dynamic load"
while the fraps is running
D:

 

[Arachnotronic]

While he may not have a source, it's pretty obvious a 450mm2-600mm2 FinFET chip with a (slight) density advantage over a 300mm2-320mm2 FinFET chip should perform far better.

However, none of the posts dealing with this topic are "on-topic" as this thread is about Polaris. Not Vega.

Do we even have die sizes for Vega?

 

[thesmokingman]

there is wrong and there is missinformation

https://www.youtube.com/watch?v=UKkFqG77-x4
glorious as they say "dx12 example of dynamic load"
while the fraps is running
D:

^^I think that is closer to disinformation tbh.

 

[RussianSensation]

The 1080 is 35% faster than 970 SLI, so in that mix of games you'd expect GTX 1080 to be ~15-20% faster than 980 SLI.

Off-topic to this thread but 1080 is not 15-20% faster than 980 SLI.
http://www.gamespot.com/articles/nvidia-geforce-gtx-1080-review/1100-6439863/

Paul from Newegg tested it too and 980 OC SLI outperformed 1080 @ 1.885mhz by even greater amounts.
https://www.youtube.com/watch?v=kYdzZzSzF2I

Of course 1080 is still a better buy than 980 SLI due to consistent performance of 1 card but this also means 1070 SLI will be a better buy than 1080.

For what it's worth, the owner of wccftech, khalid moammer, went into the comments section on his recent Polaris / computex article, and stated with a high degree of conviction that ref board core clock would be between 1300-1400 Mhz.

The reason he has so much conviction is because he claimed he has "seen them"

If he's right, it should be a very strong contender against 1070, and it wouldn't be unjustified to use g5x as AMD has hinted for some parts.

Some new specs have been floating around.

P10 = 2048 SPs @ 1340mhz or so, 224GB/sec bandwidth 150W TDP
http://www.pcper.com/news/Editorial/New-AMD-Polaris-10-and-Polaris-11-GPU-Details-Emerge

There is also a new leak on P11 = 470X.

"The Chinese website Ithome have published some of the R9 470X has to be. It is not entirely clear how the site came to this information, so it is wise to take this rumor with a grain of salt. Nevertheless Polaris getting closer and source it therefore might have to be right.

The R9 470X will be based on Polaris GPU 11. That is positioned slightly lower than Polaris 10 , which is believed to be the R 9 480 (X) will have to be controlled. Higher positioned questions will follow later in the form of Vega . The announcement follows Vega than later. These rumors just going about the R9 470X, the lowest positioned Polaris card.

According to the source the chip has 1280 shader units, 80 TMU's and 40 ROPs. The GPU does its work at about 1 GHz, which is much lower than the (higher positioned) 16nm models from Nvidia. There's 4GB of GDDR5 memory available, which has a 128-bit wide memory bus is controlled. The R9 470X must therefore perform similar to the R9 380 / 380X, though AMD has the TDP of the card while it heavily in know constrain. According to Chinese sources, the R9 470X has a TDP of only 60 watts - about a third of its counterparts.

The website also reported a sales price: 1100 yuan, which translates to about 150 euros."
Netherlands Hardware Info

If true, I am not too impressed. That means we'll be stuck between mid-range Polaris 10, the biggest gimp on x70 card and a minimum $600 USD GTX1080 cards (most of which I predict will be priced well above $600).

With Total War Warhammer getting excellent reviews, those fire sale R9 390s for $230-250 after rebates and the TW:W bundle could accidentally become the best value for 1080p 60Hz gaming.

Do we even have die sizes for Vega?

Last leak showed 27.37m2, Base Clock of 3300 with a peak GPU Boost of 9000.