Polaris 10 & 11 Benchmarks Leaked

[Sauron]

8 minutes ago, Fulgrim said:

Again, read ya dopey git. I have not once said OpenGL = Vulkan.

says the guy who did not bother to read beyond the fourth word.

[Humbug]

2 minutes ago, Fulgrim said:

Again, you didn't read my comments. I did NOT say Vulkan is OpenGL. It's derived from it.

I was making a general statement, sorry, not targeted at you, that's why I didn't quote you. I followed vulkan development from the start and people always get confused about this. Some of the openGL stigma such as bad drivers people expect vulkan to have the same issues which is unfounded.

[Notional]

Just now, Fulgrim said:

Again, you didn't read my comments. I did NOT say Vulkan is OpenGL. It's derived from it.

How about you define what you mean with "derive" then? Because this is going nowhere.

 

As for On Topic, the only reason I can think of for using such a poor benchmark (especially for an AMD product), must be that the chip was installed on a Linux or OSX based computer. That is if it's actually legit.

[Humbug]

2 minutes ago, Notional said:

How about you define what you mean with "derive" then? Because this is going nowhere.

 

As for On Topic, the only reason I can think of for using such a poor benchmark (especially for an AMD product), must be that the chip was installed on a Linux or OSX based computer. That is if it's actually legit.

Pre release benches often don't mean much because normally AMD and Nvidia have to tune the drivers for new architecture. n

 

 

 

[Notional]

Just now, Humbug said:

Pre release benches often don't mean much because normally AMD and Nvidia have to tune the drivers for new architecture. n

 

 

 

Indeed. I found the actual OpenGL results for 380, which of course are lower in numbers than the DX version:

https://gfxbench.com/device.jsp?benchmark=gfx40&testgroup=graphics&os=Windows&api=gl&hwtype=dGPU&hwname=AMD+Radeon+%28TM%29+R9+380+Series&D=AMD+Radeon+%28TM%29+R9+380+Series

 

Manhattan (on/off screen)

• 3460 Frames
  (55.8 Fps)
• 7603 Frames
  (122.6 Fps)

T-Rex (on/off screen)

• 3342 Frames
  (59.7 Fps)

• 29107 Frames

(519.8 Fps)

 

So according to the bench a Polaris 10 is like a 380 or equivalent. I don't buy it.

[Sakkura]

19 minutes ago, Notional said:

So according to the bench a Polaris 10 is like a 380 or equivalent. I don't buy it.

If the 380 has an optimized (ish) driver, and Polaris 10 has a crappy driver because it's not released yet, then it's perfectly reasonable.

[Prysin]

2 hours ago, Notional said:

This is very odd as they are comparing the Polaris chips running the obsolete OpenGL clusterfrack with a 380 running DirectX (11?). So you can't compare the results to begin with. Second problem is that a 380 is more powerful than a Polaris 10. No way in hell that is a thing.

this test is completely bonkers and prove or disprove nothing.

@zMeul will come in here and claim this is evidence of the second coming of failure. But that is nothing new.

[Prysin]

32 minutes ago, Notional said:

Indeed. I found the actual OpenGL results for 380, which of course are lower in numbers than the DX version:

https://gfxbench.com/device.jsp?benchmark=gfx40&testgroup=graphics&os=Windows&api=gl&hwtype=dGPU&hwname=AMD+Radeon+%28TM%29+R9+380+Series&D=AMD+Radeon+%28TM%29+R9+380+Series

 

Manhattan (on/off screen)

• 3460 Frames
  (55.8 Fps)
• 7603 Frames
  (122.6 Fps)

 

T-Rex (on/off screen)

• 3342 Frames
  (59.7 Fps)

• 29107 Frames

(519.8 Fps)

 

So according to the bench a Polaris 10 is like a 380 or equivalent. I don't buy it.

Polaris 10 and 11 is supposed to be mid range parts. So 390 equivalent and below.

If a unoptimized, and most likely not even running at retail clock speed 1024 core Polaris 10 = a 1800 something core R9 380 running at around 1GHz, well, that is not bad.

 

Atleast not if the power consumption is like 1/2 or 2/3rds of what the 380 uses.

 

Also, without proper DX drivers, and knowing most OpenCL stuff is massively unoptimized, it is very likely that there is a HUGE ass overhead for the new GPUs.

[deviant88]

wasted 2 mins of my useless time

[Sakkura]

35 minutes ago, Prysin said:

Polaris 10 and 11 is supposed to be mid range parts. So 390 equivalent and below.

If a unoptimized, and most likely not even running at retail clock speed 1024 core Polaris 10 = a 1800 something core R9 380 running at around 1GHz, well, that is not bad.

Midrange in the 400 series doesn't have to map to the 300 series that way. Maybe a midrange 400 series card will beat a 390X. We don't know yet, and this benchmark doesn't help. 

[LAwLz]

1 hour ago, Notional said:

So according to the bench a Polaris 10 is like a 380 or equivalent. I don't buy it.

Why not? To be clear, I don't trust these benchmarks, but I don't trust them because they can not be confirmed and are possibly using pre-release drivers coupled with non-final hardware. I think 380 performance from a mid-range part (Polaris 10 was suppose to be a mainstream part, right?) sounds pretty reasonable. Last gen high end card competing against current gen mid-range card sounds pretty trustworthy by itself.

 

Do you think Polaris 10 should perform better or worse? Higher than the 980 seems like a bit of a stretch to me. Lower than the 980 would be kind of disappointing.

[Notional]

2 minutes ago, LAwLz said:

Why not? To be clear, I don't trust these benchmarks, but I don't trust them because they can not be confirmed and are possibly using pre-release drivers coupled with non-final hardware. I think 380 performance from a mid-range part (Polaris 10 was suppose to be a mainstream part, right?) sounds pretty reasonable. Last gen high end card competing against current gen mid-range card sounds pretty trustworthy by itself.

 

Do you think Polaris 10 should perform better or worse? Higher than the 980 seems like a bit of a stretch to me. Lower than the 980 would be kind of disappointing.

Polaris 10 is the highest end part for the next half year or so. For it to only be 380 similar in performance would be a huge let down and make very little sense, considering AMD's focus on VR with Polaris. It simply doesn't add up.

 

I assume a Polaris 10 would beat a 390x for sure. And be similar to a fury.

[Delicieuxz]

Even if Polaris 10 is only equal to a Fury, then there's a question of what was AMD's bluster about making a presence in high-performance gaming all about?

[LAwLz]

1 minute ago, Notional said:

Polaris 10 is the highest end part for the next half year or so. For it to only be 380 similar in performance would be a huge let down and make very little sense, considering AMD's focus on VR with Polaris. It simply doesn't add up.

 

I assume a Polaris 10 would beat a 390x for sure. And be similar to a fury.

Oh crap, just realized that it was similar to 380 in performance. I was thinking of the 980. Yeah 380 performance seems a bit too low.

I wouldn't bet money on it beating a 390X and matching a Fury though. That sounds too optimistic to me. My expectation is more along the lines of 390X performance.

[shdowhunt60]

I'm willing to chalk this up to either being immature drivers, or early engineering samples with low clocks. We're still at least 3 months out before we see either Polaris 10 or 11.

[Prysin]

29 minutes ago, LAwLz said:

Why not? To be clear, I don't trust these benchmarks, but I don't trust them because they can not be confirmed and are possibly using pre-release drivers coupled with non-final hardware. I think 380 performance from a mid-range part (Polaris 10 was suppose to be a mainstream part, right?) sounds pretty reasonable. Last gen high end card competing against current gen mid-range card sounds pretty trustworthy by itself.

 

Do you think Polaris 10 should perform better or worse? Higher than the 980 seems like a bit of a stretch to me. Lower than the 980 would be kind of disappointing.

uhm, 380 = 200 USD, the 380 is slightly faster then a GTX 960.

 

AMD doesnt do like Nvidia, they use X90 for their "top end" and "x2" for their dual GPUs...

[LAwLz]

14 minutes ago, Prysin said:

uhm, 380 = 200 USD, the 380 is slightly faster then a GTX 960.

 

AMD doesnt do like Nvidia, they use X90 for their "top end" and "x2" for their dual GPUs...

See:

23 minutes ago, LAwLz said:

Oh crap, just realized that it was similar to 380 in performance. I was thinking of the 980. Yeah 380 performance seems a bit too low.

I wouldn't bet money on it beating a 390X and matching a Fury though. That sounds too optimistic to me. My expectation is more along the lines of 390X performance.

 

[CTR640]

Damn it! I thought games benches but instead, all I see is those stupid lame ass OpenGL/CL bullshit.

[Notional]

1 hour ago, Delicieuxz said:

Even if Polaris 10 is only equal to a Fury, then there's a question of what was AMD's bluster about making a presence in high-performance gaming all about?

Vega is their high end HBM2 based graphics card, launching at about the same time as NVidia's Big Pascal GP100 HBM2 based card as well.

[Doobeedoo]

Urm, well at least release is approaching and to see some actual benchmarks.

[AlTech]

9 hours ago, Notional said:

This is very odd as they are comparing the Polaris chips running the obsolete OpenGL clusterfrack with a 380 running DirectX (11?). So you can't compare the results to begin with. Second problem is that a 380 is more powerful than a Polaris 10. No way in hell that is a thing.

Under DX 11/12, Polaris 10 will be way faster. If Polaris 10 and Tonga Pro are comparable under OpenGL and DX respectively, then AMD's done a lot of OpenGL or Vulkan driver performance optimization.

[Energycore]

Are those benches showing both Polaris 10 and 11 performing worse than the 380 series? Or am I missing something?

[Humbug]

3 hours ago, AluminiumTech said:

Under DX 11/12, Polaris 10 will be way faster. If Polaris 10 and Tonga Pro are comparable under OpenGL and DX respectively, then AMD's done a lot of OpenGL or Vulkan driver performance optimization.

You can't equate openGL and Vulkan performance like that. They are two different things and the driver performance of each is unrelated to the other.

 

Anyway for this benchmark is a moot point. The vulkan and DX12 renderers for GFXBENCH are only being released in Q3 2016 so these numbers were obviously run on directX 11 or openGL.

 

I see now in the opening post that the R9 380 numbers have been run with directX 11 whereas the polaris numbers have been run with openGL. I presume this is because the DX11 benchmark was crashing under polaris which is not surprising (only AMD has access to proper polaris drivers). Also AMD normally loses to Nvidia in openGL performance badly, whereas they usually come out slightly ahead in DX11 numbers.

 

[FirstArmada]

looking at the info pages the polairs card could be missing upto 200 OpenGL functions also the Gl versions shows this:
AMD 67FF:08 : Gl version 4.4.13896 Compatibility Profile Context FireGL 16.300.0.0
AMD 67E8:00 : Gl version 4.4.13439 Compatibility Profile Context FireGL 16.200.0.0
R9 290 : GL version 4.5.13430 Compatibility Profile Context 16.150.1009.0
Gtx 950 : Gl version 4.5

considering this is FireGL these cards are most likely to be workstation cards

and the Gl version is 4.4 on the unknown cards, i am unsure if this is fake as i have not found any gpu that is a good match but the W7100 seems to be close in most tests.

either fake, a new workstation card, a gpu with some weird drivers or a unlocked firepro card?