Nvidia Pascal Reviews [1080XP, 1080ti, 1080, 1070ti, 1070, 1060, 1050, and 1030]

You don't need to. GP104 does not offer double-rate FP16.
Oh don't spoil it. I'm still collecting numbers. This is hilarious.

Assuming SiSoft is using OpenCL for that benchmark, I don't think it's going to reveal fp16 support unless NVIDIA's driver implements cl_khr_fp16.

Compubench doesn't report any support for that extension in the NVIDIA OpenCL driver (which is not a surprise).
SiSoft can use OpenCL, DX-CS, OpenGL, or CUDA. I can confirm the CUDA test works exactly as I need it to, and that it's using CUDA's FP16 instructions.
 
Looking at the Hardware.fr perf/W numbers, we see the 1080 having a 2.25-2.3x ratio over Fury X.

As expected, the much heralded perf/W improvements for Polaris will simply help it to catch up with Pascal.

I don't see any reasons for optimism to expect that a Polaris based laptop design will be more efficient than one that is GP106 based.

The tests from late 2014 may not be fully equivalent with today, but drilling back to Hardware.fr's numbers for Tonga's introduction shows the 1080 is just about 2.5x as efficient as the 285 at that time for BF4.
 
I can confirm the CUDA test works exactly as I need it to, and that it's using CUDA's FP16 instructions

Great! I'm surprised it would work unless there was a lot of foresight. Don't make us wait! :-|

Does FP16x2 exist in GP104 or are we getting an old FP32-only architecture thrown at our feet and making us cry?

keep_america_beautiful_tear.0.gif
 
Last edited:
I don't think I've seen that claim made on this forum... But we've definitely seen AMD hype Polaris as the second coming of Christ in the perf/W department.

Sorry. Didn't meant to imply that it was you or this board that said that, though it definitely looks like I was doing that.
I browse a lot of different forums and the general consensus of the ~2-2.5x perf/w increase of Polaris would make it even with Maxwell while Pascal would still have an advantage.
 
The tests from late 2014 may not be fully equivalent with today, but drilling back to Hardware.fr's numbers for Tonga's introduction shows the 1080 is just about 2.5x as efficient as the 285 at that time for BF4.
Yes, Fury X is really the best case.

It's going to depend on the test case which one will be ahead, but it's going to be too close to really matter.
 
Excellent card for a mediocre price. I'm not sold on the latest VR hype wagon but GP104's straight up performance and power efficiency is fantastic.

Now we wait a year or more for the true king. AMD, get your ass in the game.
 
To me it seems a lot like the Radeon 7970 launch. That averaged around 25% faster than the 580 at launch for a little more money. Nvidia have done better here, it's as if AMD had launched the 7970GHz edition up front at a price of $500 instead of $550. Still, the rough performance increase vs power decrease is comparable to the 7970 IMO. In other words it's pretty much what is to be expected for a mid sized chip on a new process.
 
To me it seems a lot like the Radeon 7970 launch. That averaged around 25% faster than the 580 at launch for a little more money. Nvidia have done better here, it's as if AMD had launched the 7970GHz edition up front at a price of $500 instead of $550. Still, the rough performance increase vs power decrease is comparable to the 7970 IMO. In other words it's pretty much what is to be expected for a mid sized chip on a new process.

Yep. And pretty fun to compare reviewer conclusions of the GTX1080 to the HD7970 conclusions.
 

Well I was more specifically talking about SkyMTL's take.

Had lots of questions about Tahiti and GCN... that one would think he would have similar questions and concerns about GP104 and Pascal, though to be fair he does admit that Pascal isn't perfect.
He slightly balked about the Tahiti MSRP but gave a "Damn good Value" award to GTX1080.
http://www.hardwarecanucks.com/foru...s/49646-amd-radeon-hd-7970-3gb-review-25.html
http://www.hardwarecanucks.com/foru.../72619-nvidia-geforce-gtx-1080-review-25.html
 
Yep. And pretty fun to compare reviewer conclusions of the GTX1080 to the HD7970 conclusions.

The 1080 is being compared to the 980 in most reviews I've seen and there it has a 60%+ advantage.

It's not quite the same as the Tahiti launch where "flagship" was much more clearly defined. nVidia really mucked things up with GK104.
 
With 100 reviews being published today, is it really too difficult to mention right from the beginning which reviews you decided to cherry pick? We aren't clairvoyant...

Sure, but how could you choose (cherry pick) techreport as a comparison when they haven't published a 1080 review? There is literally nothing to compare to.
 
With 100 reviews being published today, is it really too difficult to mention right from the beginning which reviews you decided to cherry pick? We aren't clairvoyant...

I meant going around comparing GTX1080 from a site and comparing it to their HD7970 review... you know something that is logical and makes sense since they are very similar situations.
Is everything OK with you? Is there a reason you are so salty?
 
Well I was more specifically talking about SkyMTL's take.

Had lots of questions about Tahiti and GCN... that one would think he would have similar questions and concerns about GP104 and Pascal, though to be fair he does admit that Pascal isn't perfect.
He slightly balked about the Tahiti MSRP but gave a "Damn good Value" award to GTX1080.
http://www.hardwarecanucks.com/foru...s/49646-amd-radeon-hd-7970-3gb-review-25.html
http://www.hardwarecanucks.com/foru.../72619-nvidia-geforce-gtx-1080-review-25.html

From memory, the biggest complain was about the sudden and quite substantial increase in price of the HD7970, compared to the older HD6970. It released at $550, while HD6970 had launched at $370. For 40% more performance, it was hard to swallow. 1080 is $50 higher, which is not commendable at all, but way less an increase than the HD7970 jump. Not that I was shocked with HD7970 price.. AMD needed badly to return to high end prices and was unfortunately caught short by GK104.

EDIT - I agree that Hardware Canucks conclusion is a bit too "excited" for a product that does not really bring many surprises... I cant see how Pascal can be more impressive than Fermi was... Fermi brought a completely new architecture with distributed geometry while Pascal looks just like a rehashed Maxwell...
 
Last edited:
Back
Top