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

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...

Fermi definitely brought a lot of changes but poor execution killed a lot of the excitement. People are just happy to finally see some progress with the move to 16nm. For me the price totally kills it because it probably means no big Pascal anytime soon. We need more competition.
 
Fermi was delayed (execution as you stated), but its performance to the amount of power consumption it had weren't that good, specially when you compared to what the competition had and at the price it was.
 
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...
It's just a reflection of the stagnation of the main APIs in terms of elements of the graphics pipeline. When there aren't any major new features, there's no reason to ditch an already efficient architecture for something new. Fermi added tessellation. Kepler didn't really add anything. Maxwell added some minor DX12 features. All very incremental. All mostly focused on improving efficiency. Pascal is no different.
 
Sure that? What says DX Caps Viewer?
„Full Precision“ - that's why I asked Nvidia in the first place and they as well said „nope“. Of course, it can use FP16, but not faster than FP32 and with tailored tests it maybe even slower due to possible twice the register usage (which would make sense for real 2× FP16).

Oh don't spoil it. I'm still collecting numbers. This is hilarious.
I'm not - already wrote it in my review, because... see above. :)
 
What I noticed, is that a couple of reviewers got confused by the term "Async Compute", using it both for the DX11 extension for explicit preemption by high priority context, and the asynchronous queues in DX12. And mixing these together badly, stating that Pascal would now fully support Async Compute in DX12 because it can do preemption now, or that Maxwell could perform the context switch (the reassignment of SMMs) in DX12 at draw call borders. o_O

I would say NVs marketing for fuzzing this term was a complete success.
 
What I noticed, is that a couple of reviewers got confused by the term "Async Compute", using it both for the DX11 extension for explicit preemption by high priority context, and the asynchronous queues in DX12. And mixing these together badly, stating that Pascal would now fully support Async Compute in DX12 because it can do preemption now, or that Maxwell could perform the context switch (the reassignment of SMMs) in DX12 at draw call borders. o_O

I would say NVs marketing for fuzzing this term was a complete success.
Yeah, is the same as Maxwell, the changes, if really any, must be in the compiler. See AoTS with async activated at 4K, less performance than with it desactivated:

http://www.computerbase.de/2016-05/...agramm-ashes-of-the-singularity-async-compute

Curiously that doesn´t happen at 1080p because there must be more SMs iddle that kick in the compute jobs.

Not async compute in sight.
 
Last edited:
Yeah, is the same as Maxwell, the changes, if really any, must be in the compiler. See AoTS with async activated at 4K, less performance than with it desactivated:

http://www.computerbase.de/2016-05/...agramm-ashes-of-the-singularity-async-compute

That's incorrect. Unless you're claiming nVidia is outright lying about dynamic load balancing in Pascal. AOTS performance is really not proof of anything as we have no idea if Oxide's implementation is Pascal friendly. Other reviews actually showed gains with async enabled.

The way I see it Pascal's dynamic load balancing is functionally equivalent to GCN's async shaders. At least I haven't seen anything to indicate otherwise.

Pre-emption is something else entirely, i.e. an explicit interrupt for high priority tasks.
 
Pre-emption is something else entirely, i.e. an explicit interrupt for high priority tasks.
Yes, preemption and async are different. Sebbi explained greatly. Nvidia didn´t talk about async, only preemption and it improved its granularity in Pascal, compiler level doable.

Async in AoTS are compute tasks. The feature is called async-compute but you can run it like an "async-compute" architecture (ala GCN) capable of running at the same time both compute and graphics tasks at CU level or fully using entirely a SM for compute or graphics like Nvidia does.
 
Last edited:
Funny note: AotS's settings.ini denotes „AC on“ as „AsyncComputeOff=0“. I wonder how much confusion this already created; I sure as hell had to rerun my first set of benchmarks with AotS a couple of months ago. :)
 
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.

Simply matching Pascal would be a big deal for AMD. It would bring them back to a situation they haven't known since Kepler was launched.
 
Highly unlikely they matched em, if you take their best case as 2.0 (stated at the financial call last Q's) over their current midrange, which is Tonga, their best perf/watt chip. 95 watts will give ya Tonga performance. They might have closed the gap a little bit, but I'm thinking the gap pretty much stayed the same.
 
Back
Top