AMD: Navi Speculation, Rumours and Discussion [2017-2018]

Status
Not open for further replies.
Has there ever been a PC game that actually fully used all 4096 SP, efficiently, through async compute via Parallelism?

Games like Forza 7 and Forza Horizon 4 play very well on GCN as well as other console ports (It seems UWP Games in particular). Wolfenstien 2 isn't the only 'PC' game using Async Compute.

IMO Navi's HW scheduler might be a bit more robust and their ACE could be re-organized with a lower SP count. Something that would take them beyond Async Shader quick response queuing

It might help to clarify what specifically you mean by HW scheduler.
There is an HWS unit, at least on Polaris and later GPUs. Are your data points covering older chips like Hawaii, which may mean this is not making as much of a difference?

The HWS and ACEs are on the front end of the GPU's command processing pipeline, and many games that see high performance on GCN often see big gains because their shaders and workload characteristics better fit the architecture of the back end where the workgroups launched by the front end go to be executed.

Losing resources in the back end, past a relatively modest baseline for asynchronous compute, would generally make things harder for the front end. They have limited influence on how well wavefronts execute past their early involvement with them.
If there is any expectation of improvement in the graphics pipeline, removing compute resources would make it doubly hard for the ACEs and HWS, as they exist to fill in gaps in graphics execution. The fewer spare resources, the fewer gaps.

Quick response queuing and other forms of resource reservation or pre-emption do worse if there are fewer resources to spare. They explicitly sacrifice throughput for responsiveness, either by walling off resources that could have gone somewhere else "just in case" or by forcing wavefronts that were running at very high utilization to yield to something that is not.
 
The Forza 7 situation was an NVIDIA driver bug that caused massive CPU limitations on it's GPUs. It was fixed via a driver update. Here are the results from the same link you posted:

https://www.computerbase.de/2017-09/forza-7-benchmark/2/#diagramm-forza-7-3840-2160
Hmm, that review was just a cursory update. Not a lot of effort either as it only showed 1080 on par with a 1080TI. And with 8xaa...for Forza 7.
Furthermore in the original test it made it very clear that Vega trailed Pascal at 8xaa: https://www.computerbase.de/2017-09/forza-7-benchmark/
In the link you've posted they simply carried over the 4K @8xMSAA result over to "update" review using 1080 w/387.92 driver and wallah you have your chart.
So what were the performance at 4xMSAA? We will never know from that update.

Its difficult for me to only look at such reviews when I noticed that AMD doesn't always lists performance improvements on every driver release.

Here is another review using Radeon 17.10.1:
https://us.hardware.info/reviews/76...-23-graphic-cards-test-results-wqhd-2560x1440
The above is dated in December 20, 2017. This gave AMD a opportunity to also release "a driver". Albeit my focus, in my prior post, is purely between AMD current and next gen.
So lets keep it that way.
 
Hmm, that review was just a cursory update. Not a lot of effort either as it only showed 1080 on par with a 1080TI. And with 8xaa...for Forza 7.
It's because ComputerBase didn't bother updating the 1080Ti scores with the new driver.
Furthermore in the original test it made it very clear that Vega trailed Pascal at 8xaa: https://www.computerbase.de/2017-09/forza-7-benchmark/
In the link you've posted they simply carried over the 4K @8xMSAA result over to "update" review using 1080 w/387.92 driver and wallah you have your chart.
So what were the performance at 4xMSAA? We will never know from that update.
All tests were performed with 8XAA from the start anyway. No one bothers downgrading to 4XAA when the game cranks out so many fps.

Here is another review using Radeon 17.10.1:
https://us.hardware.info/reviews/76...-23-graphic-cards-test-results-wqhd-2560x1440
The above is dated in December 20, 2017. This gave AMD a opportunity to also release "a driver". Albeit my focus, in my prior post, is purely between AMD current and next gen.
In that review, both AMD and NVIDIA GPUs appear to be lining up together at 1440p, the game appears CPU limited anyway. You can't extract much data from that situation any way. Only @4K you begin to see the separation of GPUs when the game becomes GPU limited.
https://us.hardware.info/reviews/76...graphic-cards-test-results-ultra-hd-3840x2160

Oh and BTW they tested with 8XMSAA too.

Here is another final test made this month with the 2080Ti:

RTX2080-REVIEW-47.jpg

https://www.hardwarecanucks.com/for...a-geforce-rtx-2080-ti-rtx-2080-review-10.html
 
Wolfenstien 2 is generally regarded as the best case scenario for Vega64 in the current times. Recent reviews put 2080Ti anywhere between 80~100% faster than Vega 64 in that game @4K. That's a lot of miles for Navi to cover.

Granted in other games the difference might be closer to ~70% to ~80% or more @4K (85% in TPU, 74% in CB). Still, very large ground to cover.


Not sure I understand your post.

The RTX2080ti is a cutdown TU102 @ $1,200. Since the Vega64 didn't compete with the 1080ti, but the 1080, it is only fair to compare the V64 to the RTX2080 (TU104) in terms of competing Radeon performance.


Turing GPU TU104:
12nm
545mm^2
13.6 Billion transistors
2,944 CUs
192 Texture units
64 ROPS
10.1billion FLOPS

Turing GPU TU102:
12nm
754mm^2
18.6 billion transistors
4,552 CUs
288 Texture units
96 ROPs
13.4 billion FLOPS


Vega 2.0:
7nm
20.1 billion FLOPS


You mean a lot of miles for Nvidia to cover, right..?

You are essentially suggesting, that AMD can not pack as many transistors into a 7nm die, as nVidia can in their 12nm die...? AMD can easily match Nvidia's performance and hit a $400~$500 price point using 7nm yields & efficiency. AMD doesn't have to produce anything as powerful as the 2080ti for gaming, only offer the same performance under 7nm prices.

Think of what a updated Vega128 on 7nm wafer, would offer in gaming performance..? (Specially without wasted raytracing space.)
 
Their 4K testing is made with no MSAA whatsoever, which is completely meaningless.
It is wiht MSAA in 2k 4x times and with 4k you dont Need MSAA.

Thats why i Always say Nvidia has really unbalanced Chips. They have so much geometry output and not much shader power.

And the Programs fulfill their wishes with useless MSAA. 8x MSAA is to much for 2k an 4k nobody Need it.

We should make a blindetest ;)

https://www.tomshardware.com/reviews/anti-aliasing-nvidia-geforce-amd-radeon,2868-8.html
 
Last edited:
All tests were performed with 8XAA from the start anyway. No one bothers downgrading to 4XAA when the game cranks out so many fps.

In that review, both AMD and NVIDIA GPUs appear to be lining up together at 1440p, the game appears CPU limited anyway.
It's a bit erroneous on your part to imply that the test results are cpu limited because it's not run at 4K 8xMSAA. Furthermore, it's specious to also imply that Vega is a "4K" 8xMSAA card. Lets try to keep this in prospective. 1440p, for example, is ample enough of a resolution to make comparisons with when Navi finally reaches the market. 4K, 8xMASS serves no real value if we are going to compare Vega to Navi except to say that (if true) Navi is a true 4K card. Your results seem to just make comparisons between Vega and Pascal. I'm pretty sure there is another thread for that.

Lets keep the focus on Navi. Do understand that my posts were comparing Vega to Navi in what I would think Navi would perform if GCN is still used by then.
 
Last edited:
Not really sure what are you arguing here! MSAA cleans up the image even @4K. It's even used on the Xbox X version of the game. Because IT IS NEEDED. The game has no other AA option.
Be glad you never played DOOM!
Doom has TSSAAA and multiple post process AA methods, Forza 7 has none. Hence why MSAA is needed.
1440p, for example, is ample enough of a resolution to make comparisons with when Navi finally reaches the market. 4K, 8xMASS serves no real value if we are going to compare Vega to Navi except to say that (if true) Navi is a true 4K card
You are basing your argument on one CPU limited game. Where @1080p and 1440p the difference between different cards is minimal. You need to push the game to become GPU limited to show the true difference between cards. Without this your argument is flawed. This is not about 2.5K or 4K or even 95K, it's about testing GPUs with GPU limited scenarios.
 
@DavidGraham

I don't think that they use massive MSAA on Xbox X for native 4k also vor 2k i don't think they use 8x MSAA Maybe 2x or maximum 4x. More is a waste of Energy.

If you realy have native 4k you don' t need antialysing.

In my linked Article TomsHardware stated that you can't see any big different between 4x and 8x MSAA and the Resolution was 1080p. So i think you will have better Results with 2k an 4x MSAA and of Course with native 4k.
 
Vega 2.0:
7nm
20.1 billion FLOPS
According to who?
The RTX2080ti is a cutdown TU102 @ $1,200. Since the Vega64 didn't compete with the 1080ti, but the 1080, it is only fair to compare the V64 to the RTX2080 (TU104) in terms of competing Radeon performance.
It's perfectly fine to compare top of the line GPU chips to ascertain the performance difference. AMD needs to make 7nm's Navi 80% faster than Vega 64 to even have a chance of competing with NVIDIA's top of the line 12nm! Whether you believe this will happen or not is up to you. But we are yet to talk about NVIDIA's 7nm.
Think of what a updated Vega128 on 7nm wafer, would offer in gaming performance..? (Specially without wasted raytracing space.)
Again we have yet to talk about NVIDIA's 7nm. Factor that in, and you have your answer.
 
According to who?

It's perfectly fine to compare top of the line GPU chips to ascertain the performance difference. AMD needs to make 7nm's Navi 80% faster than Vega 64 to even have a chance of competing with NVIDIA's top of the line 12nm! Whether you believe this will happen or not is up to you. But we are yet to talk about NVIDIA's 7nm.

Again we have yet to talk about NVIDIA's 7nm. Factor that in, and you have your answer.


I am confused, are you claiming that AMD doesn't have a 7nm GPU coming..?
Or are you saying you do not understand what 7nm yields, over 14nm & 12nm fabrications. In terms of performance and power usage. Or how scaling from 12nm to 7nm works in terms of power & efficiency.

You can quote me and say "according to who..?" all day long, but you are doing is feigning ignorance..? Or, you are posting as if you have not read up on anything Vega20 related in the past few months...? (This is a news article from 3 months ago.)


Essentially, a 7nm Radeon Vega 2.0 chip, (about half the size of Turing's 754mm2 TU102), would be greater in graphical power as Turing's own TU104 @ 545mm2, while still using less energy. And since these new Vega20s are so small (360mm2) they are also higher yield and easier to produce, than Nvidia's much larger GPU dies. Thus, AMD's newer 7nm Radeons coming will be much cheaper to produce.

Or, if Vega's 7nm scaling is too confusing for you to calculate/understand... just think of Vega2.0, as being a new Radeon card with about twice the performance of Vega64, for the same cost and using 30% less power.
 
Or, if Vega's 7nm scaling is too confusing for you to calculate/understand... just think of Vega2.0, as being a new Radeon card with about twice the performance of Vega64, for the same cost and using 30% less power.
Yeah, in the land of dreams maybe. The transition of AMD from 28nm to 14nm only yielded 30% more performance and a ton of power consumption increases. You expect 14nm to 7nm transition to give you a 100% uplift over Vega with less power just because it's running on 7nm? What is this 7nm magic matetial? The dreams achiever?

Ironically this 7nm only appears to work in AMD's favor too! When NVIDIA does it they will not be achieving anywhere near the results of Vega 7nm, because 7nm is the dream land of Vega only.
 
Essentially, a 7nm Radeon Vega 2.0 chip, (about half the size of Turing's 754mm2 TU102), would be greater in graphical power as Turing's own TU104 @ 545mm2, while still using less energy. And since these new Vega20s are so small (360mm2) they are also higher yield and easier to produce, than Nvidia's much larger GPU dies. Thus, AMD's newer 7nm Radeons coming will be much cheaper to produce.

Where are you getting your numbers from, 360mm2? TSMC is stating a 1.6x scaling factor for 7nm from 16nm. So VEGA 10 at 510mm2 scales down to 318mm2 on 7nm. 360mm2 would only be a 10-15% increase in die size & transistor count.

Note that 7nm isn't a true measurement. It's just a name for the node and more of a marketing term.
http://www.tsmc.com/english/dedicatedFoundry/technology/7nm.htm
 
I am confused, are you claiming that AMD doesn't have a 7nm GPU coming..?
Or are you saying you do not understand what 7nm yields, over 14nm & 12nm fabrications. In terms of performance and power usage. Or how scaling from 12nm to 7nm works in terms of power & efficiency.

You can quote me and say "according to who..?" all day long, but you are doing is feigning ignorance..? Or, you are posting as if you have not read up on anything Vega20 related in the past few months...? (This is a news article from 3 months ago.)


Essentially, a 7nm Radeon Vega 2.0 chip, (about half the size of Turing's 754mm2 TU102), would be greater in graphical power as Turing's own TU104 @ 545mm2, while still using less energy. And since these new Vega20s are so small (360mm2) they are also higher yield and easier to produce, than Nvidia's much larger GPU dies. Thus, AMD's newer 7nm Radeons coming will be much cheaper to produce.

Or, if Vega's 7nm scaling is too confusing for you to calculate/understand... just think of Vega2.0, as being a new Radeon card with about twice the performance of Vega64, for the same cost and using 30% less power.

You do realize that the entire "article" is 3rd hand layman's conjecture based on the generous reading of GlobalFoundaries fevered dreams presentation for their stillborn 7nm process, right? The last thing that needed was your ill-informed twist on top.
 
Status
Not open for further replies.
Back
Top