How so?
It's still running on "Fiji driver" in debug mode"Don't believe the framerates you saw in that demo, it's running using Fiji drivers!"
"The drivers are in debug mode, so that's cutting performance in half!"
"Don't believe the framerates you saw in that demo, it's running using Fiji drivers!"
"The drivers are in debug mode, so that's cutting performance in half!"
And plenty of other nonsensical claims that came from people praying, wishing, and hoping that the cold hard truth wasn't so.
Vega is what it is -- a power hungry GTX 1080 from AMD.
First of all, you have selectively quoted Sebbi in such a fashion as to mislead readers of this thread as to the overall thrust of his comments, which concluded that Vega would very much benefit from improved polygon throughput, and secondly you omitted to inform anyone here that Sebbi made those comments prior to being informed that primitive shaders were not yet enabled at all in Vega's driver.From Sebbi
You can't calculate triangle/primitive throughput like this anymore. Modern games don't spend their whole frame time rasterizing geometry. Lighting, post processing, etc take significant chunk of GPU time (up to 50%), and during that time the geometry pipelines are idling. Only a small part of the frame is geometry bound. Shadow map rendering is the most geometry bound step. G-buffer rendering tends to also be partially geometry bound (no matter how fat pixels), since there tends to be lots of triangles submitted that result in zero pixel shader invocations (backfacing or earlyZ/hiZ rejected). For example drawing a high poly character behind a nearby corner would cause 100k vertex shader invocations, but zero pixel shader invocations. This draw call would cause a bubble in GPU utilization, since the geometry pipeline can't process these triangles fast enough to go through these triangles before the existing pixel shader work (from previous draw calls) finish executing (on the CUs). I have found out that on GCN2, vertex shader work can only utilize roughly two CUs in common case (geom pipes simply can't feed more vertex waves). Remaining (=most) CUs will idle if there's a big chunk of sequential triangles which generate no pixel shader invocations. This is one of the reasons why you want to cull the triangles early, to avoid underutilizing the GPU.
Slide 12 of this presentation is a good example:
https://frostbite-wp-prd.s3.amazonaws.com/wp-content/uploads/2016/03/29204330/GDC_2016_Compute.pdf
Here you can see that the GPU occupancy is very low in a part of the G-buffer rendering step where most of the geometry is occluded. The green part of the occupancy graph is the vertex shader work. As you can see, VS occupancy never goes above a certain (small) portion of the whole GPU. GCN simply needs lots of pixel shader work to saturate the GPU when it is rendering geometry that results in small amount of visible pixels. This is also the main reason why async compute helps GCN so much. CUs can simply execute background compute shader work when there's not enough pixel waves spawned.
Async compute also has another advantage. It allows the developer to keep geometry units active for larger portions of the frame, because you can freely overlap compute with graphics. You don't need to dedicate a big chunk of GPU time to non-rasterization work (post processing and lighting). You can overlap this work with geometry heavy work to reduce both the time when geometry units are idling and the time when CUs are idling. This is one advantage that AMD has over the competition. Games could spend the whole frame submitting both geometry work and compute work. This way geometry units can be utilized during the whole frame (instead of <50% of the frame). The downside is unfortunately that AMD has been behind Nvidia in geometry performance, so you need to use techniques like this to reach parity, instead of gaining big advantages. Polaris improved things a bit, and Vega should improve things further, but so far the results haven't been as good as I had hoped. I would guess that the primitive shaders and DSBR still need some additional driver work to show their full potential. I just hope (for AMDs sake) that they don't need to write custom profiles for each game to utilize these... AMD simply doesn't have as much resources as Nvidia to optimize individual titles separately.
Note again that Sebbi was unaware that primitive shaders had not been enabled yet at all when he wrote this comment. If you want to see how excited Sebbi is by the possibilities of Vega's NGG fast path you can read his comments in this thread after Ryan Smith (you all may have heard of him) finally confirmed on the record that primitive shaders were not enabled in Vega's public drivers here.The downside is unfortunately that AMD has been behind Nvidia in geometry performance, so you need to use techniques like this to reach parity, instead of gaining big advantages. Polaris improved things a bit, and Vega should improve things further, but so far the results haven't been as good as I had hoped. I would guess that the primitive shaders and DSBR still need some additional driver work to show their full potential.
Actually, after reading that it sounds like improved geometry throughput will help. Sounds like GCN has a lot of troubles.
First of all, you have selectively quoted Sebbi in such a fashion as to mislead readers of this thread as to the overall thrust of his comments
If RX Vega turns out to not be any faster at all than Vega FE, then pretty much the entirety of RTG needs to get fired. The scale of failure implied would be so enormous that I'm going to hold off judgement until I see actual independent benchmarks. This would also mean that the multiple reports of an improvement in hashrate are also totally false and that RX Vega will end up no better at mining than FE is.
You fell for AMD marketing once, don't make same the mistake a second time.
In your statement you said "offerings" meaning 1070, 1080, 1080ti so yes part of your statement is false which I specifically stated. Vega 56 has more performance then the 1070 which you left out of your statement above but in your original statement you said that the 1070 out performed the Vega 56.How is it false? The 56 is not higher performaing than the 1080,1080
Ti. And the 56 is not cheaper or lower power usage vs the 1070 its competing directly with.
Actually, after reading that it sounds like improved geometry throughput will help. Sounds like GCN has a lot of troubles.
The gist seems to be that unseen geometry is a huge waste because it takes up CU time to render nothing (since it's never shaded) and forces other CUs to idle because the work doesn't fill up the wavefronts. He also mentions that the geometry engine can't finish processing the geometry due to unseen geometry before the pixel shader work is done, which implies the CUs end up waiting for more work to be issued.
Increased geometry throughput should greatly increase gcn's ability to cull the unseen geometry and improve the feeding of shaders. And obviously improved work distributor will help fill wavefronts.
At least that's how I understood it. Not entirely sure though
Maybe, just maybe, he's not locked into an ideology.I did no such thing. I used his post to refute your claim that Vega is geometry bound.
Now you're trying to move the goalposts.
Whatever happened to when you posted this?
You fell for AMD marketing once, don't make same the mistake a second time.
The full text of his post reaches the exact opposite of what you assert, and you also concealed that he wasn't even aware that the feature wasn't enabled yet when he made that comment.I did no such thing. I used his post to refute your claim that Vega is geometry bound.
Interestingly enough, when RX Vega launched without substantively improved performance I started investigating why that was case.You fell for AMD marketing once, don't make same the mistake a second time.
Whitepapers are not marketing. The entirety of your claim boils down to simply asserting that RTG is lying outright in their white paper and that primitive shaders either don't exist, or can't deliver even half of the claimed benefits. Do you have any evidence that RTG is lying in its Vega whitepaper or that it is lied previously in technical whitepapers?This i think is the heart of the situation/discussion.
On one side you have the people who are drinking the AMD kool-aid and buying into AMD's marketing, and expecting huge performance boosts from these missing at launch features.
Then on the other side you have the realists, who are basing their expectations on reality and provable performance in reviews/benchmarks as well as past AMD launches and how they went and how hardware has progressed over time in the past.
Unfortunately its hard to convince people who buy into fairy tales that the Easter bunny isnt real.
In your statement you said "offerings" meaning 1070, 1080, 1080ti so yes part of your statement is false which I specifically stated. Vega 56 has more performance then the 1070 which you left out of your statement above but in your original statement you said that the 1070 out performed the Vega 56.
Right, so you agree with me that Vega was a failure and cannot compete with current Nvidia offerings either in perf/watt or overall performance. And at current street prices not in perf/$ either.
In the end it was directly my trying to figure out what was going on with RX Vega's performance that led to Ryan Smith and several others to finally confirm on the record that primitive shaders were not enabled RX Vega's public drivers.
Whatever happened to when you posted this?
Whitepapers are not marketing. The entirety of your claim boils down to simply asserting that RTG is lying outright in their white paper and that primitive shaders either don't exist, or can't deliver even half of the claimed benefits. Do you have any evidence that RTG is lying in its Vega whitepaper or that it is lied previously in technical whitepapers?
Feel free to go read the thread on the Beyond 3D Forum for yourself here, where I started the argument about whether primitive shaders were or were not enabled in Vega's public drivers on page 57 that continued until Ryan Smith confirmed that they were not enabled in Vega's public drivers on Page 59:Oh, it was you that drove those determinations. Cool. Who else besides Ryan were you in contact with?
Ryan Smith said:Quick note on primitive shaders from my end: I had a chat with AMD PR a bit ago to clear up the earlier confusion. Primitive shaders are definitely, absolutely, 100% not enabled in any current public drivers.
The manual developer API is not ready, and the automatic feature to have the driver invoke them on its own is not enabled.
Well some of us gained some extra insight the last few days.
Right, so you agree with me that Vega was a failure and cannot compete with current Nvidia offerings either in perf/watt or overall performance. And at current street prices not in perf/$ either.
Once again, do you have any evidence whatsoever to support your implicit claim that RTG is lying in the Vega whitepaper about primitive shaders existing or being capable of delivering even half of their claimed culled polygon throughput improvements? Do you have any evidence that RTG has previously outright lied in its whitepapers? Described features which don't exist?Whitepapers, until they are backed up by benchmarks and real world testing(by independent third parties), are exactly marketing material.
Vega can't compete with Nvidia offerings and you specifically state overall performance separate from perf/watt.
Nvidia offerings did not include the 1070 in your statement above?
Vega did not include the 56?
I'm just going off of what you actually said not what you're trying to talk around.
Right, so you agree with me that Vega was a failure and cannot compete with current Nvidia offerings either in perf/watt or overall performance. And at current street prices not in perf/$ either.
Once again, do you have any evidence whatsoever to support your implicit claim that RTG is lying in the Vega whitepaper about primitive shaders existing or being capable of delivering even half of their claimed culled polygon throughput improvements? Do you have any evidence that RTG has previously outright lied in its whitepapers? Described features which don't exist?
This same pricing argument was stated before the Polaris and Ryzen launch. A few laughed at the possibility of a $200 Polaris and a $500 Ryzen.The new AMD management appears to be approaching pricing differently than the past, at least until marketshare improves. For certain, they have set up R&D and production to be very lean, and I'm certain that quite a few will also say that the marketing dept might be too lean.
We'll see.
The reason you stay here is because you also have a feeling something is missing and its out if the ordinary. So just let it go.If there is any benchmarks to backup this extra insight please post them, i would love to be proved wrong on my expectations for vega and avoid paying the Gsync tax when i upgrade monitors.
I posted what i actually said again, i suggest you read it. Nvidia offerings and Vega offerings include all models from both camps. i even broke it down for you in my last reply, im not sure how much more clearly i can say it.
edit to add last post about this to this one for clarification.
I claim that Vega cannot compete with Nvidias current offerings in overall performance, this is a true statement, they have nothing to compete with the 1080Ti, or Titan.
I claim that they cannot compete in perf/watt, this is also a true statement.
I claim that at current street prices they do not compete in perf/$ either. Also, at least in my country, a true statement.
When we look at the missing features and the performance profile - and especially inefficiency - of vega it adds up.Im not claiming they are lying, yet anyways, time will tell on that it could go either way. Ive already stated many times that i have no doubt features are disabled currently.
Im claiming that until we have proof that these features work as intended and much more importantly than just working and being enabled that they perform as intended in real games and actually offer a substantial boost in performance in the real world, that we should wait and see what happens, and not buy into the AMD marketing teams claims, until such time as they can back up those claims with real world performance numbers. Basically put me in the i need to see proof to believe it category, I dont blindly trust any companies marketing departments for any products.