Ampere is using all of these FP32 flops. Thats the reason why a RTX3080 is 60%+ faster than a 2080S and 30%+ faster than a 2080TI. Crunching through compute workload to offset other limitations.
Speculation: GPU Performance Comparisons of 2020 *Spawn*
- Thread starter eastmen
- Start date
- Status
Ampere is using all of these FP32 flops. Thats the reason why a RTX3080 is 60%+ faster than a 2080S and 30%+ faster than a 2080TI. Crunching through compute workload to offset other limitations.
I think you mean, developed to improve their competitive on CUDA compute workloads at the expense of non-CUDA compute workloads.
It still helps, but just in FP32 workloads that are shader bound, you're going to see great speed ups there.
But yea, R7 is sick in non-gaming workloads. Just library support needs to be there.
hmm regardless though, wrt to the topic of not having a supporting front end to the compute, it looks like a hard chasm to cross wrt the traditional front end without a ridiculous amount of bandwidth to support it and thus a steep increase in costs.
D
Deleted member 13524
Guest
There's absolutely no proof the RTX3080 ever reaches 30 TFLOPs when rendering a game, and that's most probably a false statement.
If you're talking about Turing then you might be right.
It is. RTX3070 will be ~30 faster than a 2080FE with a 26% smaller die and the same configuration and equal clock rates.
DegustatoR
Legend
Since this seem to be such a hard thing to grasp for some:
2080 has 10 TFLOPS + 10 TeraInt32OPS - that's 20 teraOPS in total.
3080 has 30 TFLOPS.
If we have a game which is 50/50 math between FP32 and INT32 then what is the maximum theoretical performance gain in it between 2080 and 3080? And does it mean that Ampere "reaches 30 TFLOPs when rendering" such game?
And in certain games like Doom Eternal and Crysis Remastered, those advanced engines really do take advantage of Ampere's FP32 TFLOPs.
Less carefully optimized / architected games will likely run up against other limits first.
Example from Crysis Remastered with their software RT solution running entirely on the shaders:
https://www.purepc.pl/test-wydajnosci-crysis-remastered-czy-mi-pojdzie-zalezy-co-masz?page=0,14
The 3080 manages to be +50% over the 2080Ti and more than +100% over the 2080 Super.
hmm I might be missing something here. I thought you were trying to parallel Vega/RVII and Ampere together. Vega/R7 then later became CDNA and AMD introduced RDNA to compete with game workloads.
Crysis Remastered raytracing uses RT-cores on NVIDIA
Benetanegia
Regular
Yeah, that was far from my point, keep trying. My point was actually that contrary to claims being made that just like Ampere, Vega also had "too much FP32", so it could just be "fixed" by increasing FP32 usage. that is sadly not true at all. All metrics in Vega increased by the same amount of 2x. Trying to hilariously argue about 5% differences in clocks doesn't change that fact. Ampere has a disproportionate amount of FP32 and FP32 only and an increase in that usage will undoubtely increase game performance.
DegustatoR
Legend
All cards mentioned have RT cores.
Benetanegia
Regular
Not only that, 3080 has the same amount of RT cores as the 2080 Ti and similar actual boost clocks, and is certainly not gaining +50% performance from 23% higher mem BW, or 10% higher texture or pixel fillrate. There must be something...
ModEdit: language
This 30TF GPU wont get any worse by trying to create theories that NV is lying or making things up etc. In the end though, those 30TF's (or 36 in 3090s case) are as much worth as 30TF's of RDNA2. They pretty much align there as someone noted in this thread. Turing isnt the same thing though.
Ampere is one of the biggest leaps so far, reaching 30TF of compute power for the 3080. We are not even talking about ray tracing here, or even DLSS. Its also introducing GDDR6X speeds close to 800GB/s and more. They also lowered prices. AMD will counter with close enough products (their crusing Zen2 with their Zen3 cpus already) with their Navi2 lineup, closing in on Amperes TF metrics (25+). Everyone happy, or, 99% of pc gamers atleast.
Last edited by a moderator:
Yes, point was that it's not necessarily the FP32 power that's driving the results, they have not only more but faster RT-cores too.
Huh? What theories? Who?
No-one has suggested in any imaginable way any that NVIDIA would be lying.
DegustatoR
Legend
So you're saying that s/w based RT in Crysis Remastered is limited by BVH traversal performance? I.e. the h/w part which this solution isn't even targeting since it's a s/w one in its basis?
If that would be the case shouldn't we have seen something along the lines of 2080 being 4x-6x faster than 1080Ti there?
I have no clue what other things affect performance and how much, but according to Crytek Crysis Remastered uses hardware acceleration on RTX cards
Has that been implemented yet?
According to this interview yes
https://wccftech.com/crysis-remaste...mance-boost-dlss-coming-later-with-an-update/
DegustatoR
Legend
Great. All cards mentioned are RTX cards. If this game would be limited by RT h/w - which would be really weird considering that it targets GPUs without one - we would be seing something akin to what we're seeing in pure path tracing workloads between GPUs with and without RT h/w.
Look at the difference in performance between RTX2080 Super and Vega 64. Does it look like RTX2080 is using RT cores when it is not that much faster than Vega 64? That tells you that at least RTX 2**** cards are not using RT cores in that benchmark. If all settings are the same then RTX3*** aren't either.
- Status