Battlefield V with DXR on tested from Techspot(Hardware Unboxed)

[RejZoR]

5 minutes ago, leadeater said:

Bet I can make more quotes.

 

  Reveal hidden contents

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Ok that got boring quickly, I admit defeat where I stopped.

Reported to Linus for quote spamming

[S w a t s o n]

4 minutes ago, leadeater said:

I haven't watch the video and it's rather long so without a time pointer I'll have to comment as is. I suspect that's in reference to not using Tensor cores for Ray Tracing but the actual denoising would be using the Tensor cores. At least all the past information and demos I've seen before all state using Tensot cores for that. You got a time stamp I can jump to?

They do say they are using the tensor cores but not for denoising, i dont have the time stamp but I can corrborate. DICE made some temporal denoiser and it is done on the cuda cores at least in the demo.

[leadeater]

10 minutes ago, S w a t s o n said:

Just need 3dxpoint to get cheap right?

Well if memory prices weren't insane 32GB of HBM should do it? Seeing the Witch's Cry tech demo redone with RT lighting and reflections would be interesting to see, I somehow doubt it's possible even with 4 2080 Ti/Titan V.

[leadeater]

10 minutes ago, S w a t s o n said:

They do say they are using the tensor cores but not for denoising, i dont have the time stamp but I can corrborate. DICE made some temporal denoiser and it is done on the cuda cores at least in the demo.

Awesome thanks, oh well bad assumption of mine since all facets of the 2080 Ti is 17.6% less than the Titan V so it was kinda dumb to assume it was the Tensor cores. Forgot everything scales with the SM count heh.

[S w a t s o n]

17 minutes ago, leadeater said:

Well if memory prices weren't insane 32GB of HBM should do it? Seeing the Witch's Cry tech demo redone with RT lighting and reflections would be interesting to see, I somehow doubt it's possible even with 4 2080 Ti/Titan V.

My body is ready.

15 minutes ago, leadeater said:

Awesome thanks, oh well bad assumption of mine since all facets of the 2080 Ti is 17.6% less than the Titan V so it was kinda dumb to assume it was the Tensor cores. Forgot everything scales with the SM count heh.

Well the thing is the tensor cores are doing the ray tracing in the Demo, they dont got time for that denoising shit. If they use RT cores for raytracing the obvious answer is to denoise on tensor cores

 

TL;DR RTX On Battlefield V not using RT Cores (yet), I'm pretty sure

Edited November 17, 2018 by S w a t s o n
could do it later

[leadeater]

5 minutes ago, S w a t s o n said:

Well the thing is the tensor cores are doing the ray tracing in the Demo, they dont got time for that denoising shit. If they use RT cores for raytracing the obvious answer is to denoise on tensor cores

I'm amazed they are able to do the denoising on the CUDA cores, they must not be doing full screen image processing and limiting it to the post effects/ray tracing portions some how. It did seem odd that the FPS would drop so much if the task was being done on dedicated and not utilized before hardware of the GPU.

 

Quote

In this experiment we evaluated our method on the standard set of images used by previous works. Evaluation was performed for peak values in the range of 1 to 30. PSNR values and running time presented in Table 3 show that our method outperforms all other methods by a significant margin for the large majority of the images and almost all peak values. In addition, the execution time is orders of magnitude faster on a Titan-X GPU taking only 37 milliseconds for a 256 × 256 image, and is comparable to other methods when it runs on an Intel E5-2630 2.20GHz CPU, taking 1.3 seconds. A qualitative example can be seen in Figure 1 showing the man image denoised by DenoiseNet and by I+VST+BM3D[4] for a peak value of 4.

https://arxiv.org/pdf/1701.01687.pdf

Denoising on small images really is not that fast on a GPU, Nvidia claims compared to Pascal the Tensor cores are 3 times faster. The above test I think is on the older/slower Titan X which I think is a bit slower than the P6000.

 

[S w a t s o n]

4 minutes ago, leadeater said:

 

It's built off their temporal AA engine or something like that. They are literally just AAing the shit out of the raytraced image. I think...

[leadeater]

4 minutes ago, S w a t s o n said:

It's built off their temporal AA engine or something like that. They are literally just AAing the shit out of the raytraced image. I think...

Magic happens and it looks good . I wonder if at a much later time a new patch/DLC or something might come out that completely revamps the entire RTX/DXR implementation, or maybe they will just save it for their next game.

[S w a t s o n]

Just now, leadeater said:

Magic happens and it looks good . I wonder if at a much later time a new patch/DLC or something might come out that completely revamps the entire RTX/DXR implementation, or maybe they will just save it for their next game.

Well in the video they very much stated that they would use the RT Cores but who knows if DICE/EA are gonna spend the money to have them do it for BFV

[S w a t s o n]

This is time-stamped to the mention about denoising

[leadeater]

Just now, S w a t s o n said:

This is time-stamped to the mention about denoising

Thanks, I did manage to find it earlier though.

[Princess Luna]

Why can't nVidia give the alternative to install the proper game drivers on the TITAN V? AMD *tried* it with the Vega FE.

 

I know that in no world would it make sense pay the double for about 17%~20% performance boost but why not... Also people complaining so hard that the RTX 2080 Ti is too expensive for the performance bump from the the GTX 1080 Ti.

 

The RTX 2080 Ti is actually the SAME price as the TITAN Xp that was a card that made no sense since it performed pretty much identically to the 1080 Ti needing expensive watercooling to be on the benchmark top charts.

 

The way I see it now we're ACTUALLY gaining something for that 1200$ price tag the old TITAN Xp had, we get all these Tensor Cores and RT Cores and GDDR6 to mess around with while Actual bump in CUDA Cores and Rasterization performance.

 

So really... having an useless card replaced for an actual 'useful' better card for the same price was a good deal to me.

[CarlBar]

1 hour ago, RejZoR said:

@CarlBar

Dude, he quoted like 15 of my lines individually. If I quote that and respond to each, it'll be fucking 30 of them. Sorry, just no. Not bothering to respond to a huge ass bag of noodles is not admission of anything. I've responded to all his quotes in one single paragraph.

 

Except you didn't respond at al, and you just conveniently refused to respond to me when i summarized the really important points. Admit it, you're wrong and you don't have an argument at all to respond with. Also just because he quoted you 15 times dosen;t mean you have to quote and respond to each in turn, i prefer bullet point responses myself. for responding to a signle post.

 

1 hour ago, leadeater said:

I haven't watch the video and it's rather long so without a time pointer I'll have to comment as is. I suspect that's in reference to not using Tensor cores for Ray Tracing but the actual denoising would be using the Tensor cores. At least all the past information and demos I've seen before all state using Tensor cores for that. You got a time stamp I can jump to?

 

I know you dug it up, and yeah that is what has me scratching my head and wondering if it's the problem.

 

Also regarding Titan V vs 2080 Ti. i'm just going off the Wikipedia info. That quotes them at:

 

Titan V:

 

Single Precision: 13.8 GFLOPS

Double Precision: 6.9 GFLOPS

Half-Precision: 27.6GFLOPS

 

2080Ti:

 

Single Precision: 13.448 GFLOPS

Double Precision: 0.4202 GFLOPS

Half-Precision: 26.896GFLOPS

 

That puts the 2080Ti within 97% and some decimals of the Titan V in SP and HP rate. DP is obviously utter trash, but thats the edge the Titan V has.

 

EDIT: As a further addition to the above, NVIDIA@s own demo listed the 2080Ti at 110 TFLOPS Tensor performance whilst a Tom's Hardware article lists the Titan V Tensor performance at the same level.

[RejZoR]

@CarlBar

I've literally replied on page 3 in a single paragraph. I don't know why you lot are so persistent for me to "admit" something when I know I'm right. Who gives a crap if RTX cards can do ray tracing. It's useless gimmick as no game supports it, next to none are planned and by the time it'll actually be useful, it'll be too slow anyway. So, essentially you're buying an overclocked, overpriced GTX 1080Ti with feature set you can't really use yet and won't be able for quite a while. Bloody hell, even first generation of Pixel Shader powered graphic cards (GeForce 3 and Radeon 8500) had more featured games and even though most of them only did water with it, it was kinda worth it because they weren't asking an absurd premium just because of it (like they do for RTX cards). But sure, I'm "wrong".

 

Also, drop the attitude I'm saying ray tracing is stupid. I never said that. I have and will repeat that it's stupid in current form as it just isn't mature enough on any level. Pixel Shaders were adopted quickly because all cards supported it, except most bottom feeder ones. And even those eventually got them in form of Radeon 9250 cards and stuff and people at least tried it even if it ran like crap. Having feature only on graphic cards for 1200€ means adoption of their tech will take forever.

[CarlBar]

 

Again refusing to respond to my post. Again you have no argument.

[Princess Luna]

1 hour ago, RejZoR said:

.

Like I said however, before for the 1200$ price point that some of us is fine paying all we had was the truly gimmick TITAN Xp that was no different than the 1080 Ti, in fact most of the time it was worse due to the shitty former Founders Edition blower cooler.

 

Now for those 1200$ we actually get a real bump in performance retaining single GPU solution, so the RTX 2080 Ti is not an overpriced GTX 1080 Ti, it is an actual good value for money TITAN Xp that makes sense when the former Xp didn't.

[Taf the Ghost]

3 hours ago, leadeater said:

I haven't watch the video and it's rather long so without a time pointer I'll have to comment as is. I suspect that's in reference to not using Tensor cores for Ray Tracing but the actual denoising would be using the Tensor cores. At least all the past information and demos I've seen before all state using Tensor cores for that. You got a time stamp I can jump to?

 

3 hours ago, S w a t s o n said:

They do say they are using the tensor cores but not for denoising, i dont have the time stamp but I can corrborate. DICE made some temporal denoiser and it is done on the cuda cores at least in the demo.

From the early benchmarks & images, it actually looks like the Denoising is bottlenecking BFV pretty hard. The foreground reflections on the guns is what makes me think that. They're pretty noisy for how good the rest of the reflections are. Points in that direction for issues.

[Taf the Ghost]

2 hours ago, S w a t s o n said:

This is time-stamped to the mention about denoising

So, just as a point, RTX actually doesn't even work on BFV. This explains why some have been able to get it working on RX Vega cards with some tweaks to some ini files, and why it seems to run pretty well.

 

I'm guessing Metro 2033 is the first time we'll be able to have RTX in our lives?

[S w a t s o n]

1 hour ago, Taf the Ghost said:

This explains why some have been able to get it working on RX Vega cards with some tweaks to some ini files, and why it seems to run pretty well.

Wait someone got BFV's RTX to work on Vega? That can't be right, it relies on the tensor cores for the ray calculations

[Taf the Ghost]

30 minutes ago, S w a t s o n said:

Wait someone got BFV's RTX to work on Vega? That can't be right, it relies on the tensor cores for the ray calculations

There's argument over whether it's actually working/doing, but a few seem to have been able to force it on RX Vega. 

[leadeater]

7 hours ago, CarlBar said:

I know you dug it up, and yeah that is what has me scratching my head and wondering if it's the problem.

 

Also regarding Titan V vs 2080 Ti. i'm just going off the Wikipedia info. That quotes them at:

 

Titan V:

 

Single Precision: 13.8 GFLOPS

Double Precision: 6.9 GFLOPS

Half-Precision: 27.6GFLOPS

 

2080Ti:

 

Single Precision: 13.448 GFLOPS

Double Precision: 0.4202 GFLOPS

Half-Precision: 26.896GFLOPS

 

That puts the 2080Ti within 97% and some decimals of the Titan V in SP and HP rate. DP is obviously utter trash, but thats the edge the Titan V has.

 

EDIT: As a further addition to the above, NVIDIA@s own demo listed the 2080Ti at 110 TFLOPS Tensor performance whilst a Tom's Hardware article lists the Titan V Tensor performance at the same level.

You're using the boost performance of the 2080 Ti against the base of the Titan V, wikipedia doesn't list the boost TFLOPs of the Titan V. Both GPUs can boost using GPU Boost 2.0, if the Titan V is ahead at base it's ahead at boost.

 

The 2080 Ti is slower, much slower. The Titan V is much bigger die, 17.6% SMs so 17.6% more of everything (no RT cores). It's literally the difference between a Ti card and a non-Ti card. You know the 2080 clocks higher than the 2080 Ti right? It's still slower for the exact same reason. Turing and Volta are on the same node with very similar architectures, if you scale a Turing die to 80 SMs to match a Volta GV100 die they'll be performing very similar.

 

Plus I don't even know why you care that the Titan V is a faster card, or the 2080 Ti. Only one of them has game ready drivers, the other is way more expensive purpose built for compute server workloads and comes at a very high price premium. The Titan V is just a very slightly cut down Tesla V100. Anyone that developed on the Titan V was doing so with more compute power available to them.

 

Edit:

The Titan V would be 16.732 TFLOPs at the basic boost listed on the spec sheet, both the Titan V and 2080 Ti can boost higher than those specs though.

[leadeater]

5 hours ago, Princess Cadence said:

Like I said however, before for the 1200$ price point that some of us is fine paying all we had was the truly gimmick TITAN Xp that was no different than the 1080 Ti, in fact most of the time it was worse due to the shitty former Founders Edition blower cooler.

 

Now for those 1200$ we actually get a real bump in performance retaining single GPU solution, so the RTX 2080 Ti is not an overpriced GTX 1080 Ti, it is an actual good value for money TITAN Xp that makes sense when the former Xp didn't.

Problem is before you had the early higher price card then the cheaper card with similar performance with non critical features removed (less VRAM etc). Now we don't have that, we only have the higher price card.

 

Other generations that brought in new tech, new node or something else new didn't significantly raise the asking price. Turing didn't bring with it new challenges never faced before to justify the price increase.

 

Cost has always been high for Nvidia on their large dies difference now, and for a little while, is AMD has absolutely nothing close enough to the top end performance to require Nvidia to keep competitive pricing. We're in a you either pay up or go without situation and we all know it.

 

It is what it is but I'm not sure why anyone would actually defend the price increase when there is nothing to justify it. If it were a Titan T then sure were could all assume a cheaper Ti is coming but it's not a Titan T and it is the Ti.

[Speed Weed]

[S w a t s o n]

5 hours ago, Taf the Ghost said:

There's argument over whether it's actually working/doing, but a few seem to have been able to force it on RX Vega. 

https://wccftech.com/no-radeon-rx-vega-is-not-running-dxr-with-battlefield-v/

 

hmm

[asus killer]

just played B5 on a friend's house, thank god the game has RTX, if not no one would care for it. Gimmick worked.

 

On a side i hate Battlefield's graphical engine, everything seems like it's made of plastic, way to shinny.