-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Crytek demo DXR on Vega 56
- Thread starter Harlequin
- Start date
Permabanned
- Joined
- 12 Sep 2013
- Posts
- 9,221
- Location
- Knowhere
No, they can't unfortunately
Shame , I'll stash them away till AMD have their RT capable gpu's, maybe they'll be able to then or maybe I'll get a 3rd or 4th gen RTX card one day.
Shame , I'll stash them away till AMD have their RT capable gpu's, maybe they'll be able to then or maybe I'll get a 3rd or 4th gen RTX card one day.
They are RT capable now, just as GTX card's are, they just don't have dedicated RT hardware and i think if you're expecting that from future AMD GPU's you maybe disappointed, or maybe not, its useful in AI and RT accaleration from a productivity stand point, its the real reason the RTX card's have it, but it makes them large and expensive for Retail cards and its not necessary when implemented properly.
Depends if AMD like nVidia want to flog you the same workstation card's at a premium.
The Demo was run on a Vega 56, remember... not a particularly powerful card and not "RTX".
I mean surely that's just down to how taxing doing the ray tracing is? Half resolution of 1440p is still lower than 1080p, it might not be by much, but it's still lower. If the ray tracing is such a performance tank, which I would have thought it would be, then I don't see why 1440p @half res getting 40FPS is so unbelievable. More numbers are needed obviously
Again, 720p is not half the resolution of 1440p, half is roughly about 1800*1012. I mean it's possible the guy said half resolution instead of quarter resolution by mistake, but considering he says "much better performance" I would imagine resolution has a significant impact to performance. The difference between 1080p and 1012p is about 11%.
I agree 40FPS is still too slow, shame they didn't test other card, RTX and GTX. Would be interesting to see the results, especially RTX actually. No so much to see how much better it is compared to V56, but assuming it'd take advantage of the hardware, to see how much better/worse it would be than games using RT right.
Shame , I'll stash them away till AMD have their RT capable gpu's, maybe they'll be able to then or maybe I'll get a 3rd or 4th gen RTX card one day.
In theory, AMD simply have to make a DXR compatible driver and the demos will run, as will BFV, Metro exodus etc. The fact that AMD haven't tried to release a DXR driver is somewhat telling of the performance you will get, i.e. useless. Exactly the same as Pascal running ray tracing.
No, 1080p is almost exactly half the resolution of 1440p. Remember, doubling 1 dimensions quadruples the total number of pixels. The total pixel count doubles in the sequence:
720p, 1080p, 1440p, 4K
Half 1440p is 1080p which only got 30FPS. To increase the FPS Crytek lowered the ray tracing resolution down to 720p in order to get enough performance that allowed 1440p screen at 40FPs.
If you are thinking that exactly half 1440p is slightly less than 1080p then the math does work. Half 1440p would only be 6% less than 1080p. If you believed that 100% of the performance was entirely dictated by ray tracing then that 30FPS would increase to 31.8FPS, but that is a long way form 40FPS and completely ignores the fact that the rest of the rendering has now gone form 1080pto 1440p which is almost double and will certainly have a big impact.
So no, the only way the numbers work is at 1440p the ray tracing was dropped to 720p. It is simple, it follows exactly the logic laid out in the article, and intuitively it makes sense. The wording that is confusing is when it talks about halving the ray tracing resolution, they don't mean half of 1440p, they mean half of 1080p.
They talk about a baseline figure of 1080p screen and 1080p ray tracing giving 30FPS. That is the baseline, the article states halving the ray tracing resolution then improves performance to the point that a 1440p screen resolution is acceptable.
at the end of the day we know Nvidia's dedicated hardware is about 8-10X faster than a software solution on the GPU when ray tracing is heavily utilized. There is absolutely zero reason to believe that AMD's GPus without hardware will behave much different to Pascal. AMD GPUs have slightly more raw compute but the differences these days are actually a lot smaller. Nvidia have been supporting a software ray tracing on the GPU long before AMD did and have better industry support in software.
As of today there is no apples to apples comparison possible until crytek release their demo or AMD make DXR drivers. When this happens I am absolutely certain that the difference between Pascal and vega running software ray tracing on the GPU will be within 10-15%, possible AMD on top. But the performance will be so dire compared to the RTX cards it is entirely irrelevant.
We already know you can do some very hacked and limited ray tracing in real time. For years Nvidia have had this in their games works library and it has been used in games.
There is no such thing as "Software or Hardware Ray Tracing" its just something that's built into X or Y application and uses the Hardware's compute power, the performance of [enter whatever Hardware here] simply depends on its Floating Point performance, you can include CPU's in that.
Oh so you do get it.
1080TI 11.3 TFlops FP32
Vega VII 13.4 TFlops FP32
The Crytek Neon Noir Demo is GPU, API (Driver) agnostic, which is why it works on Vega 56, why would AMD need to make a Driver for something that doesn't need a Driver, i refer you to....
As nVidia do with everything else, as they are now with RTX they have taken something already in Vendor Agnostic existence and turned it into a Propriety solution (very very badly i might add) in order to con you into thinking you need their hardware to do something you can do better outside of that Proprietary solution.
PS: the PS5 will have "Ray Tracing" not because Navi is some AMD RTX card but because Sony thinks its a good idea to jump on the bandwagon and Navi has plenty of compute power to run it, as Vega 56 has, and Pascal
Last edited:
Why do people keep measuring tflops on base clocks? Not having a dig at you humbug, you obviously grabbed the numbers off google
I guess boost clocks aren’t guaranteed but it just makes products sound worse than they are. The 1080ti is actually at 13 to 14t flop under load when it’s clocks boost up, no sure on the Radeon 7 as I don’t own one.
The 2080ti is quotes as 13.5, but I’ve benched mine at 19tflop under load (it’s overclocked as well)
I guess boost clocks aren’t guaranteed but it just makes products sound worse than they are. The 1080ti is actually at 13 to 14t flop under load when it’s clocks boost up, no sure on the Radeon 7 as I don’t own one.
The 2080ti is quotes as 13.5, but I’ve benched mine at 19tflop under load (it’s overclocked as well)
Last edited:
a bit more crytek stuff/https://wccftech.com/cryengine-5-7-...tx12-vulkan-raytracing-available-spring-2020/ 
This is where you show your ignorance. When instructions are run on general computation hardware, be that a CPU or a modern GPU's processor is is reffered to as a software solution. If there is dedicated hardware that serves to only accelerate those particular instructions then it is a hardware solution. AMD do not have hardware accelerated ray tracing, Nvidia does.
Meaningless because this doesn't reflect real world clock rates. But even if we took these figures, and assumed AMD's driver team is just as good as Nvidia's, despite the fact that nvidia have been developing GPU based ray tracing much longer and have a far larger driver team with more funding, then the Radeon 2 would be 18% faster than a 1080ti, except at the same price you could buy a 2080 which would be 8-10x faster.
DXR is GPU agnostic. if AMD ever want to support ray tracing in games they better support the official micosoft DX standard for it that is already used in multiple games.
What absolutely completely fake nonsense. Nvidia worked with Microsoft to generate an industry standard called DXR which is now part of Direct X, they did the same for Volcan. tehre is sboltuely nothing property about the API Nvidia hardware exposes.
Navi and Vega 56 absolutely do not have enough compute power to run ray tracing in any maningufl way. The Crytek demo actually proves that. Even with massive hack the vega 56 is barely playable trying the most simplest effects.
Turing cards are literrly 10x faster than Vega 56 and they only just have enopugh power and even that requires careful design choices.
So the Crux of your argument is, only nVidia have real Ray Tracing and everything else is fake.
Predictably.
What i can't work out is are you trolling or have you fallen for nVidia's marketing?
That isn't even close to what he said and no idea where you got that?
Yes it is and i'm sick of going round and round in circles correcting him over and over and over....
Sorry but you are massively wrong here and need to read what he writes.... I will pick some key points.
When instructions are run on general computation hardware, be that a CPU or a modern GPU's processor is is reffered to as a software solution. If there is dedicated hardware that serves to only accelerate those particular instructions then it is a hardware solution. AMD do not have hardware accelerated ray tracing, Nvidia does.
Which is factually correct.
DXR is GPU agnostic. if AMD ever want to support ray tracing in games they better support the official micosoft DX standard for it that is already used in multiple games.
Which is factually correct.
Nvidia worked with Microsoft to generate an industry standard called DXR which is now part of Direct X
Which is also factually correct. I am not seeing where he is saying "only NVidia have real ray tracing and everything else is fake"? What he is saying is spot on and not sure where you get he is trolling from either. He makes some very valid points.
That's a semantical argument you're trying to use as a "gotcha" thing, i know the difference between dedicated Ray Tracing acceleration on the RTX card's vs not, i have explained it a couple times in this thread which you've ignored for your fake engineered "gotcha" moment.
It changes nothing, Yes the RTX cards have dedicated RT acceleration, what you and D.P fail to understand is what nVidia using dedicated Ray Tracing acceleration hardware for Ray Tracing is, Ray Tracing is NOT nVidia RTX, nVidia RTX is acceleration hardware, Ray Tracing is entirely independent from that and any hardware with compute ability can run it, it was certainly NOT invented by nVidia or Microsoft.
AMD cannot run Ray Tracing in those nVidia Ray Tracing titles that have it, not because AMD don't have Ray tracing ability, they do, NOT because they are RTX cards, but because nVidia wrapped the Ray Tracing component in a proprietary API wrapper that locks AMD out of it.
PS: if nVidia proprietary API wrapped Ray Tracing was not so horrendously optimised you wouldn't need the dedicated RT core hardware acceleration to run it at any reasonable performance, that was done IMO for you to dig deep into your pockets for the hardware. Kerching!
Last edited:
How did you get 6%? My brain has been struggling to work out the maths of it so I'd thought I'd just ask!
For note I ended up with ~11% doing (a-b)/a where a and be are the total number of pixels for 1080p and 1440p/2As for the rest of your post, I get what you're saying, and I see your logic behind it, but I've still interpreted what they said the other way - half the resolution at the resolution they are currently talking about, in this case 1440p. The only way to really know is if Crytek come out with numbers - seems odd they didn't. Also I'd think we can completely ignore the fact that the rest of the rendering went from 1080p to 1440p. I'd assume the ray tracing is the bottleneck holding it all back tbh, even at 1440p.
I wouldn't get hung up on raw TF numbers anyhow - it would take a tech deep dive and/or specific type of benchmark to know the actual throughput of the type of calculations used for ray tracing on any given architecture which potentially can be significantly different to its on paper numbers.
It is the benefit of having a block of compute dedicated to ray tracing because it can be tuned to bulk process those kind of calculations very quickly with the pipeline optimised for handling them without having the considerations of other types of processing to worry about.
What has increasingly been lost sight of in this topic is the difference between full scene ray tracing, path tracing and the various different hybrid approaches some of which might barely use any ray tracing at all and little to none for directly shading pixels and mostly use ray picks as hints to where and how to use traditional techniques to fake up something that looks close to ray tracing and the further away from true full scene ray tracing you get the more negatives the implementation will have in terms of how plug and play it is and how much special casing you have to do to avoid issues, etc.