Navi 21/23 Cards Rumored (aka "Nvidia Killers" xD)

[Results45]

1 hour ago, nick name said:

@DrMacintosh  Doesn't the new cheese grater have a double GPU?  Anyone seen specs/performance on that bad boy yet?

 

Yes, yes it does. You can even add wheels and make it 4 GPUs for a full freakin' lump sum of $45,000! :P

 

The guys at Techpowerup have the details (1 x dual-GPU card is 28.18 TFLOPS and 2 x cards is 56.36 TFLOPS): https://www.techpowerup.com/gpu-specs/radeon-pro-vega-ii-duo.c3268

 

1 hour ago, Beskamir said:

I'll believe it when I see it but so far AMD doesn't seem to have a RT or Tensor core equivalent and I doubt they'll have one when these cards come out.

1 hour ago, DrMacintosh said:

RT cores are completely useless tho. 

40 minutes ago, ApolloFury said:

A few 3D modelling software now uses RT cores, for example in Substance Painter on a GTX 1080 it would take 60 seconds to bake an ambient occlusion map. On the RTX 2060 it will only take 1 second. Blender will use RT cores to accelerate Cycle render. Unreal Engine and Unity will soon follow for lightmap baking / realtime.

 

https://www.substance3d.com/blog/substance-painter-summer-2019

https://code.blender.org/2019/07/accelerating-cycles-using-nvidia-rtx/

 

I mean sure RTX cards shave rendering times by up to 50% compared to equivalent cards without RT cores, but why would you need those if your 45-grand Mac (or PC) happens to have more than 4 Titan Vs worth of raw rendering power?

 

On a more serious note, AMD cards have neither tensor or RT cores so if your application happens to be optimized for use with Nvidia's new rendering cores than stick with RTX. Though if you're just doing general rendering or similarly GPU-intensive workloads and can afford the new Mac Pro, then go for it!

 

[ravenshrike]

4 hours ago, huilun02 said:

I believe AMD's goal right now is to crank frames, not kill frames. Raytracing is not essential.

Except they're doing their own HW accelerated implementation, just not completely separate cores. It just wasn't on Navi 10.

2 hours ago, Arika S said:

Christ, i give Nvidia shit about their naming convention, but AMD's is all up the shit

 

the last few releases we've had

RX5##

Vega 56/64

Radeon VII

5700/5700XT

and now NAVI 21/23

Navi 21/23 would be internal code names. 5700/XT was Navi 10 internally.

[RejZoR]

Well, I hope AMD does it. RX 5700 XT proves they can do it, now it's just a matter of scaling and updating future cards accordingly to be competitive.

[suicidalfranco]

3 hours ago, DrMacintosh said:

Which the vast majority of consumers don’t use. That “industry” isn’t that big. 

by that logic:

Intel quickSync is a niche feature cause the vast majority of consumers don't go around buying macbook just to use FC

And since you believe that "that "industry" isn't big" guess who is currently smaller than it? niche upon niche.

[DrMacintosh]

1 minute ago, suicidalfranco said:

Intel quickSync is a niche feature cause the vast majority of consumers don't go around buying macbook just to use FC

QuickSync is available on basically any Intel CPU made recently. Premier uses it, Handbreak uses it, OBS uses it. QuickSync isn’t what makes FinalCut Pro the best video editor for Mac users. 

 

But you are right. The “Pro” video editors are a very small segment of consumers. 

[suicidalfranco]

1 minute ago, DrMacintosh said:

QuickSync is available on basically any Intel CPU made recently. Premier uses it, Handbreak uses it, OBS uses it. QuickSync isn’t what makes FinalCut Pro the best video editor for Mac users. 

 

But you are right. The “Pro” video editors are a very small segment of consumers. 

And RT can be used for rendering, for gaming, for even rendering in excel now. Uses for it are coming and growing and we are only a year in (or within it's first year)

Quote

QuickSync isn’t what makes FinalCut Pro the best video editor for Mac users. 

it is

[DrMacintosh]

Just now, suicidalfranco said:

it is

It’s not. But it’s irrelevant to the conversation at hand anyway. 

 

Even once AMD adds their own version of ray tracing, I’m still going to call it useless unless AMD does it without using fixed hardware like Nvidia does. 

[Zodiark1593]

14 minutes ago, DrMacintosh said:

It’s not. But it’s irrelevant to the conversation at hand anyway. 

 

Even once AMD adds their own version of ray tracing, I’m still going to call it useless unless AMD does it without using fixed hardware like Nvidia does. 

If it takes fixed function hardware to make ray tracing viable to use in real-time applications, I don't think we should be terribly picky about it. 

[mr moose]

3 hours ago, Blademaster91 said:

 

People can keep finding reasons as to how they think RT is "useless" or "niche"

Maybe they wouldn't do that if they considered RT cores as simply an extra processing option that greatly accelerates some workloads (just like GPGPU's are too CPU's). But that would require people to be rational and not blinkered.

[ravenshrike]

26 minutes ago, Zodiark1593 said:

If it takes fixed function hardware to make ray tracing viable to use in real-time applications, I don't think we should be terribly picky about it. 

Some form of path tracing(Sound as well as light I believe) is confirmed in the upcoming consoles, and AMD did patent a hybrid approach. The question is whether that patent is what they're using or if they came up with another solution.

 

https://www.tomshardware.com/news/amd-patents-hybrid-ray-tracing-solution,39761.html

[Guest]

ill wait till these come out, the bread will be in the pudding. 

[RejZoR]

1 hour ago, DrMacintosh said:

It’s not. But it’s irrelevant to the conversation at hand anyway. 

 

Even once AMD adds their own version of ray tracing, I’m still going to call it useless unless AMD does it without using fixed hardware like Nvidia does. 

RT cores are not "fixed function". They are programmable units, but are currently dedicated to ray tracing. They can be utilized to compute any kind of ray tracing like operations. Same goes for Tensor cores. Fixed function would be explicitely ray tracing and nothing else. Which is what AMD is thinking about for future generations. In all honesty, I wouldn't have any problem with that. Fixed function ray tracing will be cheaper and probably faster since it would strictly do just that, but not as flexible and potentially with compatibility problems for the future where NVIDIA could just update it on driver level, though performance is questionable at that point.

 

Ray tracing is the future of PC graphics, it's just a matter of making it readily accessible to masses for as little pemium as possible. Only that way developers will bother doing it on a large scale and not only in games paid for by one graphic chip vendor.

[MeatFeastMan]

92CU's, GOOD LORD. That has to be water-cooled, doesn't it? And surely they're gonna have to lower clocks...even at 7nm+. If that is true though, GOODNIGHT Nvidia. The 92CU is pretty much guaranteed to destroy the 2080ti, provided Navi scales perfectly. And I'd imagine the gap between it and the 7nm 3080ti won't be too much.

 

5950XT 184CU beast, please do it AMD we need the new r9 295x2.

[Trixanity]

I think I can confirm this as fake already.

We already know that either Navi 12 or 14 (but probably both) are smaller than Navi 10. Granted, that proves only half of it as fake at best but if half is fake the likelihood of the rest being fake increases considerably.

 

That's an awful lot of words to say: counterfeit and homosexual.

[leadeater]

3 hours ago, DrMacintosh said:

Even once AMD adds their own version of ray tracing, I’m still going to call it useless unless AMD does it without using fixed hardware like Nvidia does. 

Fixed function hardware is vastly superior to general purpose performance wise, I mean chefs don't just use any knife because they are all able to cut things. Fixed function makes a lot of logical sense when there is a repeated performance intensive task that is required and Ray Tracing fits in to that on every front. The only issue is just how performance intensive Ray Tracing actually is and no matter how die area efficient fixed function hardware is it still requires space and that takes away from general purpose computation, balancing this is hard when we expect every generation to get faster and anything at all close to viable in regards to Ray Tracing requires a decent portion of the die to do it.

 

GPUs themselves were once considered fixed function or application specific and if you go back far enough CPU manufacturers were also trying to say they were useless and you could do it all on CPUs, they were wrong. 

[leadeater]

1 hour ago, RejZoR said:

RT cores are not "fixed function". They are programmable units, but are currently dedicated to ray tracing. They can be utilized to compute any kind of ray tracing like operations. Same goes for Tensor cores. Fixed function would be explicitely ray tracing and nothing else. Which is what AMD is thinking about for future generations. In all honesty, I wouldn't have any problem with that. Fixed function ray tracing will be cheaper and probably faster since it would strictly do just that, but not as flexible and potentially with compatibility problems for the future where NVIDIA could just update it on driver level, though performance is questionable at that point.

 

Ray tracing is the future of PC graphics, it's just a matter of making it readily accessible to masses for as little pemium as possible. Only that way developers will bother doing it on a large scale and not only in games paid for by one graphic chip vendor.

What AMD would be doing wouldn't be any different to Nvidia RT cores, there is no 'Fixed Function Ray Tracing' it's just the type of math used and you can utilize that mathematical function in more than one way. What mathematical functions you can get those types of cores to do may be limited but possible ways of utilize them is only limited by ingenuity of the programmer.

[RejZoR]

1 minute ago, leadeater said:

What AMD would be doing wouldn't be any different to Nvidia RT cores, there is no 'Fixed Function Ray Tracing' it's just the type of math used and you can utilize that mathematical function in more than one way. What mathematical functions you can get those types of cores to do may be limited but possible ways of utilize them is only limited by ingenuity of the programmer.

Fixed function would be strictly DXR only and nothing else. Which, if fast and cheaper, I'd be perfectly fine with.

[leadeater]

11 minutes ago, RejZoR said:

Fixed function would be strictly DXR only and nothing else. Which, if fast and cheaper, I'd be perfectly fine with.

Fixed function is what ever it is and DXR uses that fixed function hardware to do it, don't be tempted to mix technology labels with what is actually happening underneath. One of the functions required currently for RT techniques is BVH so any hardware that can do that math function could be utilized for DXR. DXR makes use of it's not an actual thing.

 

Edit:

Meaning if you need to do any BVH math for anything, RT or something else you have extremely fast BVH fixed function hardware. It isn't Ray Tracing only.

[RejZoR]

1 minute ago, leadeater said:

Fixed function is what ever it is and DXR uses that fixed function hardware to do it, don't be tempted to mix technology labels with what is actually happening underneath. One of the functions required currently for RT techniques is BHV so any hardware that can do that math function could be utilized for DXR. DXR makes use of it's not an actual thing.

I'm not mixing up anything. You can use regular rasterizer shaders to do ray tracing if you want. They are general purpose shaders. And it'll make ray tracing. But it's not going to be fast. If you create a purposely crafted logic on the GPU that only does ray tracing operations and does that very fast, but can't be used for anything else, that's fixed function.

[leadeater]

6 minutes ago, RejZoR said:

I'm not mixing up anything. You can use regular rasterizer shaders to do ray tracing if you want. They are general purpose shaders. And it'll make ray tracing. But it's not going to be fast. If you create a purposely crafted logic on the GPU that only does ray tracing operations and does that very fast, but can't be used for anything else, that's fixed function.

I think you're missing my point, you can label those cores Ray Tracing only as much as you like but the actual math they are doing is able to be utilized for other purposes not Ray Tracing. Where do you think GPGPU came from? People understood the reality that GPUs are only doing math functions and those can be utilized in other ways than just to render images, math is math and can be used in as many ways as we can think of, no Ray Tracing only label will make that no longer true.

 

Edit:

For example BVH can be used for collision detection or similar where you would like to know if objects are close to each other or touching.

[Doobeedoo]

If it really ends up with such number of CUs that'd be amazing indeed. Feels so far of though. 

[RejZoR]

23 minutes ago, leadeater said:

I think you're missing my point, you can label those cores Ray Tracing only as much as you like but the actual math they are doing is able to be utilized for other purposes not Ray Tracing. Where do you think GPGPU came from? People understood the reality that GPUs are only doing math functions and those can be utilized in other ways than just to render images, math is math and can be used in as many ways as we can think of, no Ray Tracing only label will make that no longer true.

 

Edit:

For example BVH can be used for collision detection or similar where you would like to know if objects are close to each other or touching.

That's just not true. If "math is math" was true, no one would bother implementing dedicated video decoders or other similar logics within GPU's. These are essentially fixed functions. If you have something specifically made and optimized, that's fixed function. You can maybe repurpose it, but it won't be optimal. Like the way how pipelin for that functions and may work great for ray tracing, but not for something else. Which is what RT cores are in general. If that wasn't the case, NVIDIA would just ram more compute shaders and call it a day. I'm fairly certain you can go even more focused than RT cores.

[cj09beira]

10 hours ago, DrMacintosh said:

Can we get mobile Navi please? 

that should be the one to launch next as its the only one with performance leaks already out 

 

 

 

about these specs they dont seem to make much sense, why would they make 2 different dies with such similar specs 

 

8 hours ago, Blademaster91 said:

Neither is a high end GPU, but like ray tracing it's nice to have in certain games and improves the gaming experience.

As for whether the cards are an "nvidia killer" I'll believe it when I see it,AMD needs to hold back on being cocky until they actually have a high end GPU.

navi 2 should have some sort of raytracing if the rumors are correct, (its probably also the same architecture that the consoles will use and those for sure will have it)

[leadeater]

1 hour ago, RejZoR said:

That's just not true. If "math is math" was true, no one would bother implementing dedicated video decoders or other similar logics within GPU's. These are essentially fixed functions. If you have something specifically made and optimized, that's fixed function. You can maybe repurpose it, but it won't be optimal. Like the way how pipelin for that functions and may work great for ray tracing, but not for something else. Which is what RT cores are in general. If that wasn't the case, NVIDIA would just ram more compute shaders and call it a day. I'm fairly certain you can go even more focused than RT cores.

Then remind me again how GPUs can only be used for rendering images and cannot be used for other purposes and don't exist in every modern super computing cluster in the last 10 years?

 

Any hardware, any, that is doing math i.e all of them can be used for anything that math applies to. No matter how small and focused, if it's only a single thing it can be used for as many purposes as you can think of. The limitation is only in the ways you can think of to utilize that math or function. Yes not everything is easily reuable or makes sense to, as with video decoders those are not doing just one thing, one math calculation and has specific rules and flows for it's usage, not everything is that.

 

Nvidia made these cores for a specific task, purpose designed only for this one task. You know what those are called? RT cores. These cores can be used for more than just Nvidia RTX games because the math they are doing is not limited to only that usage. Why can they be used for more than that? Because all that is being done is BVH calculations.

https://www.anandtech.com/show/13282/nvidia-turing-architecture-deep-dive/5

 

If AMD is to make an accelerated hardware path for Ray Tracing it is required that they make this ASIC be able to do BVH. So if RT cores can be used for more than just Ray Tracing for RTX games AMD equivalent can too because they are doing the same thing. The only thing that would prevent this is either AMD or Nvidia not exposing that hardware through an API for you to use closing it off.

 

RT cores = BVH calculators = Applicable to more than Ray Tracing = Can do more than Ray Tracing.

[RejZoR]

7 minutes ago, leadeater said:

Then remind me again how GPUs can only be used for rendering images and cannot be used for other purposes and don't exist in every modern super computing cluster in the last 10 years?

 

Any hardware, any, that is doing math i.e all of them can be used for anything that math applies to. No matter how small and focused, if it's only a single thing it can be used for as many purposes as you can think of. The limitation is only in the ways you can think of to utilize that math or function. Yes not everything is easily reuable or makes sense to, as with video decoders those are not doing just one thing, one math calculation and has specific rules and flows for it's usage, not everything is that.

 

Nvidia made these cores for a specific task, purpose designed only for this one task. You know what those are called? RT cores. These cores can be used for more than just Nvidia RTX games because the math they are doing is not limited to only that usage. Why can they be used for more than than? Because all that is being done is BVH calculations.

https://www.anandtech.com/show/13282/nvidia-turing-architecture-deep-dive/5

 

If AMD is to make an accelerated hardware path for Ray Tracing it is required that they make this ASIC be able to do BVH. So if RT cores can be used for more than just Ray Tracing for RTX games AMD equivalent can too because they are doing the same thing. The only thing that would prevent this is either AMD or Nvidia not exposing that hardware through an API for you to use closing it off.

 

RT cores = BVH calculators = Applicable to more than Ray Tracing = Can do more than Ray Tracing.

Then why my GTX 1080Ti sucks at ray tracing? I got shitloads of compute units?!