Does the RTX series create an openning for AMD?

[maddie]

So long as Nvidia stands firm on performance-per-$ stagnation, AMD has an opening. If PP$ doesn't increase, then consumers with hardline budgets can literally never upgrade.

The problem is NV is so well executed. Yes PP$ stagnated at 2070+. But NV knows AMD will likely only be able to compete at 2060 levels (look at RX 480 vs 1060). 2060 will likely not have RTX, so they are cheaper to make and NV can increase PP$. As odd as it seems right now, we could have $600 2070 and $300 2060 and a huge gap in between. NV has essentially separated 2070 and 2060 like Intel used to separate quad core and hex core. Just speculating their strategy anyway.

So AMD needs to be able to compete a raw performance with the 2070 - 2080 Ti to really capitalize. If they follow the Polaris strategy NV will be ready.

A critically important statement.

Regarding RT tech, I've glanced over the PowerVR work and it appears to far surpass Nvidia's in efficiency. Why is it so ridiculous to imagine a license by AMD? They appeared to have done a lot of this to get Zen out quickly. AMD would also have suspected about RT for some time. Industrial snooping is eternal.

Regarding multi-die Navi, I don't know if I believe the last AMD statement that it can't work. It's obvious that Koduri released that initial slide in error. Replacement slides quickly deleted any mention of that possibility. Then we had AMD claiming that multi-die was too much like Xfire. Now we seem to have Nvidia returning to SLI with the introduction of NVlink. AMD spent a lot of R&D into IF with the proudly stated ability to scale very high. It seems that SLI/Xfire can work once the bandwidth exists. All quite interesting, in my view, and something to keep an eye on.

 

[tamz_msc]

I don't understand all of this doom and gloom talk for AMD GPUs. From all of the benchmarks I have seen (As far as gaming performance is concerned), The Vega 56 and Vega 64 are not that far off of The Nvidia 1070/1070ti and 1080. Even if the expectation was that the Vega 64 was supposed to compete with the 1080 ti, but it turns out its closer to just 1080 performance, Vega 64 still does just fine in games. Vega is not as efficient, runs hotter etc. but still I just don't understand this bandwagoning hate on AMD GPUs especially if you are just looking at game performance. Vega 64 may not be as fast as a 1080 in most games, but that doesn't mean the Vega 64 can't run games at good performance levels. They just need to lower the price of Vega. Price is the biggest issue for Vega IMO. For AMDs future competitor for RTX, if they price it competitively, it should be fine.

The doom and gloom for AMD's future discrete GPU prospects has a reason. Vega doesn't even feature in the Steam HW survey and GP106(GTX 1060) has almost 13x the number of users than Polaris 10(RX 480/580 and 470/570).

 

[richaron]

I don't understand all of this doom and gloom talk for AMD GPUs. From all of the benchmarks I have seen (As far as gaming performance is concerned), The Vega 56 and Vega 64 are not that far off of The Nvidia 1070/1070ti and 1080. Even if the expectation was that the Vega 64 was supposed to compete with the 1080 ti, but it turns out its closer to just 1080 performance, Vega 64 still does just fine in games. Vega is not as efficient, runs hotter etc. but still I just don't understand this bandwagoning hate on AMD GPUs especially if you are just looking at game performance. Vega 64 may not be as fast as a 1080 in most games, but that doesn't mean the Vega 64 can't run games at good performance levels. They just need to lower the price of Vega. Price is the biggest issue for Vega IMO. For AMDs future competitor for RTX, if they price it competitively, it should be fine.

I think the comparisons to Vega are useful for more than the reasons stated.

To simplify it; Vega also held the promise of groundbreaking new technologies which rely on drivers and developers to take full advantage of. Obviously nV has the advantage in the drivers department (I still think AMD has a distance to go with Vega & derivatives) but the question of getting developers on your side and/or actually fulfilling PR hype is where I think this thread has merit.

If nVidia makes the same mistake with lofty PR claims and essentially wasted silicone when the real world performance is less than impressive I think there's a real opening for AMD. And no, I don't think Turing's "lofty PR and 'wasted' silicone" is any more impressive than Vega's, in fact AMD has been working on their unrealized dreams for longer so I think it's just as likely they will pan out as the chance everyone will jump onto the Gigarays bandwagon.

 

[USER8000]

Don't underestimate Intel - the bigger threat for Nvidia is the lower sections of the market,where Intel and AMD can try to enter from. So all those prebuilt PCs and laptops and so on. Intel can sell at a loss and absorb it more than AMD or Nvidia can. Look at the first pictures of their future card:

https://videocardz.com/newz/intel-teases-discrete-graphics-coming-2020

From what I have heard,it will have a single slot cooler and sub 150W TDP. Intel is going after mainstream first.

 

[PeterScott]

2080ti - Titan level $999
2080 - 1080ti level $699
2070 - 1080 level $499
2060 - 1070 level $399
2050 - 1060 level $299

Perhaps?

2060 - 1070 level $399 - Isn't 1070 MSRP $349, and in theory there is no RT HW as a bonus.
2050 - 1060 level $299 - Isnt' 1060 MSRP $249, see above.

I realize those cards are selling above MSRP, but I can't see NVidia getting away with introducing the same performance ~2years later, at higher MSRP, and no compensating features like Ray Tracing. You may as well just rebrand them and keep selling them at current pricing if that is all they are going to do.

As much as people are complain about the price of the new cards, the real obscenity is these old midrange cards still selling above their MSRP.

 

[french toast]

Sad reality is this; AMD could release Vega 20 7nm for desktop, could equal 2080 on rasterized games, be 25% cheaper and STILL be outsold 10-1.
There is no consumer trust in AMD graphics anymore, the reasons have been known for years...R&D resources, bad management.
GCN is out of date, even on 7nm it will look like a dinosaur compared to Nvidia RTX architecture, combined with brand value...AMD is dead in graphics.

They could make a comeback in 2020 with a new architecture, AMD finances are MUCH better than they have been for over a decade, by 2020 they could in theory do a top to bottom next gen launch and compete well.
Just need to hang in there.

 

[LTC8K6]

2060 - 1070 level $399 - Isn't 1070 MSRP $349, and in theory there is no RT HW as a bonus.
2050 - 1060 level $299 - Isnt' 1060 MSRP $249, see above.

I realize those cards are selling above MSRP, but I can't see NVidia getting away with introducing the same performance ~2years later, at higher MSRP, and no compensating features like Ray Tracing. You may as well just rebrand them and keep selling them at current pricing if that is all they are going to do.

As much as people are complain about the price of the new cards, the real obscenity is these old midrange cards still selling above their MSRP.

I think it will be RTX2060 and RTX 2050 and they will feature ray tracing. I think I read somewhere that NV said that all of the 2000 series cards have RT. For the lower end cards, you'll just have to back off the detail, as usual.

 

[PeterScott]

I think it will be RTX2060 and RTX 2050 and they will feature ray tracing. I think I read somewhere that NV said that all of the 2000 series cards have RT. For the lower end cards, you'll just have to back off the detail, as usual.

I really can't imagine that working out. It doesn't really seem like you can scale RT and denoising to 2060/2050 class cards yet. I wouldn't even look to 2070 for Ray Tracing. Or are we going to only have Ray Tracing at 640x480 on the 2050?

The more you limit rays, they noisier your image gets, and thus you likely need even more Tensor performance to denoise it. I think there is critical size you can't go below for this kind of HW combo for RT.

It will certainly be interesting to see what happens with 2060/2050, though they look far off, maybe they are 7nm cards in the middle of next year, devoting 2/3 of the die to RT HW, but I really doubt it.

 

[LTC8K6]

I really can't imagine that working out. It doesn't really seem like you can scale RT and denoising to 2060/2050 class cards yet. I wouldn't even look to 2070 for Ray Tracing. Or are we going to only have Ray Tracing at 640x480 on the 2050?

The more you limit rays, they noisier your image gets, and thus you likely need even more Tensor performance to denoise it. I think there is critical size you can't go below for this kind of HW combo for RT.

It will certainly be interesting to see what happens with 2060/2050, though they look far off, maybe they are 7nm cards in the middle of next year, devoting 2/3 of the die to RT HW, but I really doubt it.

I would think RT at 1080 for the low end cards.

NV said you need 5 Gigarays/sec for a high res full room, I think. 2070 does 6.

 

[maddie]

I would think RT at 1080 for the low end cards.

NV said you need 5 Gigarays/sec for a high res full room, I think. 2070 does 6.

5 Gigarays/sec. Thats a very misleading number taken in isolation, and I can't understand why it would be quoted. At what framerate is the important question.

 

[PeterScott]

I would think RT at 1080 for the low end cards.

NV said you need 5 Gigarays/sec for a high res full room, I think. 2070 does 6.

That doesn't leave much room to reduce the RT HW in a 2060, let alone a 2050. That way you end up with a card that has a lot more RT than SM HW proportionally. And a much bigger potential opening for AMD to exploit.

If you are spending 2/3 of the die on RT HW and 1/3 on SM HW, AMD can do a card with 100% SM HW and surpass you in normal usage and critically in laptops.

At some point the RT HW Tax will be too high and I bet that is around 2060 level.

 

[krumme]

Sad but true. AMD could build a 2070/2080 competitor with no Ray Tracing HW and sell for $100 less than NVidia. NVidia would drop their price by $50 and everyone would still pay the $50 extra to get the NVidia card.

680 sold far better than 7970 even if it was slower. Same for 780 vs 290 series.

Nv is just well run all over and have agressive and compettent marketing and sales to top it off. Their technical marketing is just on another planet vs amd. They also kicked 3dfx even if they had the slower and worse card. And with a very small organisation. Today they are gigantic and still as effective but also damaging to the market.

Amd sells consoles and apu. Makes sense. Thats the strategy laid out years ago.

Doesnt make sense to even try to beat nv with some miracle 5850 to find it doesnt change market. Today a gpu = nvidia. Their brand is huge.

 

[Kenmitch]

https://www.techspot.com/amp/news/76103-nvidia-putting-squeeze-aib-partners.html

Doesn't look good.

 

[PeterScott]

https://www.techspot.com/amp/news/76103-nvidia-putting-squeeze-aib-partners.html

Doesn't look good.

Huge grain of salt needed there. Digitimes is a pure clickbait rumor mill. They have all the accuracy of a stopped clock.

 

[LTC8K6]

A glut of cheap 10 series cards would be the last thing NV would want, I'd think.

 

[Despoiler]

AMD definitely has an opening, but it doesn't seem like they are poised to exploit it on the consumer side or at least they are waiting to see what Turing's performance is before deciding. They are going for business dollars leading with 7nm Instinct cards late this year. That's all we know. If they were ready to release a top to bottom stack of 7nm cards and those cards were priced well for their performance they would probably mop up. That being said if AMD can fast follow a consumer GPU before mid next year I don't think Nvidia will want to move to 7nm quite that fast. They usually release a new gen somewhere around 1.5 years.

 

[ub4ty]

Beyond the glitz and glamor of these often over-hyped and over-marketed proprietary approaches to speeding up gaming performance from both AMD/Nvidia, is raw performance. In raw performance, Vega is a quite compelling compute platform that matures by the day as they get features working and drivers refined.

https://www.techpowerup.com/gpudb/b4893/powercolor-rx-vega-56
FP32 : 11Tflops
FP16 (2:1) : 21 Tflops

Find a comparable Nvidia card w/ this performance. AMD goes after the raw performance while shaping open source paradigms around it. Nvidia goes for non-standard meme cores and shapes proprietary paradigms around it.
http://blog.gpueater.com/en/2018/04/23/00011_tech_cifar10_bench_on_tf13/
http://blog.gpueater.com/en/2018/03/07/00003_tech_flops_benchmark_1/

It's the Intel/AMD paradigm.
I'm quite confident that an opening was created way back when vega was introduced. The drivers, open source ecosystem, and vulkan are evolving around it.

At these prices, Nvidia has now brought negative attention to themselves and people will now begin piecing apart their meme architecture and looking for flaws. It was exactly the wrong thing to do for a new and unproven architecture. In that, they will find many failed preparatory standards. A large footprint of hardware locked behind their new cloud services model. By comparison to AMD, people will discover similar raw compute capabilities w/o the proprietary nonsense. AMD's problem are the drivers and tooling which seem to be improving with time which was ofc going to take longer than an completely proprietary in house approach w/ a much more focused micro-architecture.

Take ray tracing for instance.... I hope people don't actually believe the new proprietary ray tracing section of Nvidia's GPU pipeline is doing all of the ray tracing compute completely in parallel. Obviously it isn't which is why you take a performance hit when you enable it... Why? Because the ray tracing cores only do a portion of the ray tracing whereas a large amount is still emulated in the traditional pipeline.

Then you have DLSS which is literally a meme feature on top of meme tensor cores. The new architecture is like an eternal beta test platform. None of the new features are standalone pipeline compute regions. They're like teaser compute units.. Just enough to call it a new feature while not achieving much. The real question is what they did in terms of the SMs to speed performance. That's all that really matters. In this, there has always been an opening.

7nm for both Nvidia and AMD is where the real comparison starts. It's a point in which they both will have had enough time to start locking in the makings of a new and bold platform outside of meme beta feature sets.

Ray tracing belong on a completely separate Die imo w/a high level, low latency copy/coherent BVH tying it to the main GPU pipeline. If you had any awareness of what it takes computationally and memory wise to do more proper ray tracing, you'd understand any reference to it on a monolithic die is a meme.

As such, everyone is on the same footing.. Gimmick emulated speed up features. Tensor cores are memes for AI and Ray tracing cores are memes for Ray tracing. Everyone is going to have to go MCM to be competitive and I don't see this in GPUs yet. This and Intel is coming into the game hard mode and many others are making dedicated accelerators just as I indicated. The future is anyone's. Were in a big hardware boom/wave. I'm not betting on any particular horse nor am I becoming indebted to someone's dev board level premature architecture.

 

[Tup3x]

^^Too much bias...


This creates an opening for Intel. I believe they even demonstrated some kind of ray tracing with Larrabee ages ago. I wouldn't be surprised if they are planning to challenge Nvidia in ray tracing too with their upcoming cards.

 

[LTC8K6]

2080ti looks pretty comparable to Vega64 in the FP32 and FP64 department.
https://www.techpowerup.com/gpudb/3305/geforce-rtx-2080-ti

 

[ub4ty]

^^Too much bias...


This creates an opening for Intel. I believe they even demonstrated some kind of ray tracing with Larrabee ages ago. I wouldn't be surprised if they are planning to challenge Nvidia in ray tracing too with their upcoming cards.

Too much bias? I mentioned Intel...
There's also NEC :
https://www.anandtech.com/show/13259/hot-chips-2018-nec-vector-processor-live-blog
Xilinx :
https://www.anandtech.com/show/13256/hot-chips-2018-xilinx-dnn-processors-live-blog
ARM :
https://www.anandtech.com/show/13253/hot-chips-2018-arm-machine-learning-core-live-blog
And all new entrants :
https://www.anandtech.com/show/13255/hot-chips-2018-tachyum-prodigy-cpu-live-blog

No one owns the future. Not AMD/Nvidia. I made this clear in my biased post and just about every post I made leading up until this point. Were in a Hardware era. In 10 years, I expect someone other than Intel/AMD/Nvidia to be dominating AI based hardware computing. Big companies get greedy and stagnant as opposed to innovating and delivering value. It's why they all end up being eventually disrupted.

Architectures for Ray tracing nor meme learning have matured. There's lots of big hail marries which one expects at the start of a race. Raw and versatile compute pan out better in the long run. Imagine an architecture where tensor/ray tracing cores merge into a more generic and singular paradigm with better coupling and integration in the GPU pipeline.... Imagine someone breaking this functionality out to a much more performant accelerator w/ 1/4th the power consumption...

Nothing is final about what anyone has presented. They're immature and first attempts at functionality hat has a long way to go. As a consumer, I'm not paying a premium for such architectures. That reflects intelligence not bias.

 

[Flayed]

Why all the hate for AMD graphics? I thought the 570 and 580 were well received and compete with the 1060 3G and 6G cards.

 

[ub4ty]

A critically important statement.

Regarding RT tech, I've glanced over the PowerVR work and it appears to far surpass Nvidia's in efficiency. Why is it so ridiculous to imagine a license by AMD? They appeared to have done a lot of this to get Zen out quickly. AMD would also have suspected about RT for some time. Industrial snooping is eternal.

Regarding multi-die Navi, I don't know if I believe the last AMD statement that it can't work. It's obvious that Koduri released that initial slide in error. Replacement slides quickly deleted any mention of that possibility. Then we had AMD claiming that multi-die was too much like Xfire. Now we seem to have Nvidia returning to SLI with the introduction of NVlink. AMD spent a lot of R&D into IF with the proudly stated ability to scale very high. It seems that SLI/Xfire can work once the bandwidth exists. All quite interesting, in my view, and something to keep an eye on.

What anyone should glean from looking into Power VR's Ray tracing feature is that they had it years ago in hardware and that it came via an acquisition of a startup operating on 3 million of capital in a ~20million buyout. Furthermore and more fundamentally that the algorithms that underly ray tracing have been pretty much solidified and standardized and that its asic level functionality. It centers on shared access of the BVH between the traditional GPU pipeline. In Nvidia's current implementation, the ray trace cores are just a ray tracing kick starter. A good amount of it still occurs in the traditional GPU pipeline. Such a feature, if it can be paralyzed nicely and interfaced at low enough latency is ripe for MCM/asic breakout. This is actually something you don't want want bloating your die foot print if you don't have to. If a small startup could have implemented the same thing over a decade ago, I'm sure both Nvidia/AMD are able to do so in a short span of time. I'm absolutely not buying Nvidia's claim that they were working on this for over a decade.. it seems more like some tensor core after thought..

 

[Ranulf]

Why all the hate for AMD graphics? I thought the 570 and 580 were well received and compete with the 1060 3G and 6G cards.

Ignorance and propaganda. AMD/retailers have already lowered the prices on the 570-580s. Amazon had a MSI 580 8gb on sale the other day for $229 with a $20 rebate. NE has cards down to $240-275 with 3 free games. 570s are going under $200 now with free games. 1060s are still near MSRP or higher, $270-310.

 

[del42sa]

To simplify it; Vega also held the promise of groundbreaking new technologies which rely on drivers and developers to take full advantage of. Obviously nV has the advantage in the drivers department (I still think AMD has a distance to go with Vega & derivatives) but the question of getting developers on your side and/or actually fulfilling PR hype is where I think this thread has merit.

year 2018, does anybody still believe in this ? That AMD needs just fix the drivers ? Ridiculous.... :-D

those "groundbreaking" technologies you mention are just HOAX. Mind you that no one except some few people from AMD labs did not seen them working actually ??? AMD never released any SDK for Primitive Shaders or NGG, so how can developers use them ??? Nobody really knows if it´s really working feature. AMD made a fancy presentation before launching VEGA 10, "groundbreaking" technologies, Primitive shaders, NGG, DBSR, etc.... just to make Vega more attractive in the eyes of buyers, but those features are just on the paper. How we can measure them ? huh? In reality it´s just Fiji with RPM and thats the only thing we can really test/measure. In 2020 there will be new GPU architecture (hopefully), nobody will care about Vega "super - truper" features. They simply don´t work. deal with it !

 

[pj-]

Beyond the glitz and glamor of these often over-hyped and over-marketed proprietary approaches to speeding up gaming performance from both AMD/Nvidia, is raw performance. In raw performance, Vega is a quite compelling compute platform that matures by the day as they get features working and drivers refined.

https://www.techpowerup.com/gpudb/b4893/powercolor-rx-vega-56
FP32 : 11Tflops
FP16 (2:1) : 21 Tflops

Find a comparable Nvidia card w/ this performance. AMD goes after the raw performance while shaping open source paradigms around it. Nvidia goes for non-standard meme cores and shapes proprietary paradigms around it.
http://blog.gpueater.com/en/2018/04/23/00011_tech_cifar10_bench_on_tf13/
http://blog.gpueater.com/en/2018/03/07/00003_tech_flops_benchmark_1/

It's the Intel/AMD paradigm.
I'm quite confident that an opening was created way back when vega was introduced. The drivers, open source ecosystem, and vulkan are evolving around it.

At these prices, Nvidia has now brought negative attention to themselves and people will now begin piecing apart their meme architecture and looking for flaws. It was exactly the wrong thing to do for a new and unproven architecture. In that, they will find many failed preparatory standards. A large footprint of hardware locked behind their new cloud services model. By comparison to AMD, people will discover similar raw compute capabilities w/o the proprietary nonsense. AMD's problem are the drivers and tooling which seem to be improving with time which was ofc going to take longer than an completely proprietary in house approach w/ a much more focused micro-architecture.

Take ray tracing for instance.... I hope people don't actually believe the new proprietary ray tracing section of Nvidia's GPU pipeline is doing all of the ray tracing compute completely in parallel. Obviously it isn't which is why you take a performance hit when you enable it... Why? Because the ray tracing cores only do a portion of the ray tracing whereas a large amount is still emulated in the traditional pipeline.

Then you have DLSS which is literally a meme feature on top of meme tensor cores. The new architecture is like an eternal beta test platform. None of the new features are standalone pipeline compute regions. They're like teaser compute units.. Just enough to call it a new feature while not achieving much. The real question is what they did in terms of the SMs to speed performance. That's all that really matters. In this, there has always been an opening.

7nm for both Nvidia and AMD is where the real comparison starts. It's a point in which they both will have had enough time to start locking in the makings of a new and bold platform outside of meme beta feature sets.

Ray tracing belong on a completely separate Die imo w/a high level, low latency copy/coherent BVH tying it to the main GPU pipeline. If you had any awareness of what it takes computationally and memory wise to do more proper ray tracing, you'd understand any reference to it on a monolithic die is a meme.

As such, everyone is on the same footing.. Gimmick emulated speed up features. Tensor cores are memes for AI and Ray tracing cores are memes for Ray tracing. Everyone is going to have to go MCM to be competitive and I don't see this in GPUs yet. This and Intel is coming into the game hard mode and many others are making dedicated accelerators just as I indicated. The future is anyone's. Were in a big hardware boom/wave. I'm not betting on any particular horse nor am I becoming indebted to someone's dev board level premature architecture.

I don't know what you think you're accomplishing by forcing the word "meme" into every post 50 times but it's not working and makes you sound petty and immature