Does the RTX series create an openning for AMD?

[Hitman928]

One just have to wonder how AMD is run, I mean how on Earth is this acceptable at any level? AMD still doesn't understand that no one will integrate / code their solutions to any software, they have to devote the resources to make things happen.

AMD knows this, the problem is they just don't have the resources. AMD was a breath away from death just a couple years ago. AMD put the last few eggs it had in the Zen basket and luckily it worked. If Zen hadn't come out as a success, AMD as we know it would have died, most likely sold off in pieces. With the success of the Zen architecture, AMD finally is getting back on its feet but it will still be a while before we see them really competing at every level, and that's if things continue to go well for them on the CPU side.

So for now, they are trying to compete only where they can get the biggest reward for the smallest investment/risk. On the discrete GPU side that means AI and pure compute. The console deals aren't huge rewards but carry very little risk, so we'll continue to see AMD go after those deals as well. Discrete GPUs for gaming will continue to be essentially an add-on from those two markets until AMD can afford to invest more resources into it. I'm not sure where rendering falls for AMD as it can give you higher margins, but also requires a lot of support.

 

[Despoiler]

Software has always been AMD's weakness. I think the problem is AMD comes from the x86 cpu background where basically you make some hardware that matches the spec and then MS, Intel and everyone else will write all your software. When AMD bought Ati they never really considered the importance of software for gpu's - it's just drivers. Mantle is an attempt to push the burden of driver writing off AMD and onto the devs, so they are still continuing with that path - release the best hardware you can and get everyone else to write the software. That's also why they do well in consoles - MS and Sony want to write all the software, they just want AMD to provide the hardware.

Anyway anything that requires AMD to write software themselves beyond a basic driver goes bad pretty quickly. The normal response is some power point slides, some talk of open source, and then nothing. So for professional workstations, gpu compute, gpu farms, linux, etc - they nearly all use Nvidia due to better software support.

It's also why Nvidia is so keen on stuff like gamesworks - they recognise AMD's weakness and prey on it by writing lots of pretty locked in custom software and then challenging AMD to make their own version knowing they mostly can't.

This is laughable. AMD still has to write the drivers. No devs are writing drivers. Mantle was a successful attempt to get MS to upgrade DX to match the reality of modern CPUs and GPUs. That is to say they are multi-threaded. DX11 at it's core was still single threaded just like all of the versions before it. Nvidia's spent 3 years building a mini multithreaded OS inside their drivers to get better threading out of DX11. That's not how APIs nor drivers are supposed to function. Drivers also should not have to fix the mistakes of bad programming. That's not its role.
https://www.gamedev.net/forums/topi...x12vulkanmantle/?tab=comments#comment-5215019

Anyway anything that requires AMD to write software themselves beyond a basic driver goes bad pretty quickly. The normal response is some power point slides, some talk of open source, and then nothing. So for professional workstations, gpu compute, gpu farms, linux, etc - they nearly all use Nvidia due to better software support.

This is dated ignorance like "AMD drivers suck". Let me bring you into the present.
https://gpuopen.com/games-cgi/
https://gpuopen.com/professional-compute/
https://rocm-documentation.readthedocs.io/en/latest

It's also why Nvidia is so keen on stuff like gamesworks - they recognise AMD's weakness and prey on it by writing lots of pretty locked in custom software and then challenging AMD to make their own version knowing they mostly can't.

Nvidia does proprietary because they are afraid of being shut out of anything that has to do with x86. Intel and AMD could collectively agree to use a bus standard that Nvidia isn't party to and it's bye bye Nvidia. Nvidia's only leverage is to lock in a base that would complain if all of their gear was suddenly obsolete. Nvidia only likes open standards when it suits them.

 

[Muhammed]

Intel and AMD could collectively agree to use a bus standard that Nvidia isn't party to and it's bye bye

Yeah in candy land maybe. They will be slammed a class action lawsuit for extreme anti competitive behaviors and monopoly, exactly like Google and Microsoft. They will be fined a legendary amount of dollars, tens of billions maybe.

Not to mention they will lock out all the peripheral card makers. Will lock out customers who will refuse to upgrade to these new castrated platforms because it gives them less option. Customers already hold on to their old i7 2600K, 3770K and the respective i5s because newer platforms gives them fewer benefits. Imagine what would happen If AMD released a Ryzen that only works on AMD GPUs. Instant dead duck in the water for multiple generations. Same for Intel.

And then Intel wouldn't dare do this alone, because too many people use NVIDIA, Professionals, people in Visualization, Data Centers, Super Computers, AI, game developers .. etc. All of those rely on NVIDIA and they will be potential customers of AMD if Intel decides to do this.

And it wouldn't stop NVIDIA selling a single GPU either, board makers will make custom boards with PCI Express standard and these boards will sell like hot cakes.

NVIDIA also relies on ARM for a lot of their HPC/AI/data center solutions. So they are truly massively less reliant on Intel right now.

NVIDIA can also start a patent war to prevent Intel from making it's GPU.

 

[Kenmitch]

Yeah in candy land maybe. They will be slammed a class action lawsuit for extreme anti competitive behaviors and monopoly, exactly like Google and Microsoft. They will be fined a legendary amount of dollars, tens of billions maybe.

So both AMD and Intel are doomed forever to innovate in a way that provides Nvidia a way to install a gpu? Are you drunk on Koolaid?

Why would the courts limit a companies rights to innovate?

 

[Muhammed]

So both AMD and Intel are doomed forever to innovate in a way that provides Nvidia a way to install a gpu? Are you drunk on Koolaid?

They are free to innovate, but they can't lock NVIDIA out of the platform without legal consequences. They can come up with a new standard other than PCI Express as much as they want. But they HAVE to make that standard open to everyone. PCI Express was created by Intel and IBM actually. And they gave it to everyone.

Microsoft got slammed hard for providing a IE as a default browser in it's Windows OS. Microsoft argued that it's decision was fair and innovative, but the court fined them over it and forced them to create openings for other browsers as well like FireFox and Chrome. Google suffered a similar fate on the Android OS.

So you see? only in candy land can a company lock another company out of it's platform. Even if it's a software one.

 

[ub4ty]

Intel and AMD could collectively agree to use a bus standard that Nvidia isn't party to and it's bye bye

Yeah in candy land maybe. ..

AMD doesn't need Intel for this ... Nor did Nvidia. The both have been going outside and around limitations with their own high speed coherent bus links. Nvidia has Nvlink :
https://www.nvidia.com/en-us/data-center/nvlink/

AMD has XGMI :
https://segmentnext.com/2018/09/07/amd-radeon-vega-20-xgmi/
What did you think Infinity Fabric was?

None of this stuff is magic or complicated :
https://en.wikipedia.org/wiki/SerDes

See slide 5 :
https://www.slideshare.net/siliconl...s-performance-for-networking-and-data-centers
Nvlink
https://en.wikichip.org/wiki/nvidia/nvlink

Everyone uses it in varying capacity. It's a generic bus mechanism you can layer in your own proprietary protocol to.
AMD can implement the equivalent of Nvlink for GPU to GPU linkage and can even layer it into their CPUs to make the sell even more attractive for the whole AMD platform.

You can only play games holding back technology for so long. AMD was in a tight position and almost about to go under and came out swinging with everything at amazing prices. There are tons of historically 'pro only' features that they put in vega. This trend will continue as it did in their CPU line for their GPU line until they get market share from Nvidia. If Nvidia continues their antics into 2019, they're screwed just like Intel is. Just like I was with Intel, I am fed up w/ Nvidia and can't wait for a competing product. Many were on the fence but this launch solidified that viewpoint. Come 2019, its' going to be ryzen all over again except in the GPU market. Like intel, nvidia is probably going to maintain their arrogant stance and pricing and they're going to pay dearly for it.

Most people aren't participating in this tie-over launch. They're waiting until 2019 when the real products are delivered.
Nvidia went for a crash grab and now has put the favor in AMD's corner just like Intel did.
Had Geforce 20 been priced more sanely Nvidia could have crushed AMD for some time. They didn't and it was a big mistake. Games by and large dont support this yet and just about everyone in the compute/rendering field is scrutinizing the hell out of what these cards can actually do and wont support it as well until 2019. One more week and this will be fully fleshed out. What a ride and product launch... I haven't seen this much discontent over a product in a while.

RAM prices are also still stupid high as well as other components. Mid/early 2017 were a rare and amazing period for all component pricing. 2019 is shaping up to be another. A really crappy time to do a build currently. Prices suck. And all of the major changes are coming next year.

 

[Thala]

Most people aren't participating in this tie-over launch. They're waiting until 2019 when the real products are delivered.
.

Even if this assumption holds, these very people are waiting for the next nVidia product. And the hope that the next generation will be significantly cheaper is a bit far fetched.

 

[AtenRa]

There is no opening space for AMD in 2019 unless they will release 7nm Consumer cards earlier than NV.
NV will continue to sell Pascal cards along RTX cards as long as they dont have an alternative.

 

[Despoiler]

So you see? only in candy land can a company lock another company out of it's platform. Even if it's a software one.

Things like CUDA, GSync, and Gameworks will surely work out for Nvidia then. That conclusion is being realized by AMD which they are exploiting by using open standards. Dude thank you for proving my point.

 

[Muhammed]

Things like CUDA, GSync, and Gameworks will surely work out for Nvidia then. That conclusion is being realized by AMD which they are exploiting by using open standards. Dude thank you for proving my point.

NONE of these things lock out whole products. They only lock certain aspects of products. And All of them were created as closed systems specific to certain hardware. Open systems such as an OS and a motherboard/platform are not the in the same category.

Read the law please before spewing nonsense.

 

[Kenmitch]

They are free to innovate, but they can't lock NVIDIA out of the platform without legal consequences. They can come up with a new standard other than PCI Express as much as they want. But they HAVE to make that standard open to everyone

Intels Kaby Lake G platform is a good example. What if AMD and Intel both decided to make all future offerings similar to this design. The courts would order them to incorporate a Nvidia gpu into the package? The courts would order them to include a full length PCI express slot that wasn't even needed? If AMD designed a new proprietary gpu bus for their add on offerings and Intel did the same the courts would order them to give away the IP for free? I don't think it really works that way. It's not like they'd be the artificial lockouts that you used as examples.

I don't think Nvidia has PCI protection anymore....At least that's the way it looks too me.

• maintain a key interface, known as the PCI Express Bus, for at least six years in a way that will not limit the performance of graphics processing chips. These assurances will provide incentives to manufacturers of complementary, and potentially competitive, products to Intel’s CPUs to continue to innovate

https://www.ftc.gov/news-events/pre...charges-anticompetitive-conduct-against-intel

End of off topic postings for me....Sorry about that guys.

 

[Muhammed]

Intels Kaby Lake G platform is a good example. What if AMD and Intel both decided to make all future offerings similar to this design.

They can't. They will have to offer standalone CPUs. For which they are mandated to provide a PCI Express solt.

If AMD designed a new proprietary gpu bus for their add on offerings and Intel did the same the courts would order them to give away the IP for free?

The court would force them to allow NVIDIA to use that same proprietary bus for a fee/or freely. If they didn't the court will fine them a heft amount.

I don't think Nvidia has PCI protection anymore....At least that's the way it looks too me.

NVIDIA doesn't need that protection though. NVIDIA can ruin Intel's plan of producing a dGPU, they already control the majority of GPU patents. They can do the same for AMD as well.

 

[crisium]

Those last few sentences are nonsense with no regard to reality. It's like claiming AMD can wave their hands and stop licensing x86-64 to Intel leading to the end of all Intel CPUs.

 

[Kenmitch]

NVIDIA doesn't need that protection though. NVIDIA can ruin Intel's plan of producing a dGPU, they already control the majority of GPU patents. They can do the same for AMD as well.

That would be a really good example of anti-competitive behavior and would most likely end not so good for Nvidia. Nice example!

 

[Muhammed]

That would be a really good example of anti-competitive behavior and would most likely end not so good for Nvidia. Nice example!

It's not. NVIDIA already went to court over the exact same thing, remember? Intel settled with them for 1.5 billion in exchange for NVIDIA not getting an X86 license or even emulating one.

Besides NVIDIA will only do so if Intel moved against them by locking them out. If Intel didn't lock them, NVIDIA won't sue them.

 

[PeterScott]

Those last few sentences are nonsense with no regard to reality. It's like claiming AMD can wave their hands and stop licensing x86-64 to Intel leading to the end of all Intel CPUs.

So is the nonsense about AMD and Intel making a proprietary bus to lock NVidia out of the dGPU business.

Everyone should get off the fairy tale track in this discussion.

It isn't about nonsensical anti-competitive moves that are potentially antitrust violations.

It's about whether the "RT Tax" on die space and thus price, creates an opportunity for AMD to release card (without RT Tax) with better price/performance than NVidia to get back market share.

Also, if that opportunity exists, is AMD well positioned enough to exploit it?

 

[Ottonomous]

Any way for AMD to sneak an open-source raytracing implementation into the upcoming consoles? Begin budget friendly ray tracing right from the bottom even if its uncompetitive

 

[Kenmitch]

It's about whether the "RT Tax" on die space and thus price, creates an opportunity for AMD to release card (without RT Tax) with better price/performance than NVidia to get back market share.

Also, if that opportunity exists, is AMD well positioned enough to exploit it?

I think the AMD response will only be what they intended to do in the 1st place. I doubt they'll deviate from their current plans much if at all.

 

[ub4ty]

Even if this assumption holds, these very people are waiting for the next nVidia product. And the hope that the next generation will be significantly cheaper is a bit far fetched.

Fine by me. Someone has to pay the early adopter tax. No way in heck I'm doing that at $1,200, $800, or even $600 (no nvlink 2070). And if Nvidia doesn't drop these stupid prices by at least $200 per card at a minimum in 2019, the answer will be heck no again. With the progress AMD is making on OpenCL/Vulkan/ and RocM, I am already salivating. I bought a Vega at launch already but sold it off because the driver/dev support wasn't there. I was already looking for an alternative. If a fanboy wants to stay loyal to Nvidia that's fine with me. I have no such loyalty. I go where the value is. I used to build completely Intel and switched to Ryzen when it was launched. I have been all Nvidia for some time and will switch in a heart beat to Radeon if the value is found there.

 

[ub4ty]

There is no opening space for AMD in 2019 unless they will release 7nm Consumer cards earlier than NV.
NV will continue to sell Pascal cards along RTX cards as long as they dont have an alternative.

7nm Pro GPUs are being launched by AMD before the close of 2018.
7nm consumer GPUs are most definitely being launched by AMD in 2019.
Nvidia can do whatever they like, they are dead to me at these prices. I already have 6 1080 Pascal cards.

 

[AtenRa]

7nm Pro GPUs are being launched by AMD before the close of 2018.
7nm consumer GPUs are most definitely being launched by AMD in 2019.
Nvidia can do whatever they like, they are dead to me at these prices. I already have 6 1080 Pascal cards.

The problem with 7nm in 2018 and 2019 will be the available Capacity and wafer prices.
Because now we know that AMD is using TSMC 7nm solo for not one but two products (GPUs and CPUs), they may not have the capacity they need to release Professional and Consumer products for both CPUs and GPUs.

And because im certain that AMD will focus heavily in CPUs early in 7nm era for both Server and Consumer, there is a possibility we may not see any 7nm consumer GPUs in 2019 at all.

 

[ub4ty]

The problem with 7nm in 2018 and 2019 will be the available Capacity and wafer prices.
Because now we know that AMD is using TSMC 7nm solo for not one but two products (GPUs and CPUs), they may not have the capacity they need to release Professional and Consumer products for both CPUs and GPUs.

And because im certain that AMD will focus heavily in CPUs early in 7nm era for both Server and Consumer, there is a possibility we may not see any 7nm consumer GPUs in 2019 at all.

It is likely were going to see consumer 7nm GPUs in 2019.
AMD is launching Professional 7nm GPUs before the end of this year.
7nm CPUs throughout 2019 starting in Jan for Pro segment.
You forget that consumer vega is literally pro vega with some features disabled.
AMD doesn't have the luxury of pulling an artificial segmentation Intel game like Nvidia w/ their GPU dies. If pro vega 7nm is launching before the end of the year, consumer gpus are close behind.

Consumer products would come in 2H 2019. If it doesn't, it's not like I'm in desperate need of a GPU. I am stocked up with 1080 Pascal. I game on a 4 year old GPU that I have zero issues with. If it dies, I'll convert one of my 1070s to my gaming GPU. I ride gaming GPUs for 5 years. I bought 1070s about 2 years ago. So, I'm set until 2021.

I only buy when I get quite excited about the performance/value. Pascal was an exciting release. Geforce20 is a joke. I bought vega but the driver/compute dev stack wasn't ready and sold it off.

As a consumer, you have to 'sell' me on a product. I'm not going to buy your product just because you're the only game in town and released something new. I'm waiting until 7nm for further CPU/GPU purchases. Nvidia could have drawn me in to buying a 12nm card but you can forget about that at these prices. So, AMD has leverage IMO. Nvidia gave them an opening. AMD delivered HBM 2.0 in consumer cards 2 years ago and its nowhere on the radar for Nvidia. I'm pretty sure AMD is going to beat them to 7nm as well. With the way the drivers/dev stack is progressing at AMD, my next GPU purchases will be with AMD.

 

[Despoiler]

Rosenblatt upgraded AMD price target from $30 to $40 based on AMD's node lead on Nvidia on the business side. Consumer will be the same as I said. Nvidia needs ~1.5 years to recoup to move to a new gen.

https://www.cnbc.com/2018/09/11/amd-can-compete-against-nvidia-in-ai-analyst-says.html

 

[Muhammed]

Nvidia needs ~1.5 years to recoup to move to a new gen.

Yeah, in candy land maybe .. the land of overzealous "analysts and their preposterous expectations.

 

[ub4ty]

Yeah, in candy land maybe .. the land of overzealous "analysts and their preposterous expectations.

Your bias is showing. Candyland was when AMD was at $2 share.
Their story has already been written now. Incredibly, companies like Intel and now Nvidia are just giving their customers to AMD. Greed knows no bounds and is completely blind it seems. The same thing that's happening to intel is going to happen to Nvidia because they're making the same exact stupid mistakes down to these idiotic and monstrous monolithic dies.