The future of AMD in graphics

[krumme]

As i recall its 1.5 year since Suzanne Plummer entered the VP role for radeon. Its 1 year back since they moved some key personel from cpu to gpu team. Gpu got new management 1 year back.
I think Suzanne is a key person to enabling a strong organization's that is the fundamental of developing a competitively gpu.
We might get lucky and get a few results of that at next gen after navi but otherwise I would guess it's still several years out.
I think and gpu devision is more than lucky now with the financial results considering the efficiency of their arch. Nv 12nm gpu is a failure and it hides the huge problems at the amd gpu group.
I am confident they will get there in time with Lisa in front and Suzanne as driver. If not in 2 years from now then in 4 years.

 

[ozzy702]

I said last quarter, you can't take the overall market that has been buying stuff over the past 10 years. Marketshare over the last quarter and last few years has been improving SIGNIFICANTLY for AMD.

No, no it hasn't, and claiming it has doesn't make it so. A percent or two gain in marketshare due to the mining boom doesn't represent "significant" improvement for AMD. I really do hope that AMD gets their act together on the GPU side of things like they have with their CPUs, but they're still likely many many years out from competing with NVIDIA in a way that increases marketshare. I fully expect to see 3000 series NVIDIA cards bring NVIDIA back with a vengeance due to the poor reception/sales of the 2000 series. If AMD hopes to compete they need a completely new architecture and direction on their gaming GPUs.

 

[Yotsugi]

I really do hope that AMD gets their act together on the GPU side of things like they have with their CPUs

They don't need to.

but they're still likely many many years out from competing with NVIDIA

This feels like it's post-R600 all over again.
Someone's gotta eat his words, and that's definitely not me.

 

[ozzy702]

They don't need to.
This feels like it's post-R600 all over again.
Someone's gotta eat his words, and that's definitely not me.

Nope. It will take many many years to turn around AMD's GPU division and yes, at that point, maybe, just maybe they can pull another 4870 miracle out of their hat, but things are different now, and the low hanging fruit is gone. It's much more difficult now to realize solid gains and make up lost ground. If AMD can pull off multi-gpu with a new architecture and catch up on the software side of things, of course they could compete, but again, we're likely many many years out from seeing that realized.

I suspect like usual, the AMD fanbois will be disappointed in NAVI because they've set their expectations unreasonably high, just like they always do. Give it some time, maybe 2021/2022 and I think AMD will strike back with some solid, compelling products, I expect very little before then.

 

[Yotsugi]

It will take many many years to turn around AMD's GPU division and yes, at that point, maybe

Like a few months left?

4870 miracle out of their hat

4870 wasn't special.
It just wasn't a broken (or undersized) mess like R600/670 were.

low hanging fruit is gone.

You don't even need brains to spot ALL the low hanging fruit left in Vega.
Even an insect can spot it.

If AMD can pull off multi-gpu with a new architecture and catch up on the software side of things,

MCM GPUs will probably never be functional for compute graphics.

 

[maddie]

Navi is gaming first.

Navi+1 is compute focus again.

Interested in this statement. Speculation or insider knowledge? We've had some posters in the past make claims as fact, when it really was their beliefs being stated. Normally it seems good practice to show the reasoning used to make a claim.

With regards to Vega. I saw a dual use card. For example the DSBR tech is mainly a gaming feature.

 

[Yotsugi]

Speculation or insider knowledge

Teeny bit of both.

With regards to Vega. I saw a dual use card. For example the DSBR tech is mainly a gaming feature.

It was a gaming first, too, just not enough changes.
Their FF setup is very much a grandpa in semi timelines, dates back to pretty much R600, they just doubled it twice and iterated on functional blocks.

 

[railven]

Yep, this train is definitely going. Hopefully this insider's crystal balls pans out a little more. Here is hoping for the return of team red! (I just found an old ATI mouse pad in the basement while cleaning! RIP ATI miss you.)

 

[Yotsugi]

Yep, this train is definitely going

It's not, AMD is still pretending Navi doesn't exist.
Which is good, everyone was complained about Vega's marketing so hard it was more obnoxious than the actual marketing.

 

[railven]

It's not, AMD is still pretending Navi doesn't exist.
Which is good, everyone was complained about Vega's marketing so hard it was more obnoxious than the actual marketing.

S'all good. As maddie said, just feels like someone is laying the tracks and starting to sell tickets. Perhaps your 4th post with odd wording. Right/Wrong hope you stick around, so many prophets have come and gone

 

[Yotsugi]

As maddie said, just feels like someone is laying the tracks and starting to sell tickets

There's still a lot that could wrong there, from slipping Si bugs to marketing messes.
The best way is to expect nothing, at least up until the actual uArch disclosure.

Right/Wrong hope you stick around, so many prophets have come and gone

I'm always on Twitter anyway.

 

[DrMrLordX]

Navi is gaming first.

Hey whatever you say buddy. To me, Navi has PS5/future Xbox written all over it. AMD will just extend/rework for desktop to keep up market presence. They know where their base is right now, and they're going to play to that first.

 

[Yotsugi]

To me, Navi has PS5/future Xbox written all over it

Nah.

AMD will just extend/rework for desktop to keep up market presence

The opposite.

They know where their base is right now, and they're going to play to that first.

Consoles are theirs either way.

 

[DrMrLordX]

Like I said, whatever you say pal.

In other news, apparently there's a device known as 66AF:F1 reported to GFXBench. Here's just one comparison:

https://gfxbench.com/compare.jsp?be...GPU&hwname1=AMD+Radeon+RX+Vega&D2=AMD+66AF:F1

It's speculated that 66AF:F1 may be Navi. Some of those scores are weird though. The RX Vega 64 numbers in that database seem kinda legit (if a bit on the low side for the offscreen renders and too high for the onscreen), but those 66AF:F1 numbers are all over the place.

 

[Yotsugi]

It's speculated that 66AF:F1 may be Navi.

That's a Vega20 ID, but it should be Navi, see the CU count.
https://compubench.com/device.jsp?benchmark=compu15d&os=Windows&api=cl&D=AMD+66AF:F1&testgroup=info

 

[Muhammed]

AMD beat Nvidia in the professional market last quarter.

hogwash, NVIDIA's data center revenue is several times bigger than AMD's RTG revenue all together.

NVIDIA reported $679 million for their Q4 data center alone, that's 2/3 of AMD's entire CPU + Graphics division revenue, which garnered $986M .

https://www.anandtech.com/show/13965/nvidia-earnings-report-q4-2019-crypto-pain-but-full-year-gain
https://www.anandtech.com/show/13917/amd-earnings-report-q4-fy-2018

 

[ThatBuzzkiller]

AMD need not worry since their graphics division will remain independent in the future while their unsuspecting competitor is at the mercy of the x86 architecture and they'll very quickly find themselves locked out of the race between AMD or Intel for sometime in the coming decade ...

Another growth potential is supplying cloud gaming hardware and having an x86 license is invaluable in that case ...

 

[Yotsugi]

AMD need not worry

They do, for Intel is big, angry and has x86 stick too.

 

[ThatBuzzkiller]

They do, for Intel is big, angry and has x86 stick too.

Which is why it'll ultimately be a race between AMD or Intel and anyone else without an x86 license are going to end up as a casualty ...

The biggest graphics chip designer will have no choice but to depend on a more desperate x86 chip designer that's clearly backed by a foreign government agenda such as Zhaoxin and in exchange for doing business will be some IP transfers I can imagine ...

Intel entering the high end graphics market could be the camel that breaks everyone's backs. The only way I can see them gaining traction in a meaningful capacity is to lock-in their customers to their graphics solution and that'll play an important role to their ambitions for compute as well. By forcing their customers to use their graphics solutions, they can also force them to use their compute solutions as well like their oneAPI project ...

For quite a while, AMD will be fine since they have an x86 license, long-term semi-custom contracts, potential for supplying the most cloud gaming hardware and they're ahead of Intel with their compute API ...

 

[Ajay]

I said last quarter, you can't take the overall market that has been buying stuff over the past 10 years. Marketshare over the last quarter and last few years has been improving SIGNIFICANTLY for AMD.

Got a link?

 

[Ajay]

MCM GPUs will probably never be functional for compute graphics.

Compute is the easiest to distribute, as it’s already being done in HPC servers across multiple GPUs. MCM GPUs May be the only way to scale out compute once large GPU designs become too expensive to be profitable.

 

[Yotsugi]

Compute is the easiest to distribute, as it’s already being done in HPC servers across multiple GPUs. MCM GPUs May be the only way to scale out compute once large GPU designs become too expensive to be profitable.

Tru, but we're >impying here applying MCM concept to graphics.
Which is beyond tricky.

 

[Dribble]

Which is why it'll ultimately be a race between AMD or Intel and anyone else without an x86 license are going to end up as a casualty ...

The biggest graphics chip designer will have no choice but to depend on a more desperate x86 chip designer that's clearly backed by a foreign government agenda such as Zhaoxin and in exchange for doing business will be some IP transfers I can imagine ...

Intel entering the high end graphics market could be the camel that breaks everyone's backs. The only way I can see them gaining traction in a meaningful capacity is to lock-in their customers to their graphics solution and that'll play an important role to their ambitions for compute as well. By forcing their customers to use their graphics solutions, they can also force them to use their compute solutions as well like their oneAPI project ...

For quite a while, AMD will be fine since they have an x86 license, long-term semi-custom contracts, potential for supplying the most cloud gaming hardware and they're ahead of Intel with their compute API ...

Sounds like you are recycling Larrabee forum posts.

 

[ThatBuzzkiller]

Sounds like you are recycling Larrabee forum posts.

I never foresaw Larrabee coming to fruition in full capacity anyway but I do foresee Intel trying lock-in their customers as much as possible ...

Anyone without an x86 license can go pout elsewhere and design their own x86 cores instead leaching off of either AMD or Intel ...

Intel are not pursuing dedicated graphics just to lose it. They will try and pull all the stops as they can such as taking the performance crown by introducing a big die and excluding the RT and tensor cores to win in benchmarks. If Apple are going to Apple then Intel are probably going to Intel as well since neither of them are compelled to make their platforms compatible with 3rd party devices ...

 

[Topweasel]

I never foresaw Larrabee coming to fruition in full capacity anyway but I do foresee Intel trying lock-in their customers as much as possible ...

Anyone without an x86 license can go pout elsewhere and design their own x86 cores instead leaching off of either AMD or Intel ...

Intel are not pursuing dedicated graphics just to lose it. They will try and pull all the stops as they can such as taking the performance crown by introducing a big die and excluding the RT and tensor cores to win in benchmarks. If Apple are going to Apple then Intel are probably going to Intel as well since neither of them are compelled to make their platforms compatible with 3rd party devices ...

Intel is not getting into the discrete graphics business to win gaming benchmarks. Intel wants to own up and down the whole server infrastructure. The compute Graphics business has quickly outpaced everything they have been trying to do since Larrabee for multi-small core compute. They don't want to be pushed out. That's why they are doing it.