AMD reports best quarterly profit in 7 years thanks to new products

[Doobeedoo]

Nice. Yeah some time was needed to past to reap the profit but great. They need it. 

[Taf the Ghost]

1 hour ago, cj09beira said:

its clear that amd will only be back to the high end with the gext gen architecture (after navi) because what is needed to be done to fix gcn would mean that the biggest and basically only gcn flaw would disappear (talking about inherent flaws here ) and moving away from gcn right after fixing it would be a waste of resources 

AMD needed a brand new GPU Architecture to land around 2019. The mistakes Raja was pointing to was the choice to iterate upon GCN longer than viable.

1 hour ago, Humbug said:

I agree that Navi will not challenge Nvidia's top end (1180ti or Titan) parts, because the rumours we are now hearing are that AMD still won't have MCM tech good enough to be able to make a gamer product.

 

Still 7nm Navi thanks to the new process is going to clock way higher using less power than current gen Vega. So that fact combined with the architectural improvements at least should be enough to compete in the GTX 1170 and 1180 range.

Gaming GPUs are not going MCM with Navi, as the new head of AMD's GPU division said as much. That wasn't anything unexpected, as GCN just can't do it. You'd need a completely redesigned GPU architecture from the front-end to pull it off. However, they *may* do a MCM compute card. I actually do expect we'll see that in 2020. It'll be with Vega on 7nm, more than likely, as the design has moved to IF-based control schemes. 

 

As for Navi, I'm expecting some Vega + new tech on GDDR6. HBM makes so much sense for AMD, but the price is such a big problem they can't really produce many products with it at Mainstream costs. It's basically the Bulldozer problem again, as the production cost vs market-clearing price is too close. So AMD will stay in the Mainstream GPU department, still be 33% of the market or so, while actually still selling far more GPUs than Nvidia does.

 

We'll probably get something like a 42 CU design with GDDR6. That should do most of the work to keep AMD right in the mix of things, up to being against the 1170.

[Humbug]

1 hour ago, Taf the Ghost said:

As for Navi, I'm expecting some Vega + new tech on GDDR6. HBM makes so much sense for AMD, but the price is such a big problem they can't really produce many products with it at Mainstream costs.

The 7nm process alone ensures that Navi will clock better with less power consumption than Vega.

On the architectural side I wonder how much different it will be to Vega. The rumours we hear about AMD putting a lot more manpower into Navi suggest a lot of changes, but we haven't heard anything official.

 

Who knows maybe by 2019 HBM will actually make sense, difficult to predict.

[AngryBeaver]

Not hard to post best Quarterly profit in 7 years when your company hasn't been competitive at all for those 7 years.

[GoldenLag]

1 hour ago, AngryBeaver said:

Not hard to post best Quarterly profit in 7 years when your company hasn't been competitive at all for those 7 years.

The question is how many "best quartely profit"s they can manage to do in a row

[Taf the Ghost]

18 minutes ago, Humbug said:

The 7nm process alone ensures that Navi will clock better with less power consumption than Vega.

On the architectural side I wonder how much different it will be to Vega. The rumours we hear about AMD putting a lot more manpower into Navi suggest a lot of changes, but we haven't heard anything official.

 

Who knows maybe by 2019 HBM will actually make sense, difficult to predict.

Vega has a pretty sizable Perf/Watt and Watt/Clock advantage over Polaris. The issue is just that, for gaming, the die just wastes a lot of resources under full load. And this is on the exact same Node as Polaris.

[Mihle]

3 hours ago, Taf the Ghost said:

AMD needed a brand new GPU Architecture to land around 2019. The mistakes Raja was pointing to was the choice to iterate upon GCN longer than viable.

Gaming GPUs are not going MCM with Navi, as the new head of AMD's GPU division said as much. That wasn't anything unexpected, as GCN just can't do it. You'd need a completely redesigned GPU architecture from the front-end to pull it off. However, they *may* do a MCM compute card. I actually do expect we'll see that in 2020. It'll be with Vega on 7nm, more than likely, as the design has moved to IF-based control schemes. 

 

As for Navi, I'm expecting some Vega + new tech on GDDR6. HBM makes so much sense for AMD, but the price is such a big problem they can't really produce many products with it at Mainstream costs. It's basically the Bulldozer problem again, as the production cost vs market-clearing price is too close. So AMD will stay in the Mainstream GPU department, still be 33% of the market or so, while actually still selling far more GPUs than Nvidia does.

 

We'll probably get something like a 42 CU design with GDDR6. That should do most of the work to keep AMD right in the mix of things, up to being against the 1170.

Vega was designed for compute as number 1. Polaris, and what I have heard Navi, is not.

[Taf the Ghost]

14 minutes ago, Mihle said:

Vega was designed for compute as number 1. Polaris, and what I have heard Navi, is not.

Correct. Navi is the Polaris replacement, but there is still core, iterative technology that Navi will incorporate from Vega. Navi is the iterative (and likely final) version of GCN. Hopefully they can save space by cutting down on some of the FP ability. 

[Christophe Corazza]

Profits are on the Ryze!

[yian88]

GO AMD

Its sad that with all the chips sales and console sales, and all those GPU's sold to miners for 3x the price, all they make is ~150m $ income, thats bullshit.

 

I dont understand why they have such a small income when they have so many products out.

If i were Dr.Lisa Su i would seriously invest in server GPU's like nvidia and make software for it, before intel starts releasing their dGPU's for desktop/server.

 

[XenosTech]

3 minutes ago, yian88 said:

GO AMD

Its sad that with all the chips sales and console sales, and all those GPU's sold to miners for 3x the price, all they make is ~150m $ income, thats bullshit.

 

I dont understand why they have such a small income when they have so many products out.

If i were Dr.Lisa Su i would seriously invest in server GPU's like nvidia and make software for it, before intel starts releasing their dGPU's for desktop/server.

 

That's what vega is for ya know. Vega has a enterprise version called MI25 I think it is

[cj09beira]

6 minutes ago, yian88 said:

GO AMD

Its sad that with all the chips sales and console sales, and all those GPU's sold to miners for 3x the price, all they make is ~150m $ income, thats bullshit.

 

I dont understand why they have such a small income when they have so many products out.

If i were Dr.Lisa Su i would seriously invest in server GPU's like nvidia and make software for it, before intel starts releasing their dGPU's for desktop/server.

 

amd has had better gpus for the server market for years but cuda is being a very hard wall to jump across, they are now trying with rocm but it will take a long time before people stop using cuda, and thats assuming they will taka the opportunity to unglue themselves from nvidea

[Drak3]

6 hours ago, Christophe Corazza said:

Profits are on the Ryze!

AMD's profits are Ryzen.

[cj09beira]

2 minutes ago, Drak3 said:

AMD's profits are Ryzen.

hey are Epyc  

[ScratchCat]

21 hours ago, Dan Castellaneta said:

 

"Dad, keyboards have keys, our grass has keys, is our grass a keyboard?"

13 hours ago, D13H4RD2L1V3 said:

Precisely that. 

 

I don't really want to see AMD dominate Intel and NVIDIA in a short span of time. That's not going to happen anytime soon. 

 

What I really want to see is them holding their ground and releasing products that prove to be great alternatives and even best the status quo at times. We need competition 

If they provide great alternatives they almost certainly will dominate over NVIDIA and Intel unless they pull anti competitive tactics.

 

Companies feel forced to innovate if they have competition. If AMD rapidly increases it's market share Intel and NVIDIA will feel threatened by the competition and try regain an advantage. We have seen how Ryzen annihilated Intels plan of 4 core mainstream chips and the rapid growth we have seen since then (i7 7700k ~ i3 8xxx K SKU with OC) while we haven't seen a new generation from NVIDIA in two years.

[Stefan Payne]

10 hours ago, Taf the Ghost said:

AMD needed a brand new GPU Architecture to land around 2019. The mistakes Raja was pointing to was the choice to iterate upon GCN longer than viable.

GCN is fine, that's why nVidia is Adopting it.

The Problem is the Frontend/Scheduling and especially scaling.

The Limit of GCN are 4k Shaders and it doesn't scale that well from 2k to 4k Shaders at all.

 

10 hours ago, Taf the Ghost said:

Gaming GPUs are not going MCM with Navi, as the new head of AMD's GPU division said as much. That wasn't anything unexpected, as GCN just can't do it. You'd need a completely redesigned GPU architecture from the front-end to pull it off. However, they *may* do a MCM compute card. I actually do expect we'll see that in 2020. It'll be with Vega on 7nm, more than likely, as the design has moved to IF-based control schemes. 

MCM is just an insane idea because of the bandwith needs of modern GPUs, that would just be too expensive in terms of power consumption and complexety that it doesn't make any sense. And you need a whole bunch of Bandwith for inter chip communication to let it act as a single GPU.

 

I think what he said might either be misspoken or misunderstood. 

And what he really meant was what AMD is doing right now -> Multi Cluster Stuff but on one Chip.

 

10 hours ago, Taf the Ghost said:

As for Navi, I'm expecting some Vega + new tech on GDDR6.

Isn't there talk about a new CU Design and AMD going a step back in the direction of the Terrascale Architecture??

I've heard them saying something about Big and Little CUs for whatever reason. 

It could also make sense to have a general scheduler and another in cluster. Because that could improve scalability.

 

10 hours ago, Taf the Ghost said:

HBM makes so much sense for AMD, but the price is such a big problem they can't really produce many products with it at Mainstream costs. 

HBM makes sense for everything as it lowers the power consumption dramatically because the power that's needed to get the signal half way across he world while with HBM it doesn't need to get off package, so that needs less driver power...

 

As for the Prie: THe Problem is putting everything together, the Cost of the Memory isn't the Problem.

But if you mess up one part in putting 2/4 HBM Stacks and one big die on the Interposer, you can throw away everything. 3 Good Memory Stacks and one 484mm² (or so) Die.

 

10 hours ago, Taf the Ghost said:

It's basically the Bulldozer problem again, as the production cost vs market-clearing price is too close.

Well, kinda, absoutely.

Because the Bulldozer Problem was starved excecution units because the Cache was shit and Northbridge clocked lower than most Athlon 64...

 

Well, the Schedular was also one of the Problems Bulldozer had at the time because with later Iterations they went for one Scheuduler per Int units and not a shared on.

 

10 hours ago, Taf the Ghost said:

So AMD will stay in the Mainstream GPU department, still be 33% of the market or so, while actually still selling far more GPUs than Nvidia does.

Lets wait and see what they do and how good Navi will scale.

What will be certain is that Sony will have much influence on the architecture.

 

10 hours ago, Taf the Ghost said:

We'll probably get something like a 42 CU design with GDDR6. That should do most of the work to keep AMD right in the mix of things, up to being against the 1170.

Well, lets wait and see how that will go...

But 42CU aren't much better than the 36 the Polaris 10 have right now.

[BuckGup]

23 hours ago, AnonymousGuy said:

Before anyone gets too excited: Intel makes > $4 billion in net income per quarter.  More than 3x what AMD makes as revenue.

 

AMD is going under / going to be acquired,  it's just a matter of when.

Except if AMD beats intel in the 7nm race then Intel is going to be hurting. Intel is already hurting in the server space

[GOTSpectrum]

23 hours ago, GoldenLag said:

The only ones that can aquire AMD without to much trouble is Intel and VIA. But if intel aquired AMD then they would be charged with monopoly charges. Dont believe VIA has the funds to buy AMD and any other would invalidate the X86-64 cross liscencing making a patent mess of current desktop processors.

 

I dont have the details, but aquiring someone who is liscensing X86 is very difficult

Don't IBM still have an x86 license too technically?

[Stefan Payne]

3 minutes ago, ScratchCat said:

If they provide great alternatives they almost certainly will dominate over NVIDIA and Intel unless they pull anti competitive tactics.

No, because many people won't buy AMD, even if they are the better alternatives.

Just look at RX470, 480, 570, 580 or Hawaii and especially Tahiti.

 

Even if they are better (RX5870) than nVidia (RX480) and have lower Power Consumption (~100W Across the Board)

[cj09beira]

3 minutes ago, Stefan Payne said:

GCN is fine, that's why nVidia is Adopting it.

The Problem is the Frontend/Scheduling and especially scaling.

The Limit of GCN are 4k Shaders and it doesn't scale that well from 2k to 4k Shaders at all.

 

MCM is just an insane idea because of the bandwith needs of modern GPUs, that would just be too expensive in terms of power consumption and complexety that it doesn't make any sense. And you need a whole bunch of Bandwith for inter chip communication to let it act as a single GPU.

 

I think what he said might either be misspoken or misunderstood. 

And what he really meant was what AMD is doing right now -> Multi Cluster Stuff but on one Chip.

 

Isn't there talk about a new CU Design and AMD going a step back in the direction of the Terrascale Architecture??

I've heard them saying something about Big and Little CUs for whatever reason. 

It could also make sense to have a general scheduler and another in cluster. Because that could improve scalability.

 

HBM makes sense for everything as it lowers the power consumption dramatically because the power that's needed to get the signal half way across he world while with HBM it doesn't need to get off package, so that needs less driver power...

 

As for the Prie: THe Problem is putting everything together, the Cost of the Memory isn't the Problem.

But if you mess up one part in putting 2/4 HBM Stacks and one big die on the Interposer, you can throw away everything. 3 Good Memory Stacks and one 484mm² (or so) Die.

 

Well, kinda, absoutely.

Because the Bulldozer Problem was starved excecution units because the Cache was shit and Northbridge clocked lower than most Athlon 64...

 

Well, the Schedular was also one of the Problems Bulldozer had at the time because with later Iterations they went for one Scheuduler per Int units and not a shared on.

 

Lets wait and see what they do and how good Navi will scale.

What will be certain is that Sony will have much influence on the architecture.

 

Well, lets wait and see how that will go...

But 42CU aren't much better than the 36 the Polaris 10 have right now.

42 cus if paired with 64 rops would be a very good gpu

[GOTSpectrum]

13 hours ago, Humbug said:

They will come back to the high end in GPUs. But the question is when. Will it be in the next two years, or do we have to wait for 5 year for them to come back... That would suck. This depends on how much they prioritize it compared to other markets.

 

The thing is that Lisa Su knows that the CPU market is much more important. If AMD puts out a bad generation of graphics chips then they go back and try again.. cycle repeats etc. But if Zen cores had sucked AMD would have been a goner, no coming back from that. Lisa Su knows that the CPU and custom market makes AMD more money than selling discrete GPUs to gamers.

 

According to Raja Koduri AMD (prior to RTG) made some bad calls on graphics where they predicted wrong. They thought that discrete GPUs will die, APUs would be the future etc, and since they were making losses at the time they decided no point spending money on R&D. At the time they had great high end GPUs (HD 7970 --> R9 290x etc) but behind the scenes they were not preparing for the future because they though the market will die and they had to prioritize.

 

They were dead wrong and PC gaming went from strength to strength demanding more horsepower than APUs could muster. According to Raja they fixed this mess of priorities and got focused with the formation of RTG, but who knows how long it will take for that to bear fruits in terms of shipping actual products. Most of us expected it to be Navi... but now there are all these stories that Lisa Su took Raja's engineers away from making PC GPUs and is more focused on designing future console chips rather than beating Nvidia on the high end. Apparently Raja was not happy and blamed that for how Vega turned out etc... If that's the case it could be a long wait for us as far as high end is concerned. Although they will continue to be competitive below the top end stuff (e.g. Vega56 or RX580 etc).

The thing is, at MSRP vega 64 is competitive with the 1080. But you just can't get them at MSRP at all. 

[Okjoek]

 

[GoldenLag]

27 minutes ago, Ben Quigley said:

Don't IBM still have an x86 license too technically?

I believe only AMD, Intel and VIA have the x86 liscence, but it is useless in convetional computers without AMD64 instruction set.

[ScratchCat]

22 minutes ago, Stefan Payne said:

No, because many people won't buy AMD, even if they are the better alternatives.

Just look at RX470, 480, 570, 580 or Hawaii and especially Tahiti.

 

Even if they are better (RX5870) than nVidia (RX480) and have lower Power Consumption (~100W Across the Board)

Hawaii was a consumer fusion reactor and pulled over 300W.

The RX series improved on that drastically but still were less efficient and back in 2016 Crimson wasn't particularly stable.

 

The RX X80s never consumed less than 150W (the power limit on the reference design) while the 1060 pulls 120W in the same situation.

[Stefan Payne]

8 minutes ago, ScratchCat said:

Hawaii was a consumer fusion reactor and pulled over 300W.

Like Skylake-X, yet nobody cares about that with Skylake-X.

 

Quote

The RX series improved on that drastically but still were less efficient and back in 2016 Crimson wasn't particularly stable.

Oh, again with the Driver Instability Lie? If you don't have any other Argument, you get that skeleton out of the closet?

Becuse we ignore all the Problems that nVidia had. Like dying Cards in WoW, crashes in the Original Tomb Raider that weren't fixed for Months (AFAIR it took them half a year to fix them).

 

So why don't you say anything about the shitty nVidia Driver or the Incompetence when there is a Problem?

 

Quote

The RX X80s never consumed less than 150W (the power limit on the reference design) while the 1060 pulls 120W in the same situation.

Yes and has 2GiB more VRAM and 64bit wider Memory Interface...

And AMD Ryzen is more Efficient than Coffee Lake and consumes less power Also something that isn't mentioned much. 

Sorry, but you are just justifying to pay more for the shittier card...

 

And since you are talking about Efficiency, do you have a Ryzen CPU?