AMD: Speculation, Rumors, and Discussion (Archive)

Status
Not open for further replies.
I think people are underestimating Polaris 10 a bit here. Sure, there will probably be around $250 model, but AMD is literally trying to cover whole market with Polaris 10 and 11, which should mean that "full Polaris 10" will fit higher, maybe not the 980 Ti level suggested by some rumour, but still notably higher than GTX 970/R9 290
Looking back up to the GTX 470, all x70 GPUs have been launched between $330 and $400. A GP104 in 1070 configuration can be expected to be priced in that range. If a full Polaris 10 comes in at roughly the same performance, I think AMD will price just below, due to the 10% Nvidia tax.
So a launch price of $300 to $350 sounds about right. Definitely higher than $250, I agree. (If they price it $250, I've lost all hope for AMD.)

But that doesn't cover the whole market. In the end, it's only a 235mm2 chip and AMD and Nvidia should have similar perf/mm2 this time around. (Adjusted for ...)
 
Perhaps AMD have a contractual obligation to purchase/have purchased a certain number of HBM1 packages/dies, and the Fury products underwhelming market performance has left a significant quantity unused. It wouldn't be the first time AMD has been left in that situation, AMD have delayed products to clear the channel before.

Although if that were the case I would have expected more Fury price cuts.
 
Even if Polaris 10 comes in a little under Fury X, I don't see how the latter has a reason to exist in an environment where it's worth $400 at best.
Also who in Hades would buy a 4GB Fury when the 8GB 1070 is on the market for even less money? I don't see Fury being viable once the new generation of cards hits.
 
They would have to drop the price..... And this is exactly why Polaris will fit well into the under $300 range. Question is how much margins do they really have on the Fury line will determine if its still a viable product to make if they are going to drop the price.
 
Last edited:
AMD has one quarter for Polaris to snag up potential marketshare (their TAM for VR) before GP106 comes out. Is that enough. And is the price/performance ratio going to be good enough, cause power usage against the gp106 most likely won't. I say this because:

Before any one says anything about their 2.5 perf/watt advantage, that advantage is from the node, not from the architecture, so if that is what they are getting in base case, man..... I don't think anything more needs to be said.

We know Maxwell 2 has an advantage architecturally for power usage and the ability to clock higher and from the p100 presentations and docs looks like that is going to get extended some more for Polaris.

Did AMD have enough time after seeing what Maxwell 2 was capable of in the perf/watt categories to make changes in Polaris, I don't think they did nor did they have the ability to. I think nV learned quite a bit from low power SOC's through Tegra development and they used that in Maxwell 2 once they saw 20nm wasn't viable, and this is learned experience, (kinda like what the lead architect of Zen stated about mixing x86 with arm), nV was able to get best of both worlds. Not saying AMD can't get the experience, they just didn't have the need for it before they saw Maxwell 2.
 
Before any one says anything about their 2.5 perf/watt advantage, that advantage is from the node, not from the architecture, so if that is what they are getting in base case, man..... I don't think anything more needs to be said.
AMD's RTG disagree that it is all coming from the node:
How is Polaris able to achieve these types of improvements? It comes from a combination of architectural changes and process technology changes. Even RTG staff were willing to admit that the move to 14nm FinFET process tech was the majority factor for the improvement we are seeing here, something on the order of a 70/30 split.
http://www.pcper.com/reviews/Graphi...hnologies-Group-Previews-Polaris-Architecture
 
AMD has one quarter for Polaris to snag up potential marketshare (their TAM for VR) before GP106 comes out. Is that enough. And is the price/performance ratio going to be good enough, cause power usage against the gp106 most likely won't. I say this because:
I think Taylor is either unaware of or unwilling to discuss GP106, which I do not think was formally announced.
All the "what we've seen" shading to his language makes me think this is a smokescreen.

Before any one says anything about their 2.5 perf/watt advantage, that advantage is from the node, not from the architecture, so if that is what they are getting in base case, man..... I don't think anything more needs to be said.
There may be some architectural advantage, AMD has give a 70:30 split on node vs architecture. Some of the given features would seemingly have to provide some power benefit.

Did AMD have enough time after seeing what Maxwell 2 was capable of in the perf/watt categories to make changes in Polaris, I don't think they did nor did they have the ability to.
CPUs have been 4-5 years for a new design. GPUs used to be able to have turnaround that might have been 2-3 years. Off-hand comments seem to be putting the more complex design work for GPUs these days pushing them closer to CPUs.

The PS4's full hardware development time seems to have been ~4 years, with 2 years to lock down the spec, going by some statements by Mark Cerny. Possibly that leaves room for alterations in response, but hopefully there were ongoing projects for improving perf/W regardless of Maxwell 2.
There are elements in GCN3 that Nvidia is implementing or has implemented. GCN3 can address 32-bit registers and extract them at 8 or 16-bit granularity, which Pascal seems to be promising as well. What GCN3 doesn't do with that currently is provide more throughput in that mode, which is being done with FP16 and potentially with 8-bit as well with Pascal.
Whether Polaris can somehow sneak that in if its IP version at a high level is unchanged, I am unsure.
Vega, it seems would benefit given its likely focus.

One thing that stood out to me about Taylor's statements is the somewhat dismissive treatment of Nvidia's automotive focus. I know that's not really an area Taylor would need to concern himself with given his position, but frankly if I'm going to weigh the significance of deep learning and automating almost everything that moves on an industrial scale in a way that could change mobility and economies forever against VR, I am unconvinced that AMD is doing the more consequential thing.

I am generally curious about VR and Nvidia's success in that field is unclear to me, but to even bring that up in comparison makes AMD look like its chasing another flash in the pan.
 
Perhaps AMD have a contractual obligation to purchase/have purchased a certain number of HBM1 packages/dies, and the Fury products underwhelming market performance has left a significant quantity unused. It wouldn't be the first time AMD has been left in that situation, AMD have delayed products to clear the channel before..
That shouldn't be a problem: AMD has years of experience in paying hundreds of millions for things it doesn't use!
 
Last edited:
I think Taylor is either unaware of or unwilling to discuss GP106, which I do not think was formally announced.
All the "what we've seen" shading to his language makes me think this is a smokescreen.

Yeah that speaks to me as well the same way. Unlike his usual belligerent responses, there seems to be quite a bit of room to play with his statements.


There may be some architectural advantage, AMD has give a 70:30 split on node vs architecture. Some of the given features would seemingly have to provide some power benefit.


Given the fact that 2.5 perf/watt is the best case (and this seems to be applications that favor AMD to begin with), I think its more like 2 times perf/watt in most cases. And this is the same thing nV stated for Pascal, an average of 2 times. Just a cursory look at things right now, AMD gets 2.5 perf/watt increase, that puts them squally equal to Maxwell 2 in the perf/watt category.....

CPUs have been 4-5 years for a new design. GPUs used to be able to have turnaround that might have been 2-3 years. Off-hand comments seem to be putting the more complex design work for GPUs these days pushing them closer to CPUs.


The PS4's full hardware development time seems to have been ~4 years, with 2 years to lock down the spec, going by some statements by Mark Cerny. Possibly that leaves room for alterations in response, but hopefully there were ongoing projects for improving perf/W regardless of Maxwell 2.
There are elements in GCN3 that Nvidia is implementing or has implemented. GCN3 can address 32-bit registers and extract them at 8 or 16-bit granularity, which Pascal seems to be promising as well. What GCN3 doesn't do with that currently is provide more throughput in that mode, which is being done with FP16 and potentially with 8-bit as well with Pascal.
Whether Polaris can somehow sneak that in if its IP version at a high level is unchanged, I am unsure.
Vega, it seems would benefit given its likely focus.

That is true but when there really is no need to focus on something deeply, will they be so intent on creating something like that? yeah I think Vega will have more time but again, looking at the charts provided, Vega will get the benefit of HBM 2 for power savings, which coincides with the perf/watt charts AMD has already provided for their generational timelines.

One thing that stood out to me about Taylor's statements is the somewhat dismissive treatment of Nvidia's automotive focus. I know that's not really an area Taylor would need to concern himself with given his position, but frankly if I'm going to weigh the significance of deep learning and automating almost everything that moves on an industrial scale in a way that could change mobility and economies forever against VR, I am unconvinced that AMD is doing the more consequential thing.

I think it was more of Roy being Roy, trying to down play something that AMD can't come into because they don't have the resources to push it.

I am generally curious about VR and Nvidia's success in that field is unclear to me, but to even bring that up in comparison makes AMD look like its chasing another flash in the pan.

From what we have seen from Pascal looks like nV has addressed some of the short comings of Maxwell for VR, but AMD does have a leg up, yeah seems to me they have an advantage.
 
From what we have seen from Pascal looks like nV has addressed some of the short comings of Maxwell for VR, but AMD does have a leg up, yeah seems to me they have an advantage.
I haven't seen anything specific in Pascal related to VR that makes it better than Maxwell. But I also have a hard time seeing what makes AMD so special.

To me, VR seems to be the 4K of the past: a hook on which to hang a marketing story, but not something that makes a major difference at the chip architecture level. In the end, triangles will still have to be rendered...
 
My interpretation of their rhetoric and road map is that AMD is at least temporarily ceding the high end to Nvidia.

-Talking up perf/W instead of absolute performance.
-Positioning against 960 -970 range, TAM, VR min spec pitches.
-Vega. Amd doesn't have the HPC and professional market share foothold to justify a non-gaming chip like GP100, and 2017 is too soon to replace Polaris, so it seems like Vega is to address the high end.

Fortunately for AMD, Nvidia abandoned high-end market back in Fall of 2009 and the entire graphics field in Jan of 2010.
 
I haven't seen anything specific in Pascal related to VR that makes it better than Maxwell. But I also have a hard time seeing what makes AMD so special.

To me, VR seems to be the 4K of the past: a hook on which to hang a marketing story, but not something that makes a major difference at the chip architecture level. In the end, triangles will still have to be rendered...


Well I was talking about preemption. Which was just a natural progression more or less.
 
My interpretation of their rhetoric and road map is that AMD is at least temporarily ceding the high end to Nvidia.

-Talking up perf/W instead of absolute performance.
-Positioning against 960 -970 range, TAM, VR min spec pitches.
-Vega. Amd doesn't have the HPC and professional market share foothold to justify a non-gaming chip like GP100, and 2017 is too soon to replace Polaris, so it seems like Vega is to address the high end.


It was never their intention to cede anything. I don't think any company would ever do that in any market, cede a market segment if at all possible. Yeah timelines can sift a little bit so there will be times where the others can take advantage. Just like what Geeforcer pointed out, the 2xx line for nV, it was a case of stretching a particular architecture for too long and giving the chance for its competition to catch up. I don't think nV will let something like that happen if at all possible. If it happens then they screwed up somewhere.

If a company has a certain advantage, its up to them not to further that advantage, if they don't they are doing something wrong on their end or they took the wrong direction to begin with and are trying to fix it.
 
For be honest, both camp put me on nerves.. look like i will need to wait 2017 for upgrade finally my hardware.. GP100-102 or Vega.. I pass my turn for the little gpus we will see in 2016 from both camp.
 
Yeah people with Fury the line or 9xx line, I really don't think these are GPU's to upgrade to, best wait for early 2017.
 
They would have to drop the price..... And this is exactly why Polaris will fit well into the under $300 range. Question is how much margins do they really have on the Fury line will determine if its still a viable product to make if they are going to drop the price.
I don't see how they could possibly sell Fury (Giant chip + Interposer + HBM) at the same price as salvaged GP104. Or rather, they could do that, but then they're back to zero or negative profit margins in a rather important segment (mid-highend).
 
However people still rocking a GTX 670 can't wait that long! Not with an Oculus Rift arriving in 3 months.

The guys who are still rocking a GTX670 excellent gpu at his time,( an excellent compromise of perf / price ) will not put 900$ in an OR in 3 months.
 
Last edited:
Status
Not open for further replies.
Back
Top