• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

[BitsAndChips]390X ready for launch - AMD ironing out drivers - Computex launch

Page 71 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
desprado said:
it is not 20 nm that is been confirmed and it is also been posted in this topic by someone.

R9 285 is based on GCN 1.2 and it is not new but it has improvements.
Click to expand...

He said regardless, the point was that the Glo-Fo silicone may clock differently.
 
Strictly dual gpu speculation.

Wouldn't it be possible to design a dual gpu that is hidden from the outside?

For clarification: To the OS and software its seen as a single gpu.

Wouldn't this eliminate the crossfire/sli issues?

Further clarification: 1 single die(piece of silicone) with multiple gpu configurations. mGPU controlling the backend. Built in ability to laser cut defects for salvage ability?

Thread is getting boring 🙂

Tic,toc....Month is a long time!
 
Last edited:
Kenmitch said:
Strictly dual gpu speculation.

Wouldn't it be possible to design a dual gpu that is hidden from the outside?

For clarification: To the OS and software its seen as a single gpu.

Wouldn't this eliminate the crossfire/sli issues?

Further clarification: 1 single die(piece of silicone) with multiple gpu configurations. mGPU controlling the backend. Built in ability to laser cut defects for salvage ability?

Thread is getting boring 🙂

Tic,toc....Month is a long time!
Click to expand...
The idea of hidden dual GPU setup would be possible if second die would be on the other side of PCB and directly connected through the board with another core.

Technically it is possible, but it is hard to understand for system from driver point.

If we will count the Dual Link interposer, and the idea, that second part of the HBM would be on the back of PCB. We have entered the world of unending possiblities. 3d Computing at its best manifestation.

I know its just and idea, however think about what we know so far. Tonga technology, HBM, Dual Tonga die, Dual link interposer, and theoretically technology to see the dual GPU by the system as one.

Doable? Of course. But would anyone be that crazy to do such thing?

Again. One PCB, HBM with dual link interposer, two dies on the same PCB, just placed on each side of PCB. Synchronising them is just a matter of.. mature drivers, if you know what im trying to tell.

Its just theorycrafting 😉. So take it like it is 😉.
 
The idea of having an interposer is to confine the interconnects to the silicon of the interposer, so interconnects through the PCB are not going to work for HBM, imo.
 
Cloudfire777 said:
So now you try to downtalk it since it shows the card might be expensive?

I already posted the same screenshot as that link in the same post showing 2 new Fiji have just shipped.
Click to expand...

Hey maybe it beats Titan X all ends up and hence is only a few hundred dollars more? Or the 'PRINTED CIRCUIT BOARD ASSEMBLY' has a markup, unless it's just a fancier way of saying 'graphics card'.

And no you didn't post the same screenshot.
 
Glo. said:
Then maybe, I have understood it wrong. I stand corrected then 🙂.
Click to expand...
I don't think there is any evidence that eight stacks of four HBM1 modules can't be positioned around the die on a large interposer. That would yield 8GB on one side of the PCB.
 
At their financial analyst day they predicted a single digit upswing in their GPU revenue. Pretty solid indication of just a couple of new top end SKUs and the rest rebrand/respins.
 
The trouble is every stack brings bandwith. So 2 stacks of 4 GB modules would bring 1 TB/s of bandwith.

1GB - 128 GB/s, 4 GB - 512 GB/s. There is no other way of doing this in HBM 1.

So, the only way to bring 8 GB's of HBM is to stuck them directly on the other side of PCB, where they can connect with other stacks. Of course, im not engineer, I can be completely wrong. But that is what I understand from the specification and idea of Dual-Link Interposer.
 
They could have some sort of interleaving or routing chip. It would impact usable bandwidth when over 4GB though.
 
Last edited:
Vesku said:
At their financial analyst day they predicted a single digit upswing in their GPU revenue. Pretty solid indication of just a couple of new top end SKUs and the rest rebrand/respins.
Click to expand...

How'd you get to that point? Their cards aren't selling now and they are posting losses. How are a couple of low volume (even if high margin) parts going to turn into growth for them if nothing else in the stack sells?
 
twjr said:
How'd you get to that point? Their cards aren't selling now and they are posting losses. How are a couple of low volume (even if high margin) parts going to turn into growth for them if nothing else in the stack sells?
Click to expand...

Because a top to bottom refresh would be expected to claw back more of the traditional marketshare and by extension revenue split that AMD has lost since the 900 series launch. A competitive top tier, however, could be expected to halt the decline even if the rest is rebrands/respins and drive a modest single digit recovery.
 
Vesku said:
Because a top to bottom refresh would be expected to claw back more of the traditional marketshare and by extension revenue split that AMD has lost since the 900 series launch. A competitive top tier, however, could be expected to halt the decline even if the rest is rebrands/respins and drive a modest single digit recovery.
Click to expand...

On what basis do you come to these conclusions? It cant be history repeating itself. For that to happen it would have had to happen before. It hasn't. AFAIK.
Why you would think a rebrand on all but the top tier would halt a decline is something im interested in hearing.
 
Last edited:
Vesku said:
At their financial analyst day they predicted a single digit upswing in their GPU revenue. Pretty solid indication of just a couple of new top end SKUs and the rest rebrand/respins.
Click to expand...

I don't see how they expect to get any upswing in GPU revenue if all they bring to the table is Fiji and a bunch of rebranded crap. Sales aren't going to remain the same on the lower-end cards if we get pure rebrands; they're going to drop. If they really do this, I expect every single review of the rebrands to be negative, and AMD to have essentially no GPU market share left by mid-2016.
 
JDG1980 said:
I don't see how they expect to get any upswing in GPU revenue if all they bring to the table is Fiji and a bunch of rebranded crap. Sales aren't going to remain the same on the lower-end cards if we get pure rebrands; they're going to drop. If they really do this, I expect every single review of the rebrands to be negative, and AMD to have essentially no GPU market share left by mid-2016.
Click to expand...

It isn't quite clear for me how the rebrands would get bad reviews and why it would drop their market share to a major extent.
Our purview is on the enthusiast side, how much do most of us know about how well the low end dgpu business is doing?
 
Keysplayr said:
On what basis do you come to these conclusions? It cant be history repeating itself. For that to happen it would have had to happen before. It hasn't. AFAIK.
Why you would think a rebrand on all but the top tier would halt a decline is something im interested in hearing.
Click to expand...

Nvidia managing to get the GTX 480 out the door finally even though it wasn't ideal it still retook fastest card spot and did a decent job of halting sales losses to AMD's 5800 series. Heck, Nvidia CEO simply promising the GTX 480 would be faster and "soon" halted the sales shift to some degree.
 
Vesku said:
Nvidia managing to get the GTX 480 out the door finally even though it wasn't ideal it still retook fastest card spot and did a decent job of halting sales losses to AMD's 5800 series. Heck, Nvidia CEO simply promising the GTX 480 would be faster and "soon" halted the sales shift to some degree.
Click to expand...

If you see Q2 2010 (GTX480 was released on 26 march 2010) AMD had 60% of the High-End ($300+ HD5870 and GTX470-80 market). GTX470/480 didnt gain market share for NVIDIA, low end Fermi cards did from Q3 2010 onwards.

But that was only because GF100 Fermi was regarded as too hot, loud and expensive compared to HD5870.
Fiji may gain High-End market share if it is a better product at lower price than GTX980/Titan X.
 
AtenRa said:
If you see Q2 2010 (GTX480 was released on 26 march 2010) AMD had 60% of the High-End ($300+ HD5870 and GTX470-80 market). GTX470/480 didnt gain market share for NVIDIA, low end Fermi cards did from Q3 2010 onwards.

But that was only because GF100 Fermi was regarded as too hot, loud and expensive compared to HD5870.
Fiji may gain High-End market share if it is a better product at lower price than GTX980/Titan X.
Click to expand...

Nvidia's revenue actually increased modestly upon launching the GTX 480 if you look at Nvidia's financial reports, which is similar to the financial guidance AMD is giving.
 
Keysplayr said:
On what basis do you come to these conclusions? It cant be history repeating itself. For that to happen it would have had to happen before. It hasn't. AFAIK.
Why you would think a rebrand on all but the top tier would halt a decline is something im interested in hearing.
Click to expand...
If the performance, features and pricing are attractive relative to Nvidia's offerings, then I don't see what the problem is. After all, an Nvidia 960 is a newer card yet I doubt most of us would take it over a 290X if we had to choose which one to put in our gaming rigs.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top