AMD: R9xx Speculation

TechRadar talked to AMD's David Hoff recently:

Hoff was also happy to speak about it's upcoming new range of DirectX 11 graphics cards. "It's certainly more than changing a sticker," he said with a grin. "I wouldn't say it's an absolute, complete from the ground up new architecture. It's a nice, different architecture."
Click to expand...

No concrete information, but better than a fake slide.
 
The earllier snippet, plus:

"A one year cadence is about the best you're ever going to see on a new architecture," he continued.
Click to expand...
does, at least, make it sound like a substantial change. Have to admit I'd started wondering if it was only going to be a refresh, constrained by 40nm.
 
LordEC911 said:
False... TSMC was completely bottlenecked from their own problems.
Click to expand...

And yet nvidia still got wafers for their gt200 series.

The point is AMD didn't get enough wafers, the ones they did get have excellent yield -- well at least much better than nvidia.

News flash, that's not AMD's fault (well, except for, like I said, bad planning), nor can they do anything about it except for book more wafers next time.

Only yielding problem is the real production problem here. So you are the one with completely misleading information.
 
Mianca said:
Sorry, but I really HOPE for the sake of AMD that whoever "leaked" that slide deliberately changed at least ONE spec and/or the performance targets (the screenshot was taken in edit mode, after all ...).
Click to expand...

You don't have to hope. Such products will never see the light when you can simply get a 5830 with 32ROP running @~725MHz which would have the upper hand on TDP, cost, performance and T2M.

Erinyes said:
AFAIK back in late 2008
Click to expand...
That's only at most one year before the production date and NI would have been or was only a few months away from being taped-out by then. And your target process should set alongside your floor plan -- not something you can just change that late. And if you do change sometime close to first silicon, that's at least half a year gone straight to waste.

jaredpace said:
So that barts slide is real :LOL:
Click to expand...

You can safely bet on the other horse. But I won't get in your way if you insist.
 
Last edited by a moderator:
racca said:
And yet nvidia still got wafers for their gt200 series.

The point is AMD didn't get enough wafers, the ones they did get have excellent yield -- well at least much better than nvidia.

News flash, that's not AMD's fault (well, except for, like I said, bad planning), nor can they do anything about it except for book more wafers next time.

Only yielding problem is the real production problem here. So you are the one with completely misleading information.
Click to expand...

So... your conclusion is that AMD didn't "book" enough wafers solely because Nvidia had 40nm GT2x0 wafers?
TSMC had stated a certain output volume for summer '09 of 40nm wafers, they didn't reach those numbers until around the end of '09/beg of '10.
 
racca said:
And yet nvidia still got wafers for their gt200 series.

The point is AMD didn't get enough wafers, the ones they did get have excellent yield -- well at least much better than nvidia.

News flash, that's not AMD's fault (well, except for, like I said, bad planning), nor can they do anything about it except for book more wafers next time.

Only yielding problem is the real production problem here. So you are the one with completely misleading information.
Click to expand...

Uh...

There was no bad planning involved. According to what TSMC was telling anyone they were planning to ramp up wafer production as normal.

AMD made wafer allocations based upon the assumption that wafer ramping would be similar to past nodes (55 nm, 65 nm, etc.) with a chance of it being slightly slower. The fact that TSMC then had significant problems producing enough wafers to meet demand on 40 nm wasn't anything AMD could have planned for in advance.

Sure if AMD could have looked into the future and known 40 nm wafer supply was going to be severly constrained, I'm sure they would have secured a larger initial allotment.

If that was bad planning, then ATI/AMD have been doing bad wafer allocation planning at TSMC going all the way back to the 90's. :p

And enough for GT200? Sure, Nvidia took a huge gamble and secured a large allotment of wafers hoping to be able to launch GT200 on a hot lot even if yield was non-optimal. That was, what, around Oct./Nov. 2009? Half a year later they finally launch and can't sell all of the few cores they did produce. Then had to go and sell back some of their allotment. And suffered massive losses. Good planning? Or yet another company suffering from TSMC problems at 40 nm? Hopefully, you'll be at least consistent in which of those choices you go with. :p

Regards,
SB
 
@ neliz: I feel Rick-Roll'd.

This is the AMD R9xx Speculation thread - and you post a link to a badly written Cypress OC analysis based on the experiences of a pubertizing kid trying to run Crysis @ different clock and memory speeds?
 
Mianca said:
@ neliz: I feel Rick-Roll'd.

This is the AMD R9xx Speculation thread - and you post a link to a badly written Cypress OC analysis based on the experiences of a pubertizing kid trying to run Crysis @ different clock and memory speeds?
Click to expand...

Yes yes, but don't you see! Barts only needs a 128-bit bus!

Like that will change the game...
 
Have you spotted this or not yet? :LOL:
Charlie says: :oops:


If a x4 cluster is only as fast as an x5 cluster, then (1920*5/4) = 2400. 50% faster there.

Now I think the real speed gain will be from the front end, if they can up efficiency by a good chunk in parsing the workloads, you would get an increase on top of that 50%. Think about having a 25% more efficient front end, that would mean an effective 2400 *1.25 = 3000 shaders.

Can you say twice as fast from a 10-20% area increase?

-Charlie
Click to expand...


http://semiaccurate.com/forums/showpost.php?p=69973&postcount=182
http://semiaccurate.com/forums/showthread.php?t=3249&page=19



Now, I am really excited and happy. Charlie, thank you for making me so happy. :p
 
neliz said:
Yes yes, but don't you see! Barts only needs a 128-bit bus!

Like that will change the game...
Click to expand...

The 256bit bus is there for the 32 ROPs and not just plain bandwith increase if the specs are true. The writer completly missed the point that 5770 had 16 ROPs.
 
GZ007 said:
The 256bit bus is there for the 32 ROPs and not just plain bandwith increase if the specs are true. The writer completly missed the point that 5770 had 16 ROPs.
Click to expand...

I'm jesting!
I just put it up to show people that with just 14 years of age and a PC, you could write articles on the interwibblez too that would seem deep and informative to the ill-informed.
 
You must log in or register to reply here.

Similar threads

F
RDNA5 Speculation
Replies
2
Views
1K
Erinyes
E
N
AMD: Volcanic Islands R1100/1200 (8***/9*** series) Speculation/ Rumour Thread
141 142 143
Replies
3K
Views
675K
pMax
pMax
S
AMD: Sea Islands R1100 (8*** series) Speculation/ Rumour Thread
69 70 71
Replies
1K
Views
389K
AnarchX
A
A
  • Locked
News & Rumors: Xbox One (codename Durango)
613 614 615
Replies
12K
Views
2M
BRiT
BRiT
Unknown Soldier
AMD Talks ATI Radeon 3000-Series
Replies
6
Views
6K
Silent_Buddha
S
Back
Top