AMD: R9xx Speculation

mczak said:
Well you started with the inconsequent usage of percentage numbers :) For size you took Barts as baseline (Cypress 30% larger) but for performance you took Cypress as baseline (Barts cards 5-10% slower)... Admittedly it makes more difference for size than performance...
Click to expand...

Yes, probably I was a little unclear, but the around 20% perf/mm2 should have been the same looking both ways :D

mczak said:
Really, as far as performance goes I can't see many other tweaks. I think it also depends what you consider a "big" increase. As far as I can tell it's the result of removal of unneeded features, simds (which don't scale too well), and higher clock, not any magic changes (except for tesselation). Maybe it's also a bit more densely packed (due to process improvements?) since die size was reduced a bit more than transistor count.
Click to expand...

It makes me wonder how much of the die area was gained removing the DP and other smaller features. Anyway, I never said that it was "magic", but only the result of a better balanced chip: I already wondered what is the cause of poor scaling with the SIMD count in Cypress. Also, I see also some improvements with regard to the Juniper chip (maybe Juniper is too bandwidth limited?) considering the 6850 and the 5770...
It would be useful to see if Cayman will solve the scaling issue in Cypress.


mczak said:
These aren't bad results (I don't like those wall measurements if you want to convince me try other numbers), but again the efficiency just isn't that much higher. For the HD6850 vs. HD5850 you get like 20% lower power draw for 10% less performance (though it seems to depend on card as it looks like some have much higher voltage than others). Which definitely IS a best-in-class result, it's just not THAT much better than the old one.
Click to expand...

Again, never said that it was magic, but at the moment there are no reviews using the measurements on the card (xbit seems late this round). In any case it is an improvement (and the AFAIK AMD slides never said "how much", they said only "there is an improvement"). I was only criticizing the quoted sentence in the Tech Report review, because the AMD claims were not so far from the reality as the sentence implied. You have to consider that the process is still the same, as well as the architecture basics, so expecting a 50% breakthrough is simply not realistic.
 
leoneazzurro said:
Yes, probably I was a little unclear, but the around 20% perf/mm2 should have been the same looking both ways :D
Click to expand...
Indeed :).

Again, never said that it was magic, but at the moment there are no reviews using the measurements on the card (xbit seems late this round). In any case it is an improvement (and the AFAIK AMD slides never said "how much", they said only "there is an improvement").
Click to expand...
I usually trust these numbers too:
http://ht4u.net/reviews/2010/amd_radeon_hd_6850_hd_6870_test/index19.php
I think (but can't find it right now) the "games" load is 3dmark06. Power draw of Barts vs. Cypress isn't so hot there. In Furmark there's some improvement, however. Also note the tested HD6850 had high voltages (same as HD6870), I'm not sure if all retail cards are like that (the AMD reference definitely is not), which of course has a big impact and it would look better otherwise.

I was only criticizing the quoted sentence in the Tech Report review, because the AMD claims were not so far from the reality as the sentence implied. You have to consider that the process is still the same, as well as the architecture basics, so expecting a 50% breakthrough is simply not realistic.
Click to expand...
Ok makes sense.
 
mczak said:
I usually trust these numbers too:
http://ht4u.net/reviews/2010/amd_radeon_hd_6850_hd_6870_test/index19.php
I think (but can't find it right now) the "games" load is 3dmark06. Power draw of Barts vs. Cypress isn't so hot there. In Furmark there's some improvement, however. Also note the tested HD6850 had high voltages (same as HD6870), I'm not sure if all retail cards are like that (the AMD reference definitely is not), which of course has a big impact and it would look better otherwise.
Click to expand...

Hmm.. but if they are measuring the 12V railcurrent, aren´t thay missing the PCI-E slot wattage?
 
leoneazzurro said:
Hmm.. but if they are measuring the 12V railcurrent, aren´t thay missing the PCI-E slot wattage?
Click to expand...

the (5?) pins for PCIE wattage are easy to measure, xbit even has a good guide about it, i think everyone knows which PEG pins to measure.
 
wrt simd scaling, I'm wondering if efficiency just goes down because there are "too many" simds per dispatch processor? Maybe ideally there really would be less than 10? Even rv770 -> rv740 showed bad simd scaling though it was more attributed to the (lack of) interpolators then. But Juniper and Cypress continue that tradition.
So maybe Cayman using less simds per partition?
 
AMD promotes Cayman as the new R300

Very confident about Cayman

We have received some new information about Cayman, claiming that the new GPU will end up big and hot, which is not surprising as it is the biggest GPU that AMD has ever made. We don't have any exact numbers yet, but the general feeling about this quite secretive project is that it might be the new R300.

If you are old enough you will remember the famous Half life 2 vouchers and ATI’s first leader series, which included the now legendary Radeon 9700. This chip dominated against Nvidia's Geforce FX 5800, the infamous NV30, so Cayman has a lot to live up to. This was some eight years ago, in summer of 2002, way before Tweeter, Facetube and Youbook launched.

AMD is either hoping that Nvidia doesn't have a chance with its GF110 aka GTX 580 or that Nvidia could fail again. However, we believe that making the same mistake twice would be something that would really upset Nvidia CEO Jensen Huang.

ATI partners are confident and it looks like they should get enough cards. If all goes well Cayman Radeon HD 6970 should launch in last week of November, of course if it doesn't get pushed back further. http://www.fudzilla.com/graphics/item/20663-amd-promotes-cayman-as-the-new-r300
Click to expand...

Where did fudzilla get this information ------>(AMD promotes Cayman as the new R300)
 
R300/9700 was a groundbreaking card. For the first time both AF and AA were useable with playable FPS. I doubt Cayman can deliver something as monumental, but one can hope.
 
mczak said:
wrt simd scaling, I'm wondering if efficiency just goes down because there are "too many" simds per dispatch processor? Maybe ideally there really would be less than 10? Even rv770 -> rv740 showed bad simd scaling though it was more attributed to the (lack of) interpolators then. But Juniper and Cypress continue that tradition.
So maybe Cayman using less simds per partition?
Click to expand...

AFAIK, Barts already has two Dispatch processors or partitions as opposed to the one Cypress and Juniper had. But maybe Cayman has other secret sauce, doubling units that were singular before. ;)
 
IIRC, Dave said Cypress has the same dual dispatch, only it wasn't reflected in its diagrams.
 
CarstenS said:
AFAIK, Barts already has two Dispatch processors or partitions as opposed to the one Cypress and Juniper had. But maybe Cayman has other secret sauce, doubling units that were singular before. ;)
Click to expand...
No, Cypress and Barts are identical in this respect. Each shader engine needs its own sequencer.
 
I was thinking that if u have already two independent shader engines, could it be possible to make something like dual core GPU with sharing ROP-s and memory controler and the rest would act like a 2 gpu-s in crossfire. (crossfire 5770 on a single die with 256bit memory :p)
If two cards in crossfire can efectivly share the work than why not on die. And u would gain also parallel geometry quite easy.
 
Jawed said:
No, Cypress and Barts are identical in this respect. Each shader engine needs its own sequencer.
Click to expand...

Then I must have been looking at either the wrong slide or the slide does have an error in it because it clearly shows only one UTDP (and no plural either).
 
cypressdetail.jpg

http://sa09.idav.ucdavis.edu/docs/SA09_AMD_IHV.pdf

Is Northern Islands Architecture Deep Dive sill under NDA?
 
neliz said:
Can't trust 'em
They don't have the right TDP in all of them either :D (as per example by no-x a couple of pages back.
Click to expand...

No, he just highlighted the same issue: Don't trust slides. The same set of press decks, I let myself mislead into believing there's only a single UTDP, also explicitly mention a 170 watts TDP for HD 5850. In the HD 5800 series expanded PDFs for HD 5830's launch, it was magically shrunk to 175 watts - maybe due to a more mature mfg. process. It wasn't communicated explicitly.

But lesson learned. Always get mail confirmation on anything. ;)

edit: Idle TDP stayed constant though through both slide decks
 
Not only th HD5850... even some of the other slides seems to be wrong:

HD5670: 61W in this slide, 70W in the other one and 64W according to their website:



Cedar at 650MHz (HD5450) consumes 19.1W, but at 750MHz (HD5470) id consumes only 13-15W:



and HD5570 consumes 42,7W here, but 38W here... and 39W according to the website:
 
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
676K
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