By flagship do you mean GK104 or the thing that comes in 3-12 months from now? If GK104 is competitive then AMD has cause for concern. Otherwise they should have their dual-GPU card out in time to greet GK100 so no worries there.
AMD: Southern Islands (7*** series) Speculation/ Rumour Thread
- Thread starter UniversalTruth
- Start date
By flagship do you mean GK104 or the thing that comes in 3-12 months from now? If GK104 is competitive then AMD has cause for concern. Otherwise they should have their dual-GPU card out in time to greet GK100 so no worries there.
Ah. Yes, I used to have an association there...
Anyway the two you mention are F1 and Batman...they had PhysX on HIGH for the SLI and none (obviously) for the AMD cards...a bit tilted, eh?
F12011 is the only one that's faster - clearly those terrible Nvidia drivers!
Doing a rough estimate, I would guess a dumb shrink of GF110 would reduce the die size from 529mm² to be something in the ~315mm² range (memory interface doesn't shrink, TSMC's HKMG processes offer only ~1.8 times the transistor density of 40G without following special layout rules). NVidia needs to put some more tricks in it, a shrunk GF110 isn't going to beat Tahiti in any convincing way. It may be able to match it, if they invest everything of the 28nm process advantage in increased clocks (meaning power consumption would be the same as a 40nm GF110). I think it is not a viable option.
FYI - redundancy means that there are extra elements built in to attain a full configuration in more cases, but those redundant elements can never be used in addition. I think your question would be whether this is harvested.
).
AFDS was talking about the architecture in terms of capabilities and scalability, not any specific products implemented on the architecture.
iMacmatician
Regular
I think GK112 will be close to 300 W, but GK104 200-250 W.
Like in Cayman?
1:2 may have been mooted as an option for GCN as the overall initiative, but it wasn't stated which, if any, specific physical implementation would have it.
(edit: I see this has just been clarified above.)
I look forward to any analysis of the architecture with more effort than a Vantage test run.
For one thing, I am curious how GCN's transcendental ops and other T-instructions have been mapped from VLIW5 to VLIW4 to GCN, and what the throughputs are.
What the atomic ops are being performed with, and the bandwidths of the now read and write interconnect.
The performance numbers seem nice, I'd be tempted to buy one, but I'm the sort who would wait.
I wouldn't get the stock cards with the stock cooler. The last time I bought an upper-tier card I waited until an aftermarket large cooler was available.
By the time there are custom 7970s or a confirmed quiet cooler is available, a competitor may be on the horizon.
(edit: I see this has just been clarified above.)
I look forward to any analysis of the architecture with more effort than a Vantage test run.
For one thing, I am curious how GCN's transcendental ops and other T-instructions have been mapped from VLIW5 to VLIW4 to GCN, and what the throughputs are.
What the atomic ops are being performed with, and the bandwidths of the now read and write interconnect.
The performance numbers seem nice, I'd be tempted to buy one, but I'm the sort who would wait.
I wouldn't get the stock cards with the stock cooler. The last time I bought an upper-tier card I waited until an aftermarket large cooler was available.
By the time there are custom 7970s or a confirmed quiet cooler is available, a competitor may be on the horizon.
The numbers are impressive it seems but I think I would still go with a pair of 580 3gb SLIed because I know Nvidia's drivers well. If I was a single card user then I would have made a beeline towards the ATI offering because of the substantial improvement over what I am running. 20% increase in performance over Nvidia is impressive especially with such early drivers. I think ATI has produced an impressive product. The ball is now in Nvidia's court. Mind I am running factory overclocked 285 gtx's so my video cards are at least a few generations behind. I have waited this long I think I should wait a bit more despite me chomping at the bit to get new video cards and another 30 inch monitor.
As someone who ran 2x6970s for the past year and dealt with some CFX issues, and now has 2x580s (3GB) I would say it would be extremely tempting to give AMD another shot at drivers for a CFX rig that might be 50% or more faster than the 2x580s for the same price. That crossfire scaling is insane and the idle power would mean not feeling compelled to shutdown my computer when I'm done. I have to see the AF improvements too as that was a big gripe of mine.
I wonder if the revamped memory pipeline and paging support lends itself to improved crossfire performance.
So has Tahiti already the capability to do FMAs in DP with a 1:2 rate or is there just the possibility to scale GCN to that in a future implementation (like the CI top model)?
Otherwise I read your statement as somewhat confirming Tahiti (not only the HD7900 line) is capable of 1:4 only.
If it were stated that Tahiti could do 1:2 DP, I'd start wondering what it could have done if it had devoted the hardware for DP to other things.
The zero power core technique would be interesting when testing crossfire rigs. It would be trivial to know when AMD disables crossfire for a game after a driver update.
The zero power core technique would be interesting when testing crossfire rigs. It would be trivial to know when AMD disables crossfire for a game after a driver update.
As in the case of GF100/110?
The thought crossed my mind back then.
The pressure seems more acute to me when it comes to a design with a smaller die, although it may be open to debate that a GCN chip allowed to grow 40% would be able to utilize all that die area on 28nm, since it can hit its TDP limit at 365mm.
Thanks to the amazing drivers, you can actually go from 580 SLI to single 7970 and get a performance boost in DX11 titles
580SLI vs single 7970
BF3 33%+
Skyrim 17%+
F1 2011 -43%
Shogun 2: 18%+
Batman:AA -10%
Another tidbit on this:
1. Batman:AA: As already stated DH decided to tilt the benchmark toward AMD by enabling PhysX at High (effectively disabling a large portion of one of the two 580s).
2. F1 2011: If you read the non-CFX review, the 580 and 7970 are pretty close to even at 19x10. DH then drops the 580 from the graph for 57x10 (why?) and then shows, in their CFX review that the 580 is now 43% slower in SLI...only they didn't mention that there was an F12011 patch released days ago that broke SLI and they apparently used this without mentioning SLI was disabled. Of note is the SLI scaling is typically ~1.6x for SLI...so the benchmark number is exactly the difference one would expect from SLI being disabled.
I'm not trying to play any type of fb here, but I call out bad/tilted reviews when I can.
C- for DH on this review (would have been a D if not for the pretty graphs)
Similar threads
