Wow Terry is leaving AMD ?
AMD: R9xx Speculation
- Thread starter Lukfi
- Start date
Wow Terry is leaving AMD ?
I'm sorry but I only see GTX580 SLI beating even the overclocked 6970 CrossFireX in the link. But it's true though that AMD has been getting better with CrossFire lately, and they are now at least on par now with Nvidia on scaling.
I'm now looking forward to the 6990 model. The combination of good CrossFire scaling and Powertune might be a recipe for a killer product. Though on absolute performance per watt, Nvidia has rather surprisingly caught up, there the GTX570 is almost identical to 6970.
I think the most recent test of G80 and R600 is on ComputerBase:I wouldn't mind seeing G80 compared from original drivers up to today's 260.xx pack. That would be interesting I think. Probably because I'm still using one.
...
I'm sure that there's zero chance of it being proved as inefficient as R600 was though. R600 was just crazy nuts.
http://www.computerbase.de/artikel/...n-im-vergleich/8/#abschnitt_performancerating
and the only problem of R600 is slow MSAA... in more actual games and with new drivers it performs a bit better than 8800GTS and 60%(!) faster than X1950XTX w/o MSAA...
No, but he's changing positions within AMD. Unless his twitter account has been hacked, that is.
http://twitter.com/Catalystmaker
UniversalTruth
Veteran
If my memory still serves well, I think someone (perhaps TPU) claimed "shortage of components" or something similar.
^^ given the way it has turned out, I believe the delay was due to GTX580 and GTX 570. It's clear that AMD aimed too low with 69XX.
Of course, AMD won't admit it but I believe the last few weeks were spent tweaking the clocks and drivers to make 69XX look somewhat decent against GTX 5XX cards.
Of course, AMD won't admit it but I believe the last few weeks were spent tweaking the clocks and drivers to make 69XX look somewhat decent against GTX 5XX cards.
Er, drivers maybe, but you can't just spend a few weeks and tweak clocks like that; adjusting the clocks would have taken much longer than the 3 week delay, and since the delay happened at pretty much the last minute, there's no way for it to have been related to that. It's unlikely that the 580 and 570 impacted anything other than perhaps pricing.
UniversalTruth
Veteran
I think this is discussed pretty well here:
http://semiaccurate.com/forums/showthread.php?t=3867&page=2
Probably better than expected yield for XT chips and following shortage of some components for the XT version. But that's of course pure speculation.
http://semiaccurate.com/forums/showthread.php?t=3867&page=2
Probably better than expected yield for XT chips and following shortage of some components for the XT version. But that's of course pure speculation.
Babel-17
Veteran
http://forums.guru3d.com/showpost.php?p=3815985&postcount=33
Originally Posted by Mineria
True, although maybe you should have found a better nick than Catalyst Maker?
Ya no kidding. I guess I didnt think of all the consequences 8 or 9 years ago when I started it to post on Rage3D.
Looking forward, is this a major architecture change for Ati? Was the old architecture maxed out in terms of delivering more performance? I wonder if they are taking a hit on performance to lay down the groundwork for a new architecture?
So far, Fermi after the initial stumble has proved to be a better product with their 570/580 revisions.
So far, Fermi after the initial stumble has proved to be a better product with their 570/580 revisions.
Ultimately faster yes. Better? I don't see how a 25% bigger chip with horrendous yields can be better tbh.
If my memory still serves well, I think someone (perhaps TPU) claimed "shortage of components" or something similar.
Yea that was another rumour, the TI mosfet thing. They claimed that the same component was also being used on the 68xx series but there was plenty of supply of those cards so i dont buy that rimour
Yea maybe it was drivers but i dont believe 3 weeks is enough time to change clocks. The cards have to enter production ~1 month before launch. This is to first stockpile some quantity of cards for launch day, then the time for testing, packing, shipping(if by sea then its give or take a week total shipping time) and then cards usually reach retailers a few days in advance. So they would have had to make changes to a LOT of cards which were already produced by the time(probably in thousands if not tens of thousands)
It was reported on some other sites as well, i think TPU and VR-Zone were among them. But then again VR-Zone also claimed launch prices were $449 and $349. And when do we ever see the high end part outselling the low end part? Im not saying it isnt possible but generally lower end cards are produced in greater quantities
We assume yields are horrendous. Reports said that 40nm yields were now on par with 55nm. For all we know Fermi's yields could be similar to GT200/GT200b
I still think that comparing Cayman to Cypress, while natural for us, isn't very realistic considering the lag time in the design of these chips. This chip was probably nearly finished when Cypress was launched.
I imagine that right now the next new GPU is nearing completion and that it has lessons learned by Cypress incorporated.
I was pondering the pathetic performance of my former notebook's 3450 the other day and it dawned on me that with AA it is only a 2 px/clk chip. Performance with 2 and 4X AA isn't much different. Its performance overall isn't much different than a 9600 Pro.
I imagine that right now the next new GPU is nearing completion and that it has lessons learned by Cypress incorporated.
The infamous halved pixel rate when using AA. It does look like 'ol R600 is proving superior in more recent games to G80. Unless you like AA. That massive bandwidth and somewhat superior shader throughput coming into play maybe.
I was pondering the pathetic performance of my former notebook's 3450 the other day and it dawned on me that with AA it is only a 2 px/clk chip. Performance with 2 and 4X AA isn't much different. Its performance overall isn't much different than a 9600 Pro.
Last edited by a moderator:
Thats something which has to be considered. Source? Probably lower yields on Cypress inspite Cayman's larger die size for the most part. I can't really see the GDDR5 being any cheaper given the Cayman boards have gone for 2Gb chips.
If I'm not mistaken, Dave has mentioned that Cayman taped-out in May. Most design decisions were probably set in stone long before that.
Lowest number for GF110 I've read was 520 mm², which is over 33% larger than Cayman's 389 mm².
GTX 580's average performance advantage is clearly less than that, so perf./mm²-wise Cayman would be the better chip even if yields were similar.
Similar threads
