Originally posted by: coldpower27
Originally posted by: keysplayr2003
Originally posted by: ronnn
Originally posted by: keysplayr2003
Originally posted by: ronnn
Of course they want to reduce costs. Very surprising to find a company that doesn't. Past history shows many second gen cards that use reduced bus, but still manage to keep pace with previous flagship., so that part is not surprising. The interesting part is that we should be seeing some real dx10 by that time - and will see if the architecture competes well. The R600 may be more efficient (not counting power use) than you give it credit for.
I agree with you for the most part. But I don't think that a reduced bus was ever done to new flagship cards. I mean, this 2950pro is aimed to replace the 2900XT, correct? Still has 320 shaders kicking, just seriously reduced manufacturing process. So the core is the same (architecturally) and pumped up to 850MHz. But, doesn't this GPU have a 512-bit internal/external Ring Bus Controller? Are they going to cut this in half as well? It is going to be very interesting to see how this setup pans out for AMD. If this thing performs at least AS good as a current 2900XT for the 250 dollar price point, I know I would be sold. I really don't care about the bus width as long as the performance is there.
It would really be something special if AMD got back into the game here with something fierce.
Was my limited understanding the the 2950pro was not a flagship, and the flagship was to be a gemini. Hence the price.
Judging from the specs, and if indeed the 2900XT would not have been memory starved at 256-bit memory interface, the 2950pro will indeed be faster than a 2900XT due to it's increased core clock. But again, only if it is not bandwidth starved with a 256-bit bus. You can't really go by price point because as I said, with these modifications, the 2950pro will be dirt cheap to manufacture compared to a 2900XT.
My understanding of this situation is not quite clear either, as you can see.
This is a truly smart move by AMD. Bout time eh? Better late than never.
Yeah if you can have HD 2900 XT like performance at more reasonable price points of 199-249US this card could sell quite nicely, and with a 80nm -> 55nm transition that is a full node shrink, though it's going from half node to the next half node.
If assuming ATI didn't remove any transistors on the R600 and simply shrunk it down to 55nm your looking at a die size of only 200mm2 or so, which is around the level of both the G71 and RV570. Great for cost reduction, the only issue here is that this is coming in in Q1 2008 I believe, so it is still quite a large distance away.
512Bit while nice is a bit excessive at this point chopping it down to 256Bit, and increasing the memory clocks to compensate is the same trick used when 256Bit cards performance levels get shifted to 128Bit cards.
76.8GB/s on the high end HD 2950 Pro as currently rumored is no slouch in terms of amount of memory bandwidth. That is more then the GTS 640 has and that card does pretty well considering it's bandwidth level.
This could be a great GPU for performance-mainstream gamers, I am eagerly hoping Nvidia will do something similar and shrink down to at least 65nm and bring us higher clocked 8800 GTS like SKU's as well.
Though 55nm manufacturing this early in the game is surprising TSMC I thought in past recent history, was usually 1 full node behind Intel and now they are only a half node behind? Interesting to say the least.
