Well, GTX 580 can be really good, especially for people like me. GTX 480 was not enough to replace OCed 260 SLI, but GTX 580 is. I decided to go 1-card now. SLI means great performance, but not all games support it, and you can occasionally get stuttering/jitter in some of them. Time to wait for a frame is also higher. Still, I'm not going to get GTX 580 till I see good review, and possibly a comparison with 6970.
You have eyes? Look at the damn picture and count those ****ing shader clusters. GF100 comes with 512 CUDA cores. They disabled one SM cluster in GTX 480.
Poor troll.............. Anyways i am really excited about the gtx 580 Rumors also say the may release a dual gpu in December! Cant wait
I doubt they'd release dual GPU. GTX 580 already has power-related "issues". With its current limiter it seems that heavily OCed and dual-GPU cards are unlikely in the near future. Hilbert, I know that you use the current drivers, but I'd like to ask you to ensure that 480 gets retested along with 580 using the same drivers. If you can compare 580 to 480 at 580's clocks, it would be so awesome too.
Still doesn't say anything is disabled in the GTX 480, they must have just built it with less, not disabled or locked anything... Think about it logically, a) why would they go to all the trouble and waste of money to laser cut something and b) have a lower performance card when this would have less people buying it and not compete aswell with AMD/Ati??? doesn't make sense m8
+1000000000 HH = Guru Of 3D @ Rag: ????? HH said: But do realize I got architecturally briefed by NVIDIA's CTO on this.
This doesn't make sense at all. a) Building a card with such number of clusters wouldn't cost less. People know it if they have just a basic knowledge about building microprocessors. Another issue is yield - if you build a GPU with 1 faulty cluster tolerance, you get much better yields than with 0 tolerance. b) 480 was already faster than 5870 - it's enough to be just faster. People were buying GTX 480 because they were faster than 5870 and 470 as they were faster than 5850. They didn't need to make card that's 2x faster. + it would be more difficult to release 580 now if 480 already had all clusters enabled. well... anyway RagDoll knows stuff better than NV guys who informed HH
Yes, they are just built with lower performance spec By your terminology, this would mean that AMD's 5870 is just a 6870, well??? Or why don't they just build a GTX 680 now and disable/lock things and call it a GTX 585 or something??? If the GTX 480 was really a 580, now competing with a 5970, why didn't they bring this out before and just have the performance crown, therefore selling more..
http://www.anandtech.com/print/2977 http://en.wikipedia.org/wiki/GeForce_400_Series Its not just Guru3d that says that it is everybody. You want to say everybody is wrong, and you are right ?
Simply because GTX480 had heat and power consumption issue thats why they couldn't release chip any bigger. It was already over the limit TBH. Same way you could ask AMD why they wont release HD6xxx with let's say 2400 Shaders, there is certainly space for it. GPU is suppose to be balanced , not overdone power monster. nvidia fixed some power and heat issues and now releases GTX480 with more shaders. Quite simple.
I can't wait for your "very impressed" posts when the 6970 comes in and beats or meets the 580's performance. Personally In looking at the 1920x1200 BFBC:2 results only ~10fps up from 480. The COD4 results were very good and I am thinking of picking up Black Ops so hope to see some benches when that comes out. But it's Faster/Quieter than the 480 and with price drops across the board I am glad they released it before the holiday rush.
Got Proof from Nvidia site http://www.nvidia.com/object/GTX_400_architecture.html Download the GTX 400 Graphics Architecture White Paper bottom right on the page. In it will say "GF100 implements 512 CUDA cores, organized as 16 SMs of 32 cores each. Each SM is a highly parallel multiprocessor supporting up to 48 warps at any given time. Each CUDA core is a unified processor core that executes vertex, pixel, geometry, and compute kernels. A unified L2 cache architecture services load, store, and texture operations."
it improves yields, if u have a defective clusters on a wafer you can still use the chips for a 480/470 etc
rag doll fail. They had such bad yeilds with all 16 SM's enabled plus it ate up too much power and was extremely hot. - even at the launch of 480 with 15 SM's enabled they had temp issues, that was the main problems and that was the problem of why Nvida's 400 series cards was so delayed. and it is easier and cheaper to produce 1 set of chips and just laser 1 SM for for the 480 and another for the 470, to save money instead of making 2 separate chips. You do realise that this generation isn't really a new generation technically speaking - the 6000 and 500 series are just modifications on the current series cards, the 6000 series card are just remodled 5000 series chips, and the 500 chips are remoddled fermi - GF100 was 480 /470/465, GF104 was 460, GF106 - 450/440 and GF108 was 420/430. And now the 580 is GF110. realistically the 580 should of been called the 485
It doesn't mentioned disabled anywhere in that nvidia pdf???, they just built the 480 with one less sm, and the 470 with 2 less... it says at launch none were available.
^ dude believe what you will, do u not realise how uch more expensive it would be to build the two cards on different designs? Add that to the fact that they are both GF100 and it is clear. As they say ignorance is bliss, enjoy