AMD Fusion Developer Summit Starts This June

[Guest]

AMD's Fusion Developer Summit is being held in Bellevue, Wash. June 11 – 14, 2012

AMD Fusion Developer Summit Starts This June : Read more

 

[Parsian]

This is something that AMD needed to do long ago ever since investing in ATi. Their platform in my opinion is better than CUDA. Yet nVIDIA pushed CUDA everywhere while AMD sit and sank. I hope they can succeed and we can see more OpenCL GPGPU.

 

[sayantan]

I don't think so..Their 7900 series is three times faster than 680 in certain GPGPU computation. In gaming they are not far behind too..they just need to drop the prices..

 

[esrever]

I really hope AMD can make openCL computing something that is used in almost all everyday tasks. They have a strong platform and by leveraging out to openCL, they can fight intel on their own terms. How fast is an i7 when the 6550D in the A3850 can beat it in many compute heavy tasks if given that chance?

 

[obsama1]

OpenCL > CUDA. CUDA is good, but the only reason it's used is because Nvidia bribes Adobe and Autodesk to implement it. In bitcoin mining, AMD >>>> Nvidia. But for gaming, 680 >>>> 7970.

 

[fyasko]

as far back as the radeon 9600 amd/ati cards have been faster than nvidia but nvidia's drivers have always been better. open CL looks promising, if it takes off like it should i will jump back over to AMD gpu's.

 

[doron]

[citation][nom]frozonic[/nom]i really dont care what AMD does, i just want it to be competiive againg in both GPU and CPU, their GPUs havent done well since the gtx 4XX series were introduced, at least in the high end market....[/citation]

Then why are you reading this? Head straight to Anandtech's bench utility and be done with it.

[citation][nom]sayantan[/nom]I don't think so..Their 7900 series is three times faster than 680 in certain GPGPU computation. In gaming they are not far behind too..they just need to drop the prices..[/citation]

Peak freq. - 7970 and 680 trade blows in gaming, 7970 better than 680 (and 580) in compute.
Size - 680 smaller than 79xx and larger than 78xx
arrival - 7970 paper launched in december, launched in january. 680 MIA.

The way I see it, despite the "gtx680 was supposed to be 660ti" rumors, AMD and Nvidia keep leap-frogging each other since the hd4xxx series.

 

[apple_eater]

only time will tell

 

[zooted]

[citation][nom]frozonic[/nom]i really dont care what AMD does, i just want it to be competiive againg in both GPU and CPU, their GPUs havent done well since the gtx 4XX series were introduced, at least in the high end market....[/citation]
Can't tell if serious....

 

[doron]

[citation][nom]obsama1[/nom]OpenCL > CUDA. CUDA is good, but the only reason it's used is because Nvidia bribes Adobe and Autodesk to implement it. In bitcoin mining, AMD >>>> Nvidia. But for gaming, 680 >>>> 7970.[/citation]

When the gtx680 will be out for real AMD will have to lower the prices and its partners will likely introduce an overclocked version which should be on par with the 680.

However, since the 680 is largely out of stock this makes the 7970 the only choice for people who need a high end gaming machine right now and the gtx680 irrelevant.

 

[gilgamex]

[citation][nom]apple_eater[/nom]only time will tell[/citation]

You must also be clear by how much better AMD's chips are at computing, it is not even remotely trading blows with Nvidia like they often do on the gaming front.

 

[gilgamex]

^^^ignore my previous post I quoted the wrong person.

[citation][nom]doron[/nom]Then why are you reading this? Head straight to Anandtech's bench utility and be done with it.Peak freq. - 7970 and 680 trade blows in gaming, 7970 better than 680 (and 580) in compute. Size - 680 smaller than 79xx and larger than 78xxarrival - 7970 paper launched in december, launched in january. 680 MIA.The way I see it, despite the "gtx680 was supposed to be 660ti" rumors, AMD and Nvidia keep leap-frogging each other since the hd4xxx series.[/citation]

You must also be clear by how much better AMD's chips are at computing, it is not even remotely trading blows with Nvidia like they often do on the gaming front.

 

[doron]

[citation][nom]gilgamex[/nom]^^^ignore my previous post I quoted the wrong person.You must also be clear by how much better AMD's chips are at computing, it is not even remotely trading blows with Nvidia like they often do on the gaming front.[/citation]

What's really going to be interesting is what compute performance will be like when Nvidia will release their "flagship" gpu, which is rumored to have high compute performance. Can't wait for the FireGL and Quadro refresh

 

[jimmysmitty]

[citation][nom]sayantan[/nom]I don't think so..Their 7900 series is three times faster than 680 in certain GPGPU computation. In gaming they are not far behind too..they just need to drop the prices..[/citation]

Thats only because nVidia purposley disabled GPGPU computations in the GTX6 series to be able to better sell their Quadro GPUs which are built for that arena.

 

[blazorthon]

[citation][nom]jimmysmitty[/nom]Thats only because nVidia purposley disabled GPGPU computations in the GTX6 series to be able to better sell their Quadro GPUs which are built for that arena.[/citation]

Actually, the Kepler gaming cards have hardware problems that cause their low GPGPU performance, not Nvidia lowering it artificially. Kepler has a much simpler scheduler that hampers GPGPU performance compared to the more complex schedulers of Fermi and GCN. Kepler's shaders are also simpler. The Kepler cads have an additional set of DP shaders that are only active for DP compute workloads. Quadro Kepler could be made entirely out of these different cores that have equal DP and SP compute performance.

Nvidia did this to decrease gaming power usage and the size of the die for the gaming Kepler cards. Compute is not disabled, it is just slow on Kepler cards. For example, the DP compute performance of the GTX 680 is about half of the GTX 580 (very slightly more than half), so it's probably comparable to the GTX 460. For single precision (the regular shaders usually used for gaming can do this), the 680 is only about 2/3 of a 7970 in compute performance. The 7970 is about 5.5 times faster than the 680 for DP compute. At least, that is what Tom's and Anand say. I did not read any reviews from anyone else yet on the 680 and 7970 for compute.

 

[Zingam_Duo]

[citation][nom]obsama1[/nom]OpenCL > CUDA. CUDA is good, but the only reason it's used is because Nvidia bribes Adobe and Autodesk to implement it. In bitcoin mining, AMD >>>> Nvidia. But for gaming, 680 >>>> 7970.[/citation]

I am developing game GPU that will be 100000000x faster than 680 but it will be available in 300 years. So NVIDIA sucks! 😀

Dude, if it is not out, it is not better!

 

[Zingam_Duo]

[citation][nom]Parsian[/nom]This is something that AMD needed to do long ago ever since investing in ATi. Their platform in my opinion is better than CUDA. Yet nVIDIA pushed CUDA everywhere while AMD sit and sank. I hope they can succeed and we can see more OpenCL GPGPU.[/citation]


Why is AMD a better platform? Their tools are barely existing.... etc? What do you mean?

 

[eddieroolz]

CUDA is a very good implementation. I'd rather hedge my bets on a proper, mature technology.