AMD: R7xx Speculation

[jimmyjames123]

AMD says its 4850 device at about 110 W and $199 will deliver about 75 percent of the performance of Nvidia's high-end GTX280 which costs $649 and dissipates 236W. Two of the AMD parts on a board will hit graphics benchmarks about 30 percent higher than the Nvidia device, the AMD spokesman added.

Oh this is priceless. The GTX 280 will have only 33% higher performance than the 4850? lol, dream on, he looks like he was smoking something hallucinogenic I might as well argue that the GTX 280 is only 33% faster than 9800 GTX! I suppose I could find a situation where that is true, but it's not very realistic in newer games at ultra high res and/or high AA.

 

[jimmyjames123]

I'm sure those numbers are skewed towards RV770 but they look damn impressive!

Someone at NVIDIA is crying right now..

The performance per watt and performance per mm^2 comparison there is quite misleading. Performance based on what game and what setting? Watts based on idle or load power consumption?

The GTX 280 would actually have a higher performance per watt than RV770 when looking at ultra high res with very high AA and using idle power consumption.

The RV770-based cards are going to be very nice all around cards, but some of these AMD and NVIDIA marketers can really get carried away

 

[AlphaWolf]

The GTX 280 would actually have a higher performance per watt than RV770 when looking at ultra high res with very high AA and using idle power consumption.

How the hell is a game at any resolution while the gpu is idle?

 

[jimmyjames123]

How the hell is a game at any resolution while the gpu is idle?

Considering that graphics card are essentially idle most of the time, why would one want to only look at performance per watt based on load power consumption, and then give no consideration to idle power consumption? Delivering higher performance with relatively low idle power consumption is a more sensible thing to look at. Point being that performance per watt is a meaningless measure without looking at idle vs load power consumption too.

 

[Ninjaprime]

The performance per watt and performance per mm^2 comparison there is quite misleading. Performance based on what game and what setting? Watts based on idle or load power consumption?

The GTX 280 would actually have a higher performance per watt than RV770 when looking at ultra high res with very high AA and using idle power consumption.

The RV770-based cards are going to be very nice all around cards, but some of these AMD and NVIDIA marketers can really get carried away

Its pretty simple:

4850 = 1 Tflops on 250mm².
GTX 280 = 933 Gflops on 576mm².

Difference = about 2.8.

4850 = 150w card with 1Tflops.
GTX 280 = 300w card with 933 Gflops.

Difference = more than 2.0, so ATI is probably selling themselves short.

 

[jimmyjames123]

Its pretty simple:

4850 = 1 Tflops on 250mm².
GTX 280 = 933 Gflops on 576mm².

Difference = about 2.8.

4850 = 150w card with 1Tflops.
GTX 280 = 300w card with 933 Gflops.

Difference = more than 2.0, so ATI is probably selling themselves short.

So we are back to the assumption that #Flops = performance [for games]? Ok

 

[AnarchX]

4850 = 150w card with 1Tflops.
GTX 280 = 300w card with 933 Gflops.
t.

GTX 280s TDP is 236W, HD4850 should be ~150W, since normal usage seems around 100-110W, as some tests show (load was higher than on GTS-512).

 

[Ninjaprime]

So we are back to the assumption that #Flops = performance [for games]? Ok

I dont see anywhere its claimed "in games." It just says performance. Flops is a type of performance. Go stroke your GTX 280, you'll feel better.

 

[AlphaWolf]

Considering that graphics card are essentially idle most of the time, why would one want to only look at performance per watt based on load power consumption, and then give no consideration to idle power consumption? Delivering higher performance with relatively low idle power consumption is a more sensible thing to look at. Point being that performance per watt is a meaningless measure without looking at idle vs load power consumption too.

Well the performance of a graphics card while idle is zero, but I bet you the 4850 uses less power than the 280 while it's idle also, so it'll still be more efficient there.

 

[Sunday]

Wirmish fire up your calc

 

[ZerazaX]

Interesting... look closely at the RV770 die shot. I see 6 rows of 16 processors. 480 SP confirmed? Or am I just seeing things?

I think you are just seeing things... I see where you can get 6 rows but definitely not 16 processors from that view. Besides, the GTX280 die shot isn't correct so I don't know how much those die shots are just mockups.

 

[Ninjaprime]

I think you are just seeing things... I see where you can get 6 rows but definitely not 16 processors from that view. Besides, the GTX280 die shot isn't correct so I don't know how much those die shots are just mockups.

Yeah its hard to tell being so small of a picture. It could be they are back to back like in Xenos, which would mean 5 of 32? I'm not sure if that first one looks the same, might be something else, 5x32 would be 800 so that fits. Or they could be side by side and the white space is the interconnects between them, which could make it the 10x16 rumored a page or two back. It still looks like six rows to me though...

Too small to tell really, need a higher rez shot.

 

[Ninjaprime]

Hmm on closer inspection, I can only really see 10 dark spots in a row, and 5 rows. Could it be 5x20 for 500? Or 10x10? I'm probably overanalyzing.

 

[ZerazaX]

My numbers keep changing everytime too

 

[Psycho]

Wirmish fire up your calc

Not very useful without some kind of known measure, be it a ruler, the pcie slot, mounting hole distance etc. Btw. my old pcie slot based calculation was off because of a wrong length of the connector (it's really 85mm), and with this number and no correction term I still get 276mm2

 

[DegustatoR]

Someone at NVIDIA is crying right now..

Not really.
But i do continue to wonder how they've packed all that into such small chip.
It almost looks like something isn't right here...

 

[Lukfi]

no-X says the "sampler units" that used to be part of the TUs are not there.

 

[Shadowmage]

Not really.
But i do continue to wonder how they've packed all that into such small chip.
It almost looks like something isn't right here...

My guess is that they did the shaders with custom layout, or perhaps.... Fast14?

 

[Jawed]

This is incredible -> Short video

This is real-time rendered !

Where's that :drool: smiley? I'm really looking forward to seeing the whole thing

Jawed

 

[Arun]

My first estimate based on that die shot, if it is indeed a RV770 die shot, is that AMD is the first to a 2GHz shader clock with the HD4870...