Not a Troll Thread: HD7870 LE vs GTX680

[Borealis7]

TechPowerUp has posted a review of the TVX3D HD7870 Black Edition which got me thinking:

http://www.techpowerup.com/reviews/VTX3D/Radeon_HD_7870_XT_Black/1.html

7870 LE (Tahiti - not Pitcairn) has: 1536 shaders, 32 ROPs, 4310M Xtors, 2GB VRAM, 256Bit bus, ~1GHz core clock, 6GHz GDDR5.
GTX680 (GK104) has: 1536 shaders, 32 ROPs, 3500M Xtors, 2GB VRAM, 256Bit bus, 1.1GHz core clock 6GHz GDDR5.

see where i'm getting at? ---> time for some Architectural Comparisons!

lets get down to the bottom line: http://tpucdn.com/reviews/VTX3D/Radeon_HD_7870_XT_Black/images/perfrel_1920.gif

at 1920X1200, 4AA: the GTX680 is 27% faster than the 7870LE, but it also enjoys a 14% core clock speed advantage (accounting for NV dynamic boost, this gets tricky).

so i would think that an NV Shader is, shader for shader & clock for clock, 1.27/1.14 = 11% "stronger" architecturally than an AMD shader at this configuration. plus, the GK104 does that while having 19% less transistors to work with.

lets see if we can extrapolate this to other cards:
7870LE vs GTX670 : 7870LE has 14% more shaders and 6% core clock advantage. the GTX670 is 17% stronger so: 1.17/0.875/0.93 = 10% close enough!

i'm onto something here, aren't I?
of course i'm aware than raw performance is the product of the shaders, rops, drivers, memory management and whole bunch of other stuff but i think it's always cool to have a common ground for comparison.

 

[3DVagabond]

I think the real measurement for efficiency of a design is perf/Watt. Number of shaders, etc. is really a bit arbitrary because of different architectures. I would also take more than gaming performance into consideration if I'm going to try and show which design is "stronger". I do realize that we here are mostly concerned with gaming only.

 

[DominionSeraph]

I think the real measurement for efficiency of a design is perf/Watt.

Then I think Intel's integrated graphics wins.

 

[SPBHM]

you can't compare transistors count, when a big portion is disabled for the 7870,
also the efficiency can be related to other aspects and not just shader,

 

[Borealis7]

do we know how many active Xtors are in a 7870 chip?

 

[parvadomus]

Why not use pitcairn, a pure gaming GPU just like gk104? and scale all metrics based on performance?
I bet at same clocks (turbo disabled, at 1ghz) gk104 is only 25% faster than pitcairn.
PC has 2800m transistors, gk 104 3540m, about 25%.
Architectures are very very close. However pitcairn still destroys GK104 at gpgpu, so...

 

[Borealis7]

AMD wanted a stop-gap card to challenge the GTX660TI, and the 7870 wasn't good enough, so they took a defective ("presumably") 7950 and called it a "limited edition".

 

[Haserath]

One difference between them, on a high level, seems to be the amount of TMUs. However much difference that makes.

 

[Lonyo]

AMD wanted a stop-gap card to challenge the GTX660TI, and the 7870 wasn't good enough, so they took a defective ("presumably") 7950 and called it a "limited edition".

They did exactly what they did with the 4000/5000 series etc, and what NV did with the 500 series, and probably others too.

And they did what they were rumoured to be doing 1 year ago, when initial specs were leaked which included this card, but then named a 7930, followed by 7890.

It's more using up old chips than a stop gap for competition purposes.

 

[Tuna-Fish]

so i would think that an NV Shader is, shader for shader & clock for clock, 1.27/1.14 = 11% "stronger" architecturally than an AMD shader at this configuration. plus, the GK104 does that while having 19% less transistors to work with.

That's not really true. The LE Tahiti has had a quarter of it's SIMD arrays and a third of it's memory bus disabled, you should account for that when comparing transistor counts. Also, GK104 has had DP capabilities stripped, so it's better to compare it to Pitcairn (which is also similarly defanged).

 

[Gikaseixas]

that has been debated before, Pitcain is a better comparison

 

[Imouto]

Way to go TPU, way to go.

 

[Borealis7]

what are you insinuating? that TPU reuses numbers without testing?

 

[raghu78]

you can't compare two different architectures like that. also the GTX 680 is a purpose designed gaming GPU with top bins while the Tahiti LE is a salvage SKU of a compute / gaming designed GPU.

 

[Imouto]

what are you insinuating? that TPU reuses numbers without testing?

I'm just saying that they didn't even question why 2 cards with the same BIOS and drivers (hell, both come from the same assembler) have so different performance with OC.