Geforce GTX 260 Core 216

Naturally, especially since it's speedy GDDR5.
I was just adding that since he said the 4870 falls behind because of the 512MB framebuffer

 

Yeah fair enough, prices will certainly vary from country to country and until we start seeing some decent availability of the 216sp variant I don't expect its street prices to match the recommendations from NV.

Give it a month though and I expect we could be seeing some amazing value from this part.

 

Of course. My point was simply that if you increase the price of the 4870, you lose any pricing advantage over a GTX 260 part

 

Ah, I thought you meant people were expecting this to be faster than the 280. In the rare cases were it is, its because they are testing heavily overclocked variants. At stock there is no way this card can be faster than the 280.

I think you might have misinterpreted their results. Here is how the 260 216sp faired against the HD4780 in anands tests:

Age of Conan - Lose at all resolutions
Crysis - Win at all resolutions
Quake Wars - win at 2560x1600 only
Oblivion - Win at all resolutions
GRID - Lose (by a lot!) in all resolutions
Assassins Creed - Lose at 2560x1600 (only test performed)
The Witcher - Win at 2560x1600 (only test performed)

Overall thats pretty much a tie unless you take into account 2560x1600 in which case the 260 wins by one benchmark.

 

That's good to hear -- was wondering about that!

 

Erm, aren't you just clarifying my point?

 

But will the entire core of the GTX 260 216 work, or will it disable a SIMD core to run in tandem with the regular 260?

 

I suspect the placement of the "primary" graphics adapter would determine the answer to your question. IOW, if you put the 260-216 part in the primary PEG slot it should have full functionality since it would be the "master" card in the SLI configuration. If you were to slave it to an original 260 though, I could see the extra SIMD being disabled by the driver.

 

I can't see that being a good solution for a performance set up. One frame renders faster than the other? Wouldn't you get a smoother experience by having 2 cards the same? I don't think you would gain in terms of performance over 2 vanilla 260s (which would save you money as well). But I'd be willing to look at some numbers if anyone has any with this set up.

 

Welcome to the world of AFR. Uneven frame rendering times are the norm.

 

Yes, but would an enthusiast do this with intent? I'm just suggesting that buying a Core 216 to put into SLI with an original 260 might not be the best investment. At least not while the base 260s are readily available and significantly cheaper.

 

I think the reason they don't rename things sometimes is because a new name makes interest in the old product wane. So there is not necessarily anything for NV to gain by changing the name. The performance isn't THAT different anyway.

 

Right, which I guess would make sense since it has more memory and faster/more expensive memory.
Just wait for sites to offer the 1GB cards with rebates I guess

 

No. You said the 4870 is faster up to 1920x1200 and then falls behind at 2560x1600.

In only one of the 7 tests does that occur.

In 2 tests the 260 wins everything including lower resolutions and in another 2 tests it loses everything including 2560x1600.

In the final two tests which only test 2560x1600, the 260 wins one and loses one. Apart from the Quake Wars example the pattern you mention doesn't seem to exist (at least in those benchmarks).

 

Wouldn't that be comparing rebate prices with none rebate prices?

Without taking rebates into account for either card, the stock 216 should go for the same price as the 512MB 4870. A 1GB 4870 would put you into o/c territory price wise which is approaching 280 performance.

 

If you can get either with a rebate you shouldnt take that out of the equation just because you cant get a rebate for one of them.

 

And to answer the question. ALL SP/TMUS are utilized. Even in a mixed setup. Since the performance variance between the two is not huge ((10% roughly)) AFR works pretty much the same. I have a 216 SP card coming and I intend to investigate the performance variables of the old 260 with the new ones in dual/3way SLI setups. The uneven frame distribution shouldnt be that much of a problem with the way cards perform. But it's something I intend to look at it. Its alot more even than say "Geforce Boost" feature of Hybrid SLI. ((Which is not my favorite solution due to much larger uneven AFR frame synchronization))

Chris

 

Thanks for the info, Chris. Do you know if my hypothesis about slot placement is correct?

 

The HD4870 also incurs a proportionately larger performance drop going from 1920x1200 to 2560x1600 compared to the GTX 260/280 (from other sources). I think, for 2560x1600, your best bet would be to either get the GTX 260 216, HD4870 1GB, HD4850 X2 (TBA) or HD4870 X2.

 

I think the issue is stemming from where you said:

"the HD4870 is still ahead of the 216 up to 1920x1200"

Because its not. By Anands reckoning its dead even (in terms of the number of benchmarks won).

You then said:

"2560x1600, in most titles, seems too much for the HD4870's 512MB memory pool"

But again, this isn't true because the 4870 wins 3 out of the 7 2560x1600 tests. In only one does it fall behind at the higher res were it was ahead at the lowe res. I don't think thats enough evidence to base such a conclusion on.

I take on board that its taking a bigger hit than the 260 when moving to that res whether it wins or not but that could just as easily be down to a lower fill rate. I think when you run out of framebuffer the drop in performance is generally much sharper than what we're seeing here.

All that said though, if your going to be gaming at that res I don't think either GPU is suitable. Your probably best off with an X2 or an SLI config.