GTX260 vs 9800GTX+ (Lots of pics)

[blazer]

Well i just obtained an Inno3D GTX260 (55nm) and was wondering what the difference was between that and my old Asus 9800GTX+ here are the stats.

Benchmark performed on: Core2 8500 3.16Ghz, Gigabyte EP45-DS4P, Windows 7 RC1 64-Bit,
Drivers:185.85_desktop_win7_64bit_english_whql
Multiforcer 0.7Beta -m 500 -threads 256 -blocks 256

9800GTX+ MD5 1 Hash // GTX260 MD5 1 Hash

//

9800GTX+ MD5 Few Hashes// GTX260 MD5 Few Hashes

//

9800GTX+ MD5 Many Hashes // GTX260 MD5 Many Hashes

//

9800GTX+ NTLM 1 Hash // GTX260 NTLM 1 Hash

//

9800GTX+ NTLM Few Hashes // GTX260 NTLM Few Hashes

//

9800GTX+ NTLM Many Hashes // GTX260 NTLM Many Hashes

//

Finally some 3D Marks

3DMark 06

9800GTX + // GTX260

//

3DMark Vantage

9800GTX + // GTX260

//

=============================================================================================================================

O well i personally expected the 260 to have been much better however 3Dmark06 shows only slight improvement and using for Hash cracking only has minor improvements. I wonder why 3DVantage shows such a difference between the cards.
Also changing the --thread/blocks to 512 seem to do nothing and even decrease performance at times on the 260.

[Bitweasil]

I'm still working on optimum tuning for the GT200 cores.

One thing that makes a big difference is the clock rate.

If you look, the 9800 has 128 SPs at 1.84ghz
The GTX260 has 216 PSs at 1.24ghz

128 * 184 = 23552
216 * 124 = 26784

As cracking performance more or less scales linearly with (SPs * frequency), this is about right.

The GTX260, in this metric, is 13% faster than the 9800.

It looks like you are consistently seeing 6-7% faster performance with the GTX260 over the 9800GTX. Given that they both are having to make global memory calls, and not *everything* is compute bound, this seems like a reasonable difference given the compute difference in the two cards.

Thanks for all the benchmarks, though! I really need to get the front page updated.