980 TI Bench
#11
Ivan Golubev had a comprehensive list of GPUs and their effectiveness, but his domain is parked, so I consider that info beyond reach.
Based on limited, but practical info provided by Kartan, here are the values:
MD5 performance: 16.92M p/s / $ for Titan X and 21.181 p/s / $ for 980 Ti.
However, I can't proclaim 980 Ti king, as I haven't looked through all the GPUs and compared them.
Reply
#12
The 290X makes over 35M p/s / $ in MD5. Not too shabby for that old stick.
Reply
#13
(06-06-2015, 06:41 PM)Flomac Wrote: The 290X makes over 35M p/s / $ in MD5. Not too shabby for that old stick.
According to Jeremi's benchmarks, 290X has 18.57M p/s / $.
You probably used current price and not the release MSRP, that's why your number is so high.
Reply
#14
Ok, let's math:

Perf/$ (MSRP)

R9 290X: 18.67M
GTX 970: 30.8M
GTX 980: 23.5M
GTX 980Ti: 30.3M
GTX Titan X: 18.9M

Perf/Watt

R9 290X: 34.2M
GTX 970: 72M
GTX 980: 78.3M
GTX 980Ti: 73.9M
GTX Titan X: 80.2M
Reply
#15
That's better.

980 Ti is close to 970 by effective performance, but GTX Titan X is substantially better than 980 Ti (980 gets the silver medal).

'course, this is using actual TDP vs the one manufacturers provided, which has been gotten using means known only to them.
I suspect the actual TDP can float on certain algos (blowfishcrypt vs NTLM, for example) and that introduces extra complexity, unless we're talking about MD5 and MD5 only.
Reply
#16
Usually on register-bound algorithms like MD4 and MD5 the card runs close to TDP (or over TDP in the case of AMD, which is why we have to set Powertune to +50%.) Titan X for example draws 235W at +200Mhz which is pretty close to its 250W TDP, but only draws 125W on bcrypt.
Reply
#17
(06-07-2015, 05:34 AM)Rolf Wrote:
(06-06-2015, 06:41 PM)Flomac Wrote: The 290X makes over 35M p/s / $ in MD5. Not too shabby for that old stick.
According to Jeremi's benchmarks, 290X has 18.57M p/s / $.
You probably used current price and not the release MSRP, that's why your number is so high.

Ok, but who cares about MSRP? These comparisons are very useful for actual buyers. And they don't pay MSRP. So street price is everything that count and yes, that makes them harder to compare. But far more useful.

Same with TDP, which i.e. is only 90% of power a GTX980 needs to fullfil a steady max clocking, so real world full speed TDP is more 180W. 980Ti and Titan X share the same TDP and it seems they operate close to TDP in gaming and computing, but don't quite touch it.

I'm still thinking if I switch my GTX980 for the Ti version, since the latter is more noisy and I might not like that for some reasons. Maybe I just give it a try and decide it on my experiences.
Reply
#18
MSRP and TDP are fixed variables, and Ivan Golubev's original tables used MSRP.
Reply
#19
blx, which driver version did you use for this card? i keep getting ERROR: cuMemAllocHost() 2 when i try to run a benchmark.
Reply
#20
346 is the stable branch, so 346.82 would be the recommended driver.
Reply