You are talking for games yes? - For CUDA these 'professionals' probably
haven't got a clue what they are talking about. The Titan is an awesome
card and massively beats the 690 in terms of single GPU performance.
111ns/day for JAC NVE plays about 67ns/day for 690. Granted you can run 2
calculations at once on a 690 but you can't put 4 of them in a box for
power reasons so I think most people would agree that 4 GTX Titan's is
likely the optimum option. Put 4 of these in a homebuilt box all for less
than $6K and you are golden. I'll be posting the shopping list for
building such a system on the AMBER webpage shortly.
All the best
Ross
On 3/18/13 11:29 AM, "Albert" <mailmd2011.gmail.com> wrote:
>it looks good.
>But As far as I know, it is no better than GTX690 according to some
>professional computer journal testing.
>
>
>
>
>
>On 03/18/2013 06:58 PM, filip fratev wrote:
>> Hi Ross and Mohd,
>>
>>
>>> >Hi, the clock speeds that you reported, are those with full double
>>>precision on? Or off? From
>>>thishttp://www.anandtech.com/show/6760/nvidias-geforce-gtx-titan-part-1/
>>>4 article, I get the impression that the boost 2.0 or what ever they
>>>>call it, is disabled if full DP is enabled.So, can you please check,
>>>>that when you report the higher clock under
>> windows, is this with full DP or with DP disabled. And can you please
>> also check if under Linux, DP can be disabled, say through NVIDIA
>> settings?
>>
>> I reported the clocks with DP
>> disabled (the default mode). You can see the CUDA-Z output here:
>> http://imageshack.us/photo/my-images/259/cudaz.png/
>>
>>
>> DP mode should be and is in
>> reality restricted to only 836MHz:
>> http://imageshack.us/photo/my-images/26/cudazdp.png/
>>
>> About DP mode under Linux I will check that today.
>>
>>
>>> >However, this appears to be for gaming, have you managed to run any
>>>cuda code under Windows that can use the GPU running at that clock
>>>speed?
>> Ross this is very interesting
>> question! I tested Hex_cuda and Molegro MVD (cuda docking software's).
>>Both Hex
>> and MVD work on 836MHz:
>> http://imageshack.us/photo/my-images/835/mvda.gif/
>>
>>
>> However, the GPU load was very
>> small, around 30-55%, thus I tested if I have more realistic load what
>>would be
>> the core speed and execute both programs simultaneously. Also tested
>>what would
>> happen if I run two MVD jobs on same GPU. The load increased to 75% and
>>the speed
>> immediately goes to 993MHz! The result is here:
>> http://imageshack.us/photo/my-images/13/2procesa1.gif/
>>
>> These results are also supported
>> by CUDA-Z image above. Thus the answer is that definitely the Cuda
>>applications
>> work under windows with 993MHz core speed.
>>
>> My report is just informative. My
>> opinion is that rather Nvidia developers should be pressured by some
>>serious
>> institutions to unlock their cards under Linux than a new windows amber
>> modification to be developed..
>>
>> All the best,
>> Filip
>>
>
>
>_______________________________________________
>AMBER mailing list
>AMBER.ambermd.org
>http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Mar 18 2013 - 16:00:04 PDT
