I am still considering the Titan. However, it is not clear to me it will do much more then two times a GTX 660 on twice the power. Here is my thinking so far. First, all of my five different Nvidia cards draw about 75% of power listed on the NVIDIA site. Further, gpu benchmarks have roughly approximated actual production. So:
Cost Difference. Well, the cost of the Titan does not make me feel actually faint, but if two 660 can approximate the Titan I won't bother. Remember, my goal is 35watt/10k stones, but might settle at 40 or 45watt. My test 660 is at the post office and so we shall know in a few days.
Then there is the matter of my already purchased gtx 650's. They benchmark at 1,880 and produce a 500 point BRP in about 3,000 seconds at a measured draw of 35 watts (plus PC overhead). Thats more then 14k stones/day both by calculation and actual observation. The machine I ordered with seven card slots will mount all four. It remains to be seen if the two core 35watt 2.3ghz processor can feed them all. Plus there is some question as to whether more then two 650's can work together.
Anyway, lets assume the best. 56k stones @35watt x 4 + 35watt processor = 175 watt total. That's 10k stones/31.25 watts. But I am now skeptical about the build. First, because the builder has not said to me "come and get it". In addition, various info you and others have provided cast doubt on the project. But it is an experimental project. What fun are sequential experiments without a failure or two. Sort of like when I got all optimistic about the failed gt 620s I got based on the success of the 610's!
Arecibo 19 Oct 2012
Just Because The Space Alien Is Green
Does Not Mean You Should Go
The number of cuda cores has not proven as good at predicting production as GPU Benchmark. For instance, the 620 has twice the cores as the 610 [48/96] but produces the same; but only half of them worked at all. Further, my GTX 650 has twice the cores as my GTS 450 [384/192] and they both produce a BRP in about 3,000 seconds. And the 450 draws way more power.
GPU bus 64 v 128 etc has a lot to do with it. But I have not got all mathematical enough to calculate differentials over the entire nvidia product line.
Arecibo 19 Oct 2012
Just Because The Space Alien Is Green
Does Not Mean You Should Go
IF you were willing to use even more power then the 768 cuda cores of a $169 Nvidia 650 could be easily outshone by a $999 Nvidia Ttan with 2688 cuda cores! Not exactly cost effective but it would produce ALOT more work in ALOT less time! If you were to get an AMD 7970 you could do so for #399 and get 2048 stream processors, which is alot more cost effective by comparison. I believe cuda cores and stream processors are the same measurement using different words.
Not to mention that the 7970 is considerably faster than the Titan at this project and costs $600 less. The red team wins this shootout. BTW as a disclaimer, I run both NV and AMD cards. Right now the NVs are all on GPUGrid. The AMDs are split between Einstein and WCG, except for the 4770s which are now at SETI as they only support OpenCL v1.0.
I am still considering the Titan. However, it is not clear to me it will do much more then two times a GTX 660 on twice the power. Here is my thinking so far. First, all of my five different Nvidia cards draw about 75% of power listed on the NVIDIA site. Further, gpu benchmarks have roughly approximated actual production. So:
Cost Difference. Well, the cost of the Titan does not make me feel actually faint, but if two 660 can approximate the Titan I won't bother. Remember, my goal is 35watt/10k stones, but might settle at 40 or 45watt. My test 660 is at the post office and so we shall know in a few days.
Then there is the matter of my already purchased gtx 650's. They benchmark at 1,880 and produce a 500 point BRP in about 3,000 seconds at a measured draw of 35 watts (plus PC overhead). Thats more then 14k stones/day both by calculation and actual observation. The machine I ordered with seven card slots will mount all four. It remains to be seen if the two core 35watt 2.3ghz processor can feed them all. Plus there is some question as to whether more then two 650's can work together.
Anyway, lets assume the best. 56k stones @35watt x 4 + 35watt processor = 175 watt total. That's 10k stones/31.25 watts. But I am now skeptical about the build. First, because the builder has not said to me "come and get it". In addition, various info you and others have provided cast doubt on the project. But it is an experimental project. What fun are sequential experiments without a failure or two. Sort of like when I got all optimistic about the failed gt 620s I got based on the success of the 610's!
Considering the BRP units use ALOT of the cpu in their crunching I would not expect a dual core cpu to be able to keep a quad gpu system fed at 100%, but it should do okay. Just do not try to run any cpu units on their own on it. In fact this message http://einsteinathome.org/node/196822&nowrap=true#123901 says that the new units coming will take upto 0.7% of each cpu to keep a gpu fed and happy. What I am trying to say is that it might work, but it won't be as fast as with a quad or more core cpu.
The only reason I am now dedicated to nvidia is I am familiar with many of their cards. Eventually I may try the other guys. The trick is nvidia has the lock on the term 'cuda'. Beginners like me hear 'cudda' task. Its a good trick.
Arecibo 19 Oct 2012
Just Because The Space Alien Is Green
Does Not Mean You Should Go
First and formost is that my hp core duo can not feed the new GTX660 any better then the GtX 650. But at 80watts unstead if 35watts. Reminds me of my GT610/620 experience. The 620 has twice the cuda cores, draws lots more power, and, when they work, don't do any more then the 610. So, a single 650 or 660 in THIS machine produce the same results. But the 660 draws more then twice the juice. Time to punt.
Second, my builder confirmed the new build can only carry two GTX 650's, not the four we originally fitted; even though the Express Lanes have added power, they can only power two of the 650's. But the good news is the machine only draws 100 watts total with dual 2.3 processors and the two 650. So IF the dual processors can feed each 650 with 24-28 BRP tasks per day, that would still yield 24-28k stones per day. Equals 10k@ 36-42watt/10k stones.
But might need a bigger processor and 30 more watts. Lots of fun going on in Newbie Town!
Arecibo 19 Oct 2012
Just Because The Space Alien Is Green
Does Not Mean You Should Go
I am still considering the
)
I am still considering the Titan. However, it is not clear to me it will do much more then two times a GTX 660 on twice the power. Here is my thinking so far. First, all of my five different Nvidia cards draw about 75% of power listed on the NVIDIA site. Further, gpu benchmarks have roughly approximated actual production. So:
GTX 660 Benchmark 4,073 140watts x 2 = benchmark 8,146 @ 280watts
Titan Benchmark 8,400 250watts
Cost Difference. Well, the cost of the Titan does not make me feel actually faint, but if two 660 can approximate the Titan I won't bother. Remember, my goal is 35watt/10k stones, but might settle at 40 or 45watt. My test 660 is at the post office and so we shall know in a few days.
Then there is the matter of my already purchased gtx 650's. They benchmark at 1,880 and produce a 500 point BRP in about 3,000 seconds at a measured draw of 35 watts (plus PC overhead). Thats more then 14k stones/day both by calculation and actual observation. The machine I ordered with seven card slots will mount all four. It remains to be seen if the two core 35watt 2.3ghz processor can feed them all. Plus there is some question as to whether more then two 650's can work together.
Anyway, lets assume the best. 56k stones @35watt x 4 + 35watt processor = 175 watt total. That's 10k stones/31.25 watts. But I am now skeptical about the build. First, because the builder has not said to me "come and get it". In addition, various info you and others have provided cast doubt on the project. But it is an experimental project. What fun are sequential experiments without a failure or two. Sort of like when I got all optimistic about the failed gt 620s I got based on the success of the 610's!
Arecibo 19 Oct 2012
Just Because The Space Alien Is Green
Does Not Mean You Should Go
PS The number of cuda
)
PS
The number of cuda cores has not proven as good at predicting production as GPU Benchmark. For instance, the 620 has twice the cores as the 610 [48/96] but produces the same; but only half of them worked at all. Further, my GTX 650 has twice the cores as my GTS 450 [384/192] and they both produce a BRP in about 3,000 seconds. And the 450 draws way more power.
GPU bus 64 v 128 etc has a lot to do with it. But I have not got all mathematical enough to calculate differentials over the entire nvidia product line.
Arecibo 19 Oct 2012
Just Because The Space Alien Is Green
Does Not Mean You Should Go
RE: IF you were willing to
)
Not to mention that the 7970 is considerably faster than the Titan at this project and costs $600 less. The red team wins this shootout. BTW as a disclaimer, I run both NV and AMD cards. Right now the NVs are all on GPUGrid. The AMDs are split between Einstein and WCG, except for the 4770s which are now at SETI as they only support OpenCL v1.0.
RE: I am still considering
)
Considering the BRP units use ALOT of the cpu in their crunching I would not expect a dual core cpu to be able to keep a quad gpu system fed at 100%, but it should do okay. Just do not try to run any cpu units on their own on it. In fact this message http://einsteinathome.org/node/196822&nowrap=true#123901 says that the new units coming will take upto 0.7% of each cpu to keep a gpu fed and happy. What I am trying to say is that it might work, but it won't be as fast as with a quad or more core cpu.
Beyond, The only reason I
)
Beyond,
The only reason I am now dedicated to nvidia is I am familiar with many of their cards. Eventually I may try the other guys. The trick is nvidia has the lock on the term 'cuda'. Beginners like me hear 'cudda' task. Its a good trick.
Arecibo 19 Oct 2012
Just Because The Space Alien Is Green
Does Not Mean You Should Go
Results to report! First
)
Results to report!
First and formost is that my hp core duo can not feed the new GTX660 any better then the GtX 650. But at 80watts unstead if 35watts. Reminds me of my GT610/620 experience. The 620 has twice the cuda cores, draws lots more power, and, when they work, don't do any more then the 610. So, a single 650 or 660 in THIS machine produce the same results. But the 660 draws more then twice the juice. Time to punt.
Second, my builder confirmed the new build can only carry two GTX 650's, not the four we originally fitted; even though the Express Lanes have added power, they can only power two of the 650's. But the good news is the machine only draws 100 watts total with dual 2.3 processors and the two 650. So IF the dual processors can feed each 650 with 24-28 BRP tasks per day, that would still yield 24-28k stones per day. Equals 10k@ 36-42watt/10k stones.
But might need a bigger processor and 30 more watts. Lots of fun going on in Newbie Town!
Arecibo 19 Oct 2012
Just Because The Space Alien Is Green
Does Not Mean You Should Go
How many tasks are you trying
)
How many tasks are you trying to run at once on the 650 & 660???
Those numbers are indicative
)
Those numbers are indicative of one task per GPU.