GPU mining slower than CPU mining?

[nlai]

Laptop:
Core2Duo 1.8ghz
1gb RAM
1500 khash/sec on CPU
?? khash/sec on GPU (it's intel video card, doubt it can do GPU)

Desktop:
Intel® Core™ i7-870 w/VT
8GB RAM
4500 khash/sec on CPU without limit
5000 khash/sec on CPU with limit to 4 cores (why is this higher?)
3500 khash/sec on GPU (I thought GPU>CPU?

1)Why is limiting to 4 cores faster than no limit (on 8 cores)?
2)Why is GPU slower than CPU when ppl say GPU is supposed to be an order of magnitude faster?
3)Why is my Core2Duo 1/3rd as fast as my i7 when my i7 has 8 cores vs 2 cores and a lot more ram/better architecture/etc?

[nlai]

Oh and why do people get like 15K off an AMD Phenom but I only get 5K off an i7? I thought they're comparable processors?

Is it my software setup?

[nlai]

Sorry, I should've posted what GPU

It's a NVidia Geforce 310

It says it supports OpenCL and CUDA... which should I be using for mining?

[LZ]

Try puddinpop's RPC CUDA miner and his Screensaver RPC CUDA miner.

[nlai]

Try puddinpop's RPC CUDA miner and his Screensaver RPC CUDA miner.

That's what I was using to get those stats. The RPC CUDA miner. Should I be using OpenCL instead since GPU Caps Viewer says my card supports both?

[LZ]

I just read an information about your GPU. It has a very low config core. So that speed is normal for it.

http://en.wikipedia.org/wiki/Comparison_of_Nvidia_graphics_processing_units#GeForce_300_Series T_T

[Big Bear]

 If you have a high end ATI, there is a problem.

Sorry, is this a typo? Whats the problem with a high end ATI card? Or do you mean it's no problem?
 

[LZ]

No, it is not a typo. He had in mind that having a low mhps for some high end ATI card is a problem.

[nlai]

So in other words my GPU sucks If I knew about bitcoin I would've opted for a better one when I bought my computer


Now one more question... it says my GPU supports both CUDA and OpenCL.

Which should I be using for better performance?

[Bengie]

The i7 quad cores(8 threads) may times do out perform the sexacore AMD chips. I think what the issue may be here though, is each i7 core only has one 128bit SSE unit which is shared between both kernel threads.

You should try running an SSE miner on 4 threads and another 4 non-SSE threads. The i7 should be able to handle 4 sse and 4 fpu threads without contention.

The new Sandy Bridge Intel CPUs should not have this limitation as they have two 128bit units. But SB chips do support AVX which is even better than SSE. Once miners start to support AVX, then you will have the same issue with a 4 thread limit on a quad core because AVX is 256bit and it will require taking both of those 128bit units and using both at the same time. So again, only on thread per core may use both units at once. But until AVX, I would think 2 SSE threads per core on SB chips should work.

[dishwara]

With 16 cores what u get is ok.
http://en.wikipedia.org/wiki/Comparison_of_Nvidia_graphics_processing_units#GeForce_300_Series

It actually don't support opencl, even though says supports opencl 1.1
My card 9400GT has same specs & it mines only 2500 khash/s at 550MHZ & at 700MHz, mines at 3400 khash/s

So, it normal.
poclbm gives much higher rate in cuda than RPC.
Try poclbm.
http://bitcointalk.org/index.php?topic=1334.0

[LeiKriS]

No, it is not a typo. He had in mind that having a low mhps for some high end ATI card is a problem.

Can you elaborate more on this? I have a problem with my CUDA program. GPU appears to be slower than CPU. I'm using GeForce 9500 GT.

[urizane]

Desktop:
Intel® Core™ i7-870 w/VT
8GB RAM
4500 khash/sec on CPU without limit
5000 khash/sec on CPU with limit to 4 cores (why is this higher?)

Give ufasoft's SSE2 miner a go.  I have an i7-950 running at 3.83 GHz that can produce 18.9 Mhash/s.  You should probably get somewhere in the 14.5-15.8 Mhash/s range, depending on the thermals in your case.  Also, an nVidia G 310 has a really hard time post processing HD video clips, so mining isn't going to be worthwhile on this card.

AMD Processors with 4way instructions are much more efficient than intel processors.

i7 doesn't have 8 cores. It has 4 cores with hyperthreading. This has been discussed a few times if you want more info do a search for what's out there.

Disabling hashing on the hyperthreads will increase your hash rate.

AMD vs. Intel depends on the miner you run.  I know on my quad core I get roughly 16 Mhash/s at stock speeds (I'm assuming an AMD 975 BE would be somewhere in this range, as well) so be sure you use the best miner for your hardware.  In my experience, disabling HyperThreading doesn't really help when running ufasoft's miner.  Besides, if you set a miner up for 4 threads, Windows will do the job of parking the virtual threads unless another CPU heavy application suddenly needs them.

The i7 quad cores(8 threads) may times do out perform the sexacore AMD chips. I think what the issue may be here though, is each i7 core only has one 128bit SSE unit which is shared between both kernel threads.

You should try running an SSE miner on 4 threads and another 4 non-SSE threads. The i7 should be able to handle 4 sse and 4 fpu threads without contention.

Ehm...  I guess this sounds like something you can try, but I'm under the assumption that SSE2 floating point operations still utilize the FPU SSE2 ALU operations utilize the same ALUs that non-SSE2 instructions would use, so running two types of miners at once would be somewhat counter-productive.  Also, using ufasoft's miner, I got better results running with 8 threads over 4, but that could be limited to the i7-9xx series parts.

With 16 cores what u get is ok.
http://en.wikipedia.org/wiki/Comparison_of_Nvidia_graphics_processing_units#GeForce_300_Series

It actually don't support opencl, even though says supports opencl 1.1
My card 9400GT has same specs & it mines only 2500 khash/s at 550MHZ & at 700MHz, mines at 3400 khash/s

So, it normal.
poclbm gives much higher rate in cuda than RPC.
Try poclbm.
http://bitcointalk.org/index.php?topic=1334.0

Some of this seems a little off.  The nVidia G 310 is a GT200 based GPU.  It does in fact support both CUDA and OpenCL.  If you're comfortable tweaking command line parameters, I would suggest looking into puddinpop's RPC CUDA miner as well as m0mchil's Python OpenCL miner.  After tweaking paramaters, you'll probably find that, for your G 310, the CUDA miner will be about 5% faster.

Can you elaborate more on this? I have a problem with my CUDA program. GPU appears to be slower than CPU. I'm using GeForce 9500 GT.

Depending on the CPU, you're probably in the same boat.  There isn't enough processing power on one of these lower end cards to do integer operations with any significant speed.  If you have a particularly fast CPU, these GPUs aren't going to match the performance you'll get on the CPU.

EDIT: I removed references to FPUs.  Hashing and FPUs shouldn't go together.

[LeiKriS]

Desktop:
Intel® Core™ i7-870 w/VT
8GB RAM
4500 khash/sec on CPU without limit
5000 khash/sec on CPU with limit to 4 cores (why is this higher?)

Give ufasoft's SSE2 miner a go.  I have an i7-950 running at 3.83 GHz that can produce 18.9 Mhash/s.  You should probably get somewhere in the 14.5-15.8 Mhash/s range, depending on the thermals in your case.  Also, an nVidia G 310 has a really hard time post processing HD video clips, so mining isn't going to be worthwhile on this card.

AMD Processors with 4way instructions are much more efficient than intel processors.

i7 doesn't have 8 cores. It has 4 cores with hyperthreading. This has been discussed a few times if you want more info do a search for what's out there.

Disabling hashing on the hyperthreads will increase your hash rate.

AMD vs. Intel depends on the miner you run.  I know on my quad core I get roughly 16 Mhash/s at stock speeds (I'm assuming an AMD 975 BE would be somewhere in this range, as well) so be sure you use the best miner for your hardware.  In my experience, disabling HyperThreading doesn't really help when running ufasoft's miner.  Besides, if you set a miner up for 4 threads, Windows will do the job of parking the virtual threads unless another CPU heavy application suddenly needs them.

The i7 quad cores(8 threads) may times do out perform the sexacore AMD chips. I think what the issue may be here though, is each i7 core only has one 128bit SSE unit which is shared between both kernel threads.

You should try running an SSE miner on 4 threads and another 4 non-SSE threads. The i7 should be able to handle 4 sse and 4 fpu threads without contention.

Ehm...  I guess this sounds like something you can try, but I'm under the assumption that SSE2 floating point operations still utilize the FPU SSE2 ALU operations utilize the same ALUs that non-SSE2 instructions would use, so running two types of miners at once would be somewhat counter-productive.  Also, using ufasoft's miner, I got better results running with 8 threads over 4, but that could be limited to the i7-9xx series parts.

With 16 cores what u get is ok.
http://en.wikipedia.org/wiki/Comparison_of_Nvidia_graphics_processing_units#GeForce_300_Series

It actually don't support opencl, even though says supports opencl 1.1
My card 9400GT has same specs & it mines only 2500 khash/s at 550MHZ & at 700MHz, mines at 3400 khash/s

So, it normal.
poclbm gives much higher rate in cuda than RPC.
Try poclbm.
http://bitcointalk.org/index.php?topic=1334.0

Some of this seems a little off.  The nVidia G 310 is a GT200 based GPU.  It does in fact support both CUDA and OpenCL.  If you're comfortable tweaking command line parameters, I would suggest looking into puddinpop's RPC CUDA miner as well as m0mchil's Python OpenCL miner.  After tweaking paramaters, you'll probably find that, for your G 310, the CUDA miner will be about 5% faster.

Can you elaborate more on this? I have a problem with my CUDA program. GPU appears to be slower than CPU. I'm using GeForce 9500 GT.

Depending on the CPU, you're probably in the same boat.  There isn't enough processing power on one of these lower end cards to do integer operations with any significant speed.  If you have a particularly fast CPU, these GPUs aren't going to match the performance you'll get on the CPU.

EDIT: I removed references to FPUs.  Hashing and FPUs shouldn't go together.

Thanks urizane for the reply. I have:

Intel(R) Core(TM)2 Duo CPU
E7500 @ 2.93 GHz
2.93 GHz, 2.00 GB of RAM

so does your statement mean that even if i use GPU but if ever I have a more powerful CPU then GPU wont beat CPU?

[urizane]

Can you elaborate more on this? I have a problem with my CUDA program. GPU appears to be slower than CPU. I'm using GeForce 9500 GT.

Thanks urizane for the reply. I have:

Intel(R) Core(TM)2 Duo CPU
E7500 @ 2.93 GHz
2.93 GHz, 2.00 GB of RAM

so does your statement mean that even if i use GPU but if ever I have a more powerful CPU then GPU wont beat CPU?

From what I understand of the first of these two posts is that your Core2 Duo is already faster than your GPU.  However, if you intend on this system being used primarily for mining, you could keep the CPU you have and replace the GPU (given your power supply can handle the higher wattage).  In fact, the CPU you have still has some legs in some of the more recent games (except Civ V and some Starcraft II battles) and could see some benefit to having a better GPU and another 2GB of RAM.  The RAM wouldn't be necessary for a purely mining machine, though.

[LeiKriS]

Can you elaborate more on this? I have a problem with my CUDA program. GPU appears to be slower than CPU. I'm using GeForce 9500 GT.

Thanks urizane for the reply. I have:

Intel(R) Core(TM)2 Duo CPU
E7500 @ 2.93 GHz
2.93 GHz, 2.00 GB of RAM

so does your statement mean that even if i use GPU but if ever I have a more powerful CPU then GPU wont beat CPU?

From what I understand of the first of these two posts is that your Core2 Duo is already faster than your GPU.  However, if you intend on this system being used primarily for mining, you could keep the CPU you have and replace the GPU (given your power supply can handle the higher wattage).  In fact, the CPU you have still has some legs in some of the more recent games (except Civ V and some Starcraft II battles) and could see some benefit to having a better GPU and another 2GB of RAM.  The RAM wouldn't be necessary for a purely mining machine, though.

Thanks so much for your reply. so does this mean that my CUDA program which uses GPU can't beat the CPU?

[urizane]

You won't be getting much out of a 9500 GT.  Even with CUDA, it just can't do integer operations as quickly as recent CPUs.  Take a tour of https://en.bitcoin.it/wiki/Mining_Hardware_Comparison to see what your other options could be.

[LeiKriS]

You won't be getting much out of a 9500 GT.  Even with CUDA, it just can't do integer operations as quickly as recent CPUs.  Take a tour of https://en.bitcoin.it/wiki/Mining_Hardware_Comparison to see what your other options could be.

Thanks. uhm, it appears that my cpu is not in the list. do you happen to know how many mhash/s does the e7500 have? Thanks a lot for your help, I don't really know much about hardware.

[nster]

My i7 920 @ 4GHz does 20~21 Mh/s (8 threads, didn't try with HT off) with ufasoft's miner

[urizane]

Thanks. uhm, it appears that my cpu is not in the list. do you happen to know how many mhash/s does the e7500 have? Thanks a lot for your help, I don't really know much about hardware.

Well, the best way to find out is to run ufasoft's SSE2 miner for yourself.  My guess is that you'll get roughly 7 Mhash/s for your two cores if they're running at 2.93 GHz.