CPU or GPU?
 

Hi, is my Haswell i5-4670K or my Kepler GTX 660 better for trial factoring?

Thanks in advance.

With a rate of [URL="http://www.mersenne.ca/mfaktc.php?show=533&sort=ghdpd"]about 180GHzD/D[/URL] at TF for a 660, I don't think that [B][U]any[/U][/B] CPU can keep, even with a quarter of it. Additionally, your CPU is extremely good at LL, so better use it at LL/DC, especially with a good/fast/wide/channeled memory.

OK, thanks for the reply.

EDIT: You mentioned fast memory, is this memory good? That's the only stick I have. [url]http://imgur.com/pEKKs1n[/url]

That's a slow stick (1333 MHz). If you have a z97 motherboard, you would do best with a pair of 2133 or 2400 MHz memory. Even a second stick of 1333 would double your throughput. Prime95 is memory bandwidth bound for performance.

It also looks like that stick should hopefully run at 1600.

[QUOTE=Mark Rose;407669]That's a slow stick (1333 MHz). If you have a z97 motherboard, you would do best with a pair of 2133 or 2400 MHz memory. Even a second stick of 1333 would double your throughput. Prime95 is memory bandwidth bound for performance.[/QUOTE]

It's an H87 board, though it allows me to overclock both RAM and CPU even after a bios update, at first I thought I was locked out from OCing forever on this board, but it looks like MSI didn't include the microcode update yet.

The CPU is dual channel, I was referring more like 2 sticks against a single stick. If it is also faster clock, then even better, but the LL performance comes from the dual channel and the huge (6MB), "super-fast" cache of [URL="http://ark.intel.com/products/75048/Intel-Core-i5-4670K-Processor-6M-Cache-up-to-3_80-GHz"]the CPU[/URL].

So I decided to go into the BIOS and fiddle with the RAM OC settings. I set the DDR frequency to 1600MHz and turned on the Extreme Memory Profile, as actually that Kingston stick supports these settings.[URL="http://imgur.com/YN3SOXO"]http://imgur.com/YN3SOXO[/URL] So now I'm running on this frequency. [URL="http://imgur.com/iITROfF"]http://imgur.com/iITROfF[/URL]

Oh, and by the way is 30ms per iteration good while all 4 workers are active?

Which exponent (range?). Assuming it is at LL front (~59M), it is slow.
Can you run a benchmark (from the options menu) and see (or post here) time difference between 2-3-4 workers? With a single stick, you are most probably limited by the memory bandwidth, and those milliseconds could be (almost) halved if you add an identical stick, as Mark Rose also said above. The CPU which I post this from is a i5-3340M at 2.7G, (so, half of your cache, slower clock, 2 physical cores, 4 HT threads) running two workers, and it gets a bit better times than you posted... Your CPU can do more, for sure.

Yeah sure, I'll post it when it's finished.