|
Title: Testing odd-ball PCI-e cards for mining. Post by: GhostPlayer on February 12, 2014, 04:54:26 PM Hi gang,
I'm no programmer, just a general end user geek, completely overshadowed by the knowledge in this forum!! But I do plan on mining. I have a test PC at home, and I probably will have access to dozen of PCIe cards, some with very weird processors in them. All of them have drivers and are known to work. These range from old GFX cards, to university unbranded prototypes, to sound cards etc. please bear with me, I have access to no programmer. This is a project. I'd like to run some tests though. I slapped an audio card in the PCIe slot and tried cgminer SHA-256. This card, albeit very old, was primed for being a breakthrough 32bit floating point processor... thats whats needed right? I couldn't get it go hash, but it did run and started looking for USB FPGA's, but could find anything so it stopped. Maybe a config is wrong, I'm all too new to this, but have managed to hash with another PC (GPU & script), so I know what to look for Am I right to assume that if a PCIe has good working drivers, then CGminer will communicate and do some processing? There is a thread where someone got an old crappy sound card (with DSP) to hash... He apparently only used this... Quote ./miner --combined-worker-pool --miner-urn http://[elided]:[elided]@pit.deepbit.net:8332 and has raw output of what happened. https://bitcointalk.org/index.php?topic=12782.0 Anyone care to lend a hand? Title: Re: Testing off-ball PCI-e cards for mining. Post by: nachius on February 12, 2014, 05:27:16 PM No that assumption is incorrect:
Just because it is a PCIE device does not mean it has any capability to hash. PCIE Video cards radeon 4000 series or newer with opengl support, nvidia cuda chipset with opencl support (not sure on earliest model). Overall currently mining with a video card is only profitable for scrypt mining not SHA256 and the card needs to hash above 100KHs (typically want at least 300KHs per card for it to make sense) under 200 watts and paying less than .10 per kWh to even come close to breaking even. Old devices just don't cut it...... Title: Re: Testing off-ball PCI-e cards for mining. Post by: GhostPlayer on February 12, 2014, 10:55:02 PM Sorry, I didn't explain myself properly, english is not my native language.
In the posted link, someone got a PCIe audio card to work! Basically, a DSP chip from "ancient" times, and EMUxx. Sounds bogus at first, bug digging into the thread it sure seems a reality. Unfortunately, the OP has 3 posts activity since 2011. People say sound blaster whatever hi-end... for get that. I'm talking serious DSP power. I have a studio, and doing maintenance it popped in my mind. The last few years DSP took off. Hundreds of products out there that are truly high end. I don't know the requisites for mining script vs sha-256 alga's, but I am definitely game to contribute to the community I have one of these babies currently not used in the studio. Just sitting at home in a little quad-core PC. Very happy to try something with it if someone is willing to point the way. The card has 4x these in it. Total of 9,6 Gflops and 1600mhz. I don't even know if these numbers can be summed like that ;D Quote 400MHz SIMD Sharc Core, capable of 2.4 GFLOPS peak performance 2Mbits SRAM; 6 Mbits customer-definable ROM High bandwidth, 32-bit external memory interface supporting glue-less interface to SDRAM, SRAM, and FLASH Digital Audio Interface (DAI) enabling user-definable access to system peripherals including 8 serial ports (SPORTs), S/PDIF Tx/Rx, 8-channel asynchronous sample rate converter, and 4 precision clock generators. Digital Peripheral Interface (DPI) enabling user-definable access to system peripherals including 2 SPI-compatible ports, 2 UARTs, 3 full-featured timers, and a two-wire interface compliant to the I2C standard. 34 zero-overhead DMA channels 16 Pulse Width Modulation (PWM) channels 208-lead LQFP-EP and 256-ball SBGA packages Commercial and industrial temperature ranges Maximum core performance is reduced for LQFP package http://fr.audiofanzine.com/tests_v3/universal_audio/uad2/uad_2_quad_front_hq.jpg I also have access to competing products with different architectures/chips etc. Or maybe just a silly idea? Title: Re: Testing off-ball PCI-e cards for mining. Post by: GhostPlayer on February 12, 2014, 11:05:58 PM Forgot... link to the data sheet http://www.analog.com/static/imported-files/data_sheets/ADSP-21367_21368_21369.pdf Title: Re: Testing off-ball PCI-e cards for mining. Post by: synclab on February 13, 2014, 12:33:47 AM The card has 4x these in it. Total of 9,6 Gflops and 1600mhz. I don't even know if these numbers can be summed like that ;D I don't believe that FLOPS correlates very well to mining speed since I believe that mining relies more on integer math. But that being said a standard Radeon 7970 does about 4000 GFLOPS. I don't think the effort required to get your DSP card to work would be worth the time unless it's just for fun. http://en.wikipedia.org/wiki/List_of_AMD_graphics_processing_units Title: Re: Testing off-ball PCI-e cards for mining. Post by: GhostPlayer on February 13, 2014, 04:35:27 AM I don't know, I mean, what are FPGA's's basically? What these DSP chips do very very well is 32 floating point math, using 7 watts, and apparently fully programable.
If some dude got an EMU working, c'mon, why not give it a shot? I'm gearing towards building a my first GPU rig for scrypt and devoting all my effort there, but very happy to take some pointers on what might work here, trying config setting/posting back, whatever. First generation of these cards are bought for peanuts today. Wouldn't it be cool if we managed to harness some hash juice out of them? Surely there a code wiz around that might have 0.02 fiat cents to lend...? Title: Re: Testing odd-ball PCI-e cards for mining. Post by: sirlimpy on February 13, 2014, 09:23:06 AM No one in that thread managed to get it to work, nor has anyone. They agree it might be possible but mostly not easy and in no way cost effective. You could most likely with a hell of a lot of effort make one work and a god of a programmer.
Title: Re: Testing odd-ball PCI-e cards for mining. Post by: GhostPlayer on February 13, 2014, 09:26:10 PM ok then, fun idea, thought it'd be as simple as firing up cpuminer or similar and tweaking around
case closed it seems... thanks! :-* |
