If I recall correctly, cryptonight uses a 2MB buffer, which means you need MORE than 2MB of cache to fit it entirely and avoid using main gpu ram. Moreover, the miners must work on many cards so they probably make use of gpu ram anyway. Cache helps, but there will be a lot of misses. Thus, cache size and speed matters but so does global memory latency.
If your rig has 6 gpu's they have a total of 12 MB cache. instead of computing the complete hash each of the gpu's, it might be possible to spread the work. not sure the internal state can be split... maybe someone with better knowledge on cryptonight, like wolf0, can comment. |
|
|
|
|
FPGA mining... any details available? Which language, board type will it support?
|
|
|
|
Thanks for the response.
I tried running my sgminer with the setx GPU_MAX_ALLOC_PERCENT 100 command, and noticed no difference, GPU still uses ~1.5gb of memory out of avalible 3gb
Should it have made a difference?
I don't think it makes any difference for algos that don't use the global ram (like groestl) or use it but only for scratch registers (i.e. no explicit buffer allocation) (like x11). As far as I know, only scrypt uses those kind of buffers in sgminer. |
|
|
|
Regarding cryptonight: for the ccminer cryptonight kernel it seems like the level 2 cache is a big deciding factor. 750ti / 2mb l2 cache 960 / 1mb l2 cache 970 / 1.75mb l2 cache 980 / 2mb l2 cache Titan X / 3mb l2 cache (anyone got one yet  ) But how come the amd cards are that fast with only 768kb for the 280x or 1mb for the 290x?? Is memory bandwidth and latency that much more important for this algo? If I recall correctly, cryptonight uses a 2MB buffer, which means you need MORE than 2MB of cache to fit it entirely and avoid using main gpu ram. Moreover, the miners must work on many cards so they probably make use of gpu ram anyway. Cache helps, but there will be a lot of misses. Thus, cache size and speed matters but so does global memory latency. |
|
|
|
well thats funny cause it is totally hashing away and like I stated I am solving blocks and receiving my coins just get bouts of large amounts of share above target errors
did you try --difficulty-multiplier ? though intel graphics is nevertheless hmmmmmm no i didn't how does that work? and what would be your recommended multiplier? generally 2 values are worth trying, 256 and 1/256, but in rare cases you might experiment, put in your conf "difficulty-multiplier" : "256" or "difficulty-multiplier" : "0.00390625" i doubt that helps since intel's not supported but... it should work on some OpenCL devices other than AMD, including intel and nvidia. |
|
|
|
I also have problems on 280x with high temps.
Can someone tell me how can this afterburner help to run my GPUs cooler? What can it do that I cannot with miner config or AMD Catalyst (--gpu-fan, --gpu-engine...)?
and what can you do with miner config actually ? on my 280x, miner config can't set clocks lower, only higher, but afterburner can, that trouble started with 14.6 drivers. gpu-z will show you everything why would I lower my clocks anyway?! to lower the temperature? lower clocks might give more hashes per watt in some cases. for example when the bottleneck is the memory, or when higher clocks means higher voltage + higher temperature which means higher fan rotation speed which in turn means higher power draw. Some fans can take 10 or even 20 watts. That's why you should always use a power meter on your rigs. |
|
|
|
What's the highest hashrate for 280x with private kernel at this time?İf it's effective possible to buy ))... i have 25 280x for gpu mining.
My miner is doing about 450 Mh/s at 1175/1025 MHz on 14.12. But I didn't test it with 15.3 drivers yet, which are supposed to give a good boost, probably it'll reach 500. |
|
|
|
Why is the difficulty so low ?
Farms kicking in and out, playing with the difficulty. The coin is great, but when farms kick in, it's pretty much their coin. algo-wise this coin is dead, wpx icecream heats as hell now Every algo will produce heat when fully optimized, but I wouldn't call cards running at 70C hot. Consider a better cooling and not staking tons of cards next to eachother  i mean watts per coin, cooling won't help as you said "Every algo will produce heat when fully optimized" let me ask then why launch a coin with new algo so underoptimized at start and claim it energy efficient. the story reminds me of groestlcoin Algo isn't the only thing what is energy efficient, the whole source is. If you follow news you will see that the staking iOS and Android wallets will use ~0% cpu. i follow only numbers, if you launch a coin with public miner at 100MH/card, mining enough the same time with custom one at 300MH/card, sell the coins, then everything else is just for fun Your 300% example is quite conservative when compared to neoscrypt and it's potential 1000-2000%. no need to compare, neo's "potential" is already cashed out, look at ftc price, now wpx turn. it's all the same old groestl story, with variations Please show me what groestl coins had to offer the world except a new algo and another Bitcoin clone. Point it out. You can't even compare this project, it's codebase and the future potential of it with those. I feel that you still talk about this project as if the only thing it would offer is a new algo and when people took their profit this would just die off, no chance that this is going to happen. I think he means that the groestl algo is the classic example of a very slow public miner which had much faster private implementations since the beginning: coins using that algo have been mined like that for months until the profits went over and finally someone (me) released an opersource miner up to speed ;-) |
|
|
|
guys who experience slot 0 problems,
do you have some driver dll's in sgminer folder ?
(i got some other weird effects on 15.3 with 14.7 dll's in there,
when mining on pregenerated bin, not when compiling it)
Can't be, I'm on linux ;-) |
|
|
|
Hey Pallas,
It's been a couple of weeks. Any notable improvements?
I'm struggling to make it hash significantly faster than stock kernel on 280x. I may release the full miner earlier so everybody can enjoy. I finally acquired a 280x. It hashes at 1.115 Kh/s, clocks 835/1500, intensity 16 (instead of 15 used on Hawaii). Did you mean 1.115MHs? Locale mistake: I used . instead of , as thousands separator 😉 |
|
|
|
Hey Pallas,
It's been a couple of weeks. Any notable improvements?
I'm struggling to make it hash significantly faster than stock kernel on 280x. I may release the full miner earlier so everybody can enjoy. I finally acquired a 280x. It hashes at 1.115 Kh/s, clocks 835/1500, intensity 16 (instead of 15 used on Hawaii). |
|
|
|
Is VNL still one of the most profitable coins to mine? Or in other words, what is more profitable: x11 with 6,6 mh/s or VNL with 260MH/s (R9 280x) ? http://www.coinwarz.com/cryptocurrency nor http://whatmine.com/ doesn't show profitability for WhirlpoolX. Is there any other site which does show? you can have a look at orders on nicehash to have an idea of VNL profit. |
|
|
|
Mining x11, gpu0 hashes but no submitted shares and no hardware errors, gpu1 is ok.
Tried switching the two gpus but still gpu0 doesn't work (i.e. the gpu in the top pci slot).
I'm using wolf0's binaries.
Mining other algos works fine.
Also tried older sgminer versions but no change.
Any clues?
I'd say the problem is between Wolf's x11 bins and new AMD drivers (14.12 and 15.3beta). Standard x11 kernel works with any driver version. Sgminer version is irrelevant here. Wolf, do you have any explanation? I can add the information that I had this problem with pregenerated bins: if I compile them myself, they work. Maybe it's a 32/64 bit issue? |
|
|
|
|
Programming is not easy. Nothing is easy.
If it's easy, than it's of no value :-)
And unlike most people say, it's not for everyone. Like playing an instrument or piloting a plane.
If he's smart and has the will, he'll learn which is much more important than some more hashes per second ;-)
|
|
|
|
R9 290 @ 1000MHz: 547MH/s R9 280x @ 1150MHz: 504MH/s R9 270 @ 1123MHz: 309MH/s 7990 @ 1000MHz: 438MH/s (per GPU)  is Real ? It's real if you can make it :-) |
|
|
|
there are at least a couple very low hanging fruits for us developers to enjoy, thanks wolf0 for helping preserve the specie ;-)
Tsss! Don't tell anyone about prepatched kernels!  I'm in the 600 Mh/s club: 621 on r9 290x@1100. Now wolf0 will reply "I'm at 700+"... ;-) |
|
|
|
I also figured out the Catalyst 15.3 Beta seems to be even better for this miner.
Any way to run on linux? Extracting libopencl from Ubuntu vivid package? Running Catalyst 15.3? If you can rip it out of shitbuntu's package they got from AMD, awesome. Also, please upload it somewhere.  I've extracted the libs and rebuilt the (modified) kernel with those, it became 104 bytes fatter and hashrate went from 377 to 374 on a 290x@936 MHz; don't know how it works on your stock kernel, though... Still want the files? |
|
|
|
I also figured out the Catalyst 15.3 Beta seems to be even better for this miner.
Any way to run on linux? Extracting libopencl from Ubuntu vivid package? |
|
|
|
|
xintensity.... not the same as intensity ;-)
|
|
|
|
|
Show Posts
