[Dev] NeoScrypt GPU Miner - Public Beta Test
-
We all want Wolf’s personal kernels, but he’s said a few times that he won’t be giving them to us. He’s already done enough. Let the Wolf be free.
This is getting weird. I’m used to getting hated on, not used to being defended. :P
Anyway, maybe I’ll start a thread with those hashrates, as well as interesting info about Neoscrypt on a technical level - maybe some basic descriptions of what I did to it, as well, and things I’d like to do.
-
This is getting weird. I’m used to getting hated on, not used to being defended. :P
Anyway, maybe I’ll start a thread with those hashrates, as well as interesting info about Neoscrypt on a technical level - maybe some basic descriptions of what I did to it, as well, and things I’d like to do.
Yes, this is not bitcointalk :)
I wonder what the temperatures are on your cards.
-
Yes, this is not bitcointalk :)
I wonder what the temperatures are on your cards.
I’m quite surprised. Not in this forum not being Bitcointalk, but in people’s general attitude. I was becoming a little jaded, having most of my interactions with others in crypto on Bitcointalk - but maybe it’s limited to there. I’ve actually had Freya up and stable for 6h, so I’ll post a screenshot in a minute with all the different cards.
-
Okay, here you go (NSFW): https://ottrbutt.com/miner/neoscryptwolf-11042014.png
As you can see, every card is around mid-70C or below, except that 270X with the dead fan. What you don’t see is that all of them have a decent sized voltage bump, except that 270X, which is slightly undervolted to compensate for its lack of cooling. So, even at this hashrate, the cards are running quite cool - too cool, as my room is rather warm (probably from all three desktops with high end GPUs…) - so I need to trade off some of the memory usage for extra computations. Compute power is likely still plentiful, unused because of memory access times.
-
Okay, here you go (NSFW): https://ottrbutt.com/miner/neoscryptwolf-11042014.png
As you can see, every card is around mid-70C or below, except that 270X with the dead fan. What you don’t see is that all of them have a decent sized voltage bump, except that 270X, which is slightly undervolted to compensate for its lack of cooling. So, even at this hashrate, the cards are running quite cool - too cool, as my room is rather warm (probably from all three desktops with high end GPUs…) - so I need to trade off some of the memory usage for extra computations. Compute power is likely still plentiful, unused because of memory access times.
This is very impressive. Thank you for the settings you are using at least. I know I am going to try some of those settings on my GPUs.
LOL @ your NSFW screenshots. The legendary Wolf0 = no fucks given :)
Edit:
Sapphire 290x Vapor-x (Hynix) gets poor hashing performance with w64 I14 g2 tc32768. w48 I15 g2 tc32768 is the sweet spot for this card using the sgminer 5 you compiled for us.
-
This is very impressive. Thank you for the settings you are using at least. I know I am going to try some of those settings on my GPUs.
LOL @ your NSFW screenshots. The legendary Wolf0 = no fucks given :)
Pretty much; also, a Hynix 280X does quite a bit better than Elpida - I believe I’m getting 290+ kh/s from it. As I decreased memory fetching, useless code, and other stuff that can cause the GPU to stall, I found running LESS threads per GPU began getting far better hashrates. With the stock miner, running more threads may help, because the more threads you’re running, the more the GPU can hide that it’s waiting on a memory access. The idea is this: if a compute unit has the resources, and is stalling on a memory access, the card can schedule another wavefront to run on the CU - this means you’re hashing even when waiting for memory.
EDIT:
I should also note that --gpu-threads is a REALLY stupid name, your GPU is running at least tens of thousands of threads, --gpu-threads means spawn x amount of threads on the CPU, each giving work to the GPU.
-
Pretty much; also, a Hynix 280X does quite a bit better than Elpida - I believe I’m getting 290+ kh/s from it. As I decreased memory fetching, useless code, and other stuff that can cause the GPU to stall, I found running LESS threads per GPU began getting far better hashrates. With the stock miner, running more threads may help, because the more threads you’re running, the more the GPU can hide that it’s waiting on a memory access. The idea is this: if a compute unit has the resources, and is stalling on a memory access, the card can schedule another wavefront to run on the CU - this means you’re hashing even when waiting for memory.
EDIT:
I should also note that --gpu-threads is a REALLY stupid name, your GPU is running at least tens of thousands of threads, --gpu-threads means spawn x amount of threads on the CPU, each giving work to the GPU.
May I ask what queue, expiry, and scan-time settings you are using?
-
May I ask what queue, expiry, and scan-time settings you are using?
It’s really totally irrelevant, but I think queue is 0 in my config, scan-time is 5, and expiry I forget - maybe not set.
-
Getting 145kh/s with my 280x with sgminer 5 settings : sgminer.exe --algorithm neoscrypt -I 15 -w 48 -g 2 --no-submit-stale -o -p
That is 60kh/s faster than 3.7.7b and the same as with 3.7.8.
Thanks Wolf, your contribution is greatly appreciated.
-
Anyone except me having problems even getting sgminer started? It comes to the jansson line and then just dies. No binary created
Tried with as few settings as you can possibly use without no difference…
Oh and btw, great job Wolf0! Accidently opened your screenshot here at work and noticed your NSFW warning afterwards, LOL :D
-
Anyone except me having problems even getting sgminer started? It comes to the jansson line and then just dies. No binary created
Tried with as few settings as you can possibly use without no difference…
Oh and btw, great job Wolf0! Accidently opened your screenshot here at work and noticed your NSFW warning afterwards, LOL :D
Delete your sgminer folder and do a recursive clone - git clone --recursive.
-
It’s really totally irrelevant, but I think queue is 0 in my config, scan-time is 5, and expiry I forget - maybe not set.
Thank you. Just trying to increase WU/m.
-
Just in case some of you haven’t seen them before, my screenshots are NSFW.
My mixed card rig, Freya, with card types and settings shown (you won’t get the same hash, but the settings should work well): https://ottrbutt.com/miner/neoscryptwolf-11022014-2.png
Wolf0 , First thank you for great work .
I tested R 290x R290 and R280 but i can’t get more than 160kh for R280 and for R290, R290x 190kh/s
Can you please help more .
How you got this high hashrate !!?
-
Wolf0 , First thank you for great work .
I tested R 290x R290 and R280 but i can’t get more than 160kh for R280 and for R290, R290x 190kh/s
Can you please help more .
How you got this high hashrate !!?
You need to read the thread a bit more. He is using his own PRIVATE kernels. As long as he never releases his private kernels you will never get the same speeds as him. 190Kh/s for a 290x is pretty much the highest I have seen posted in this thread, so why would you need help? From the looks of it you are doing just fine without any help whatsoever. How about you tell us how you got a 290x at 190 Kh/s?
-
He probably has non-reference cards, or at least Hynix memory, on the 290x. I’ve gotten a TRI-X 290 card to 190kh/s (it has Hynix memory). My reference 290x are around 178kh/s max at the moment.
Elpida, I believe. My reference Sapphire, which is kind of a dog, does 270 - 280, while my non-ref Asus does like 310kh/s. I believe they’re both Elpida, because my Hynix 280X does over 290kh/s.
-
He probably has non-reference cards, or at least Hynix memory, on the 290x. I’ve gotten a TRI-X 290 card to 190kh/s (it has Hynix memory). My reference 290x are around 178kh/s max at the moment.
I got 200 kh/s on R290x refrence and 190kh/s on 290 refrence
-
Elpida, I believe. My reference Sapphire, which is kind of a dog, does 270 - 280, while my non-ref Asus does like 310kh/s. I believe they’re both Elpida, because my Hynix 280X does over 290kh/s.
Do you have any plan for publish new kernel ?
-
Do you have any plan for publish new kernel ?
Not at the moment - however, I am considering fixing chacha and salsa for pre-GCN cards. Would even the playing field for people with 6xxx and below.
-
Not at the moment - however, I am considering fixing chacha and salsa for pre-GCN cards. Would even the playing field for people with 6xxx and below.
We would sincerely appreciate a fix for pre-GCN cards. I think people need to donate to you also for all the work you have done for us already, including outside of FTC. If it weren’t for you Wolf, we wouldn’t even have sgminer.
-
We would sincerely appreciate a fix for pre-GCN cards. I think people need to donate to you also for all the work you have done for us already, including outside of FTC. If it weren’t for you Wolf, we wouldn’t even have sgminer.
Well, salsa and chacha won’t be too difficult, but does a 6xxx card mine right now? At all? If not, then it may not be able to address global memory using unaligned addresses, and then it’d be a nightmare to fix.