Gateless Gate Sharp 1.3.8: 30Mh/s (Ethash) on RX 480!

[nerdralph]

On Linux with my Rx470 I get 29.3Mh with sgminer-gm 5.5.5 (xI:1155) while gg does 29.0 (with xI:2304 to keep HW errors under control).  Claymore gets ~30.3, so we know there is room to improve ethash-new.

There is - but not a ton. I can get ~1.5% more than Claymore's, but that's about it.

Which is quite good considering 32Mh/s is the absolute maximum possible with memory clocked at 2GHz if there was no refresh overhead and no channel contention.
Release it closed source with a 0.5% fee, and I think you'd do well.

Besides absolute performance I think there's room to improve power efficiency.  To get 31Mh@2.1Ghz you probably need to run the core at 1150-1175Mhz.  31Mh/s uses 248GB/s of L2 bandwidth, which can theoretically be achieved with a 970Mhz core (64 bytes/L2 channel per core clock * 4 channels).  I've been playing around with CU utilization (disabling CUs and just running kernels on a limited number of CUs), and while using 8 of 32 CUs I only see a 1% hit on memory throughput.  Power reduction isn't very impressive, but I'm still figuring things out.

[doktor83]

how do you measure 1.5% increase in performance when hashing speed is not constant ? 

[UnclWish]

how do you measure 1.5% increase in performance when hashing speed is not constant ? 

It's need several hours to measure.

[one level]

hi. how to mine ubiq or music by gatelessgate? I have message probing for an alive pool and that's all.

[cryptominer420]

most UBQ and Music pools use a http proxy and not stratum which is what appears to be the cause. I got it working using the following config for UBQ using my R9 270 2GB card making $38 per month.

Code:

@echo off
set GPU_FORCE_64BIT_PTR 0
set GPU_MAX_HEAP_SIZE 100
set GPU_USE_SYNC_OBJECTS 1s
set GPU_MAX_ALLOC_PERCENT 100
set GPU_SINGLE_ALLOC_PERCENT 100
set CUDA_CACHE_DISABLE 1
gatelessgate.exe -k ethash --gpu-threads 2 --gpu-platform 0 -o http://ubiqpool.io:8888/UBQADDRESSHERE -u UBQADDRESSHERE -p x --xintensity 1512 --worksize 192 --api-listen --api-groups P:enablepool:*

NOTE: this was also for a old version of GG.

@zawawa Sent a BTC donation it isn't much but I try to send something every month.
TxId: b7ff373a5a2d5cb554139c467f357ed1bc0dffd176f43e56bcc868dbf1c64a0c

[nerdralph]

how do you measure 1.5% increase in performance when hashing speed is not constant ? 

Eth hashing speed is a function of memory bandwith (hashing speed = bandwidth/8MB).
A sub-optimal kernel may not be tuned to use the maximum bandwdith, and therefore may show some variability in performance.  An optimally-tuned kernel would have no memory bandwidth contention and therefore no variability in hashrate.  Practically speaking, a 5-10 minute test is all you need to tell if eth performance is better.

For ZEC the solution rate is statistically variable, so for performance testing you pay more attention to the itterations/s than the solutions/s.

[UnclWish]

how do you measure 1.5% increase in performance when hashing speed is not constant ? 

Eth hashing speed is a function of memory bandwith (hashing speed = bandwidth/8MB).
A sub-optimal kernel may not be tuned to use the maximum bandwdith, and therefore may show some variability in performance.  An optimally-tuned kernel would have no memory bandwidth contention and therefore no variability in hashrate.  Practically speaking, a 5-10 minute test is all you need to tell if eth performance is better.

For ZEC the solution rate is statistically variable, so for performance testing you pay more attention to the itterations/s than the solutions/s.

Hmmm, that formula works not on all cards. F.e. on 280X memory bandwidth is higher but eth hashing speed not.

[nerdralph]

how do you measure 1.5% increase in performance when hashing speed is not constant ? 

Eth hashing speed is a function of memory bandwith (hashing speed = bandwidth/8MB).
A sub-optimal kernel may not be tuned to use the maximum bandwdith, and therefore may show some variability in performance.  An optimally-tuned kernel would have no memory bandwidth contention and therefore no variability in hashrate.  Practically speaking, a 5-10 minute test is all you need to tell if eth performance is better.

For ZEC the solution rate is statistically variable, so for performance testing you pay more attention to the itterations/s than the solutions/s.

Hmmm, that formula works not on all cards. F.e. on 280X memory bandwidth is higher but eth hashing speed not.

Sure it does.  You're probably mistaking maximum peak bandwidth with sustained bandwidth for a large working set.  GCN1.0 devices like Tahiti & Pitcairn show a material decrease in effective memory throughput as the working set (DAG size in the case of ethash) increases beyond 1GB.  If you've mined eth for any length of time you'd have noticed that a 280X performed at ~27Mh when the DAG was just over 1GB, and now that is almost 2GB, you're lucky to get 20Mh from it.
Core speed also can play a factor in older devices that may not have enough compute power to do the keccak function as fast as the data is coming in from memory.  For GCN1.2 and later (i.e. Tonga, & Polaris), default core clocks are more than enough even with overclocked memory.

[sp_]

Ethereum:

The r9 390 should be able to do 36mhash. When undervolted the card use 260watt...

[nerdralph]

Ethereum:

The r9 390 should be able to do 36mhash. When undervolted the card use 260watt...

I think Wolf's private ASM kernel is the best one for Hawaii.  I don't see much point in trying to optimize for the older cards.  Even though I have Pitcairn, Hawaii, Tonga, Tahiti & Polaris cards, I'm only working on a eth miner for Tonga & Polaris on Linux.

[newmz]

Ethereum:

The r9 390 should be able to do 36mhash. When undervolted the card use 260watt...

I think Wolf's private ASM kernel is the best one for Hawaii.  I don't see much point in trying to optimize for the older cards.  Even though I have Pitcairn, Hawaii, Tonga, Tahiti & Polaris cards, I'm only working on a eth miner for Tonga & Polaris on Linux.

I know this is probably a very "noob" question - but how does someone obtain a "Wolf's private ASM kernel" and is it worth it for someone who has only 6 Hawaii GPUs?

Or is it only economically practical for large mining farms?

I'm currently using the version of sgminer-gm supplied with Ethos 1.2.0 and for me it does better than Claymore, in terms of hash-rate vs power consumption. Hawaii GPUs are great at hashing but they just use SO MUCH power. I also have a rig with a couple of gtx1070s on it and they hash at approximately the same rate as my 290s but using maybe 60% of the power consumption.

I have not yet tried Gateless Gate but I am going to out of curiosity try it for ETH and ZEC.

I think that all those people out there using Claymore because it does dual mining are going to learn in the long term that it will just make their cards fail much more quickly than they would have just "single" mining. I doubt there are many scenarios in the intended usage of these GPUs (gaming, etc) that demand what dual mining demands of a GPU for sustained periods of time, but since I'm not an engineer that is just a guess.

[doktor83]

Ethereum:

The r9 390 should be able to do 36mhash. When undervolted the card use 260watt...

I think Wolf's private ASM kernel is the best one for Hawaii.  I don't see much point in trying to optimize for the older cards.  Even though I have Pitcairn, Hawaii, Tonga, Tahiti & Polaris cards, I'm only working on a eth miner for Tonga & Polaris on Linux.

I know this is probably a very "noob" question - but how does someone obtain a "Wolf's private ASM kernel" and is it worth it for someone who has only 6 Hawaii GPUs?

Or is it only economically practical for large mining farms?

I'm currently using the version of sgminer-gm supplied with Ethos 1.2.0 and for me it does better than Claymore, in terms of hash-rate vs power consumption. Hawaii GPUs are great at hashing but they just use SO MUCH power. I also have a rig with a couple of gtx1070s on it and they hash at approximately the same rate as my 290s but using maybe 60% of the power consumption.

I have not yet tried Gateless Gate but I am going to out of curiosity try it for ETH and ZEC.

I think that all those people out there using Claymore because it does dual mining are going to learn in the long term that it will just make their cards fail much more quickly than they would have just "single" mining. I doubt there are many scenarios in the intended usage of these GPUs (gaming, etc) that demand what dual mining demands of a GPU for sustained periods of time, but since I'm not an engineer that is just a guess.

You contact Wolf and make a deal with him.
But 6 cards is way not profitable to pay for a kernel.

[xeridea]

Ethereum:

The r9 390 should be able to do 36mhash. When undervolted the card use 260watt...

I think Wolf's private ASM kernel is the best one for Hawaii.  I don't see much point in trying to optimize for the older cards.  Even though I have Pitcairn, Hawaii, Tonga, Tahiti & Polaris cards, I'm only working on a eth miner for Tonga & Polaris on Linux.

I know this is probably a very "noob" question - but how does someone obtain a "Wolf's private ASM kernel" and is it worth it for someone who has only 6 Hawaii GPUs?

Or is it only economically practical for large mining farms?

I'm currently using the version of sgminer-gm supplied with Ethos 1.2.0 and for me it does better than Claymore, in terms of hash-rate vs power consumption. Hawaii GPUs are great at hashing but they just use SO MUCH power. I also have a rig with a couple of gtx1070s on it and they hash at approximately the same rate as my 290s but using maybe 60% of the power consumption.

I have not yet tried Gateless Gate but I am going to out of curiosity try it for ETH and ZEC.

I think that all those people out there using Claymore because it does dual mining are going to learn in the long term that it will just make their cards fail much more quickly than they would have just "single" mining. I doubt there are many scenarios in the intended usage of these GPUs (gaming, etc) that demand what dual mining demands of a GPU for sustained periods of time, but since I'm not an engineer that is just a guess.

I dual mine with underclock and undervolt, so longevity shouldn't be an issue. Been mining since Polaris was released, no failures yet.  Have RMAed 2 out of 80 cards, but they were unstable from the start. Keep em cool (<70C for me), and they will last a while. You could also lower intensity of dual mining to lessen the extra load. Mysteriously Claymore miner mines Eth slightly faster in dual mode, so long as your dcri isn't too aggressive. A 480 gets ~$20/month extra profit, so it is totally worth it if you PSU, air conditioning, and breaker panel can handle it, and your power doesn't cost too much.

[zzzzzzzzzz]

Hopefully, zawawa's absence from the thread means he got that job!!

[zawawa]

I just came back to the States.
I am not going to give out details, but it seems like I will do a full-time miner development for the foreseeable future.
Hopefully, that will have synergistic effects on GG. We will see.

[frostminer]

I just came back to the States.
I am not going to give out details, but it seems like I will do a full-time miner development for the foreseeable future.
Hopefully, that will have synergistic effects on GG. We will see.

Congrats!

Fix some more stratum support, and id gladly use this with donation enabled.

[hannusolo]

The best job is always doing one's hobby
Let's hope it will benefit your knowledge and Gateless Gate!

[agis6]

I just came back to the States.
I am not going to give out details, but it seems like I will do a full-time miner development for the foreseeable future.
Hopefully, that will have synergistic effects on GG. We will see.

Congrats!!!
Looks like your efforts paid back. I guess your wife is smiling now

[yoyo1]

My only hope is that Sgminer-GG won't stop its development.

[Trimegistus]

I just came back to the States.
I am not going to give out details, but it seems like I will do a full-time miner development for the foreseeable future.
Hopefully, that will have synergistic effects on GG. We will see.

Can we expect an updated miner anytime soon?