Topic: r9 280x eth mining (Read 1997 times)

cleaning up some of my storage stuff.

played with two 280x sapphires.

both burned something in the board, lighted and smoked....aaand it both runs.

4 screws, you clean and you are good to go...with just 4 SCREWS!

edit: my bad, should be in the "AMD Tahiti GPUs (Radeon 7950 / 7970 / 280X / 7990 ) Owners Appreciation Thread" https://bitcointalksearch.org/topic/amd-tahiti-gpus-radeon-7950-7970-280x-7990-owners-appreciation-thread-4740661

Yes he is pretty much spot on... at least with the Daggerhash algo which is for Ethereum mining.

The R9 290 from what I recall didn't make any speed increases by changing the memory clock or even strapping the memory. I remember I had bad luck with trying to strap those. Because a little change would cause the entire system to hang and you had to pretty much do a cold power cycle. So I just left it the way it was.

Changing the memory had little effect... all it did was cause freezing with no advantages. The only way to get faster speeds was the actual engine core. But you couldn't go too much because it would eat 300 Watts of power.

If you were trying to change memory freq. on GCN1 cards, some of them have driver-capped frequencies, but upping core frequency on my 7850 with some timing improvements got me more on ethash than just one or the other. My R9 290s were mostly core limited on ethash so didn't bother with OCing memory, but I did see some small improvements on I believe cryptonight variants with tightened timings. Maybe c29 too, not sure about that one.

When you change the memory stripe, did you actually see the performance increase? I changed memory frequency from 1125 to 1250MHz, do not see any hash rate increase.

HiveOS won't spin up the fans on my Hawaii cards, but SMOS is working with it, so I very much suggest SMOS over HiveOS for GCN1 and GCN2 cards. I personally used Hawaii Bios Editor to change clocks/volts/fan curve on my 290s, and also a hex editor to change memory straps a little, but I changed when they change over rather than copying lower tier up (I was doing it for benchmarking purposes with firestrike, not mining.). I use VBE7, anorak's tools, and a hex editor for my 7850, I believe it works on the GCN1 refreshes too, I'd have to grab a vbios off of techpowerup to check if it handles it right.
edit: also to change DPM on a per-card basis on SMOS, you need to use ssh and unless you want to su to root, make a .sh file containing something like "echo "7" > /sys/class/drm/card1/device/pp_dpm_sclk " that will set a DPM level of 7 to the second card in the system,as the first card is card0. needs to be a .sh file run with sudo because the ">" happens to break sudo, applying to echo but not opening of the "file" so it can't write to it, while making it a .sh file to run has sudo apply to everything in it. I currently don't have a refresh of the GCN1 cards to test how their powerplay settings are in vbios. my 7850 doesn't seem to really have a full DPM profile in the vbios and I can only set maximums, which it runs at by default so I don't know how 270(x) or 280(x)s would behave in that regard. Also, sorry if I ramble a bit too much.

Yes, I tested and it works with R9 290 and also R9 280x. Higher hashrate also gives higher power consumption...
I used google translate to translate guide to English...

If the Hawaii is similiar to the Tahiti's then the reason why you aren't getting the highest GPU frequency is because the DPM states need to go from the lowest to highest in frequency. They need to be in order like 300Mhz,500Mhz,800Mhz, 947Mhz, 1000Mhz, 1050Mhz, etc. You can't do something like 300Mhz, 500Mhz, 800Mhz, 1000Mhz, 1050Mhz, 947Mhz, because you will run into issues that way.

What software are you using to modify your BIOS?

I remember with the Hawaii GPUs, I didn't have issues with the clockrates but I had issues with the fan speeds. I could never get them to automatically set but had to manually set the fan speeds because modifying the bios didn't do anything.

Guys anyone knows how to set the DPM on HiveOS for r9 290/390 GPUs? I have overclocked the cards from bios but the OS doesn't pick the highest GPU frequency. I tried with the wolfamdctrl command but it says: Specified core state does not exist. Is it ok to modify again the bios with the same frequency on all states and the same voltage to bypass the bug from the HiveOS? or the card will get damaged?

the labor will be too small. If you add these 3 things then you can still create a mini farm. And about one card is childish pranks

I mentioned previously FGLRX has the problem with thrashing, AMDGPU-PRO doesn't. Also, Pitcairn is much much lower power design than Tahiti. Same process node, but the 7850s were only 120W TDP cards. Not sure if they had theirs modified to get the lower power. (VBE7  can adjust frequency/voltage settings on GCN1 cards vbioses, but amdgpu ignores any memory setting above 1250mhz and caps it there on Pitcairn, there's a file in the opensource side of the amdgpu-pro driver specifying "quirks" that needs modified and recompiled to OC past that)
EDIT: Also, I had some trouble getting amdgpu-pro to work with that 7850 on vanilla unmodified ubuntu besides installing amdgpu-pro myself last year, it'd only enable less than half of the memory and 8CUs, so I've been exclusively using SMOS, which is derived from ubuntu, to test it. I probably should check the state of things now

Just look at the first picture in this article  https://www.cryptoprofi.info/?p=3313 and you will see that R9 290 give almost 29 mh/s. There is 1000 mhz at core with downvolted consumption and bad Hynix memory. If you use more power and better chips of memory it is easy to achieve 30-31 mh/s.

I don't know about the Hawaii but I do know that the Tahiti and the Pitcairn running under Ubuntu suffered the same DAG thrashing bug on Linux as it did with Windows. This was with the Claymore miner and using various AMD drivers.

This however was a long time ago, probably about 2 years ago and I haven't tested it since. Maybe out of pure boredom I will test out the Tahiti's using PhoenixMiner on some low DAG # coin and see what the results are.

90 Watts is pretty good however are you sure its accurate. I remember my Tahiti's back in 2016 mined at 20Mhs and used exactly 200 Watts each.

I'm using phoenixminer myself. I found it odd that I've only found 1 other person that seemed to realize the old GCN1 and GCN2 cards work so much better on (modern) linux for pure stock, and they were mining ubiq on a rack's worth of the same powercolor HD 7850 I have, tuned to ~90W and 17MH/s. I have yet to see how Tahiti cards do, but I really want to know now that I've seen Pitcairn and Hawaii do so well.

Hmm, I did not know this.

So basically the Hawaii slowdown with the DAG is not related to the GPU architecture but to the operating system and drivers being used.

Thanks for telling me this, I might dust them off and power them up again if I can get close to 28MH/s. Really surprised nobody has mentioned this before on the Claymore thread. Tons of people complaining why their 290/390 is slowing down and nobody could figure out the issue. Most just eBay'd or Craiglisted their R9 290 and got an RX 470 instead.

If ETH can get back above the $300 area again then with the huge power hog that the R9 290 is, it still might be a little profitable. The Reference design is rock-solid built and very stable at any temps.

One of my R9 290s is mining ETC at 29.8 right now, on SMOS. 1040 core, 1250 memory. My other is doing 28.2 at 977mhz core 1250 memory (these are the cards stock settings from factory). edit: I do have ref set to 40 on both, but I don't believe it makes a difference with how much bandwidth they just naturally have, so have it set Just Because.
The 7850 was mining Callisto on SMOS at 17MH/s before it got just a little too big to fit even on linux, the 13MH/s I managed on windows was with a wildly OC'd core and memory with callisto. Expanse was about the same as callisto, Ubiq did better with its smaller DAG (I think its ethash is also a slight modification over normal?? Still don't know for sure). On ubiq I still got 17 on linux, but was able to pull 18MH/s on windows with a high OC.

It depends what coin you are mining on Daggerhash.

You said you got 7850 running at 13MH/s, since every 7850 has max 2GB of ram you most likely are mining some extremely low DAG coin like Expanse.

So with DAG #0, an R9 280X can mine at 27MH/s and an R9 290 can mine at 31MH/s. I know because it was there when ETH had low DAG numbers.

However you can't mine ETH with an R9 290 at 30MH/s anymore, because sometime last year when the DAG increased the speed slowed down and you get like 15MH/s or less with the R9 290. With the Pitcairns or Tahiti's you can't even mine ETH.

You can mine those other coins but their profitability is way low. And ETH profitability is not that great either. So no point in using those GPUs. Just sell to gamers instead and get a cheap RX 470.

related to assumptions I made on previous posts, I'm assuming it's not eth, but an ethash coin, right? ETH and ETC won't fit.I've still had it run on a few cards where it doesn't fit using system memory, but at like 2-4MH/s.
the speed on GCN1 and GCN2 cards will go down as DAG size grows, if you're on windows or using the fglrx driver on old linuxes. Radeon HD 7950/7970/R9 280/280X are all GCN1 "Tahiti" cards.
Claymore added a "benchmark" mode for testing how speed will be impacted by different DAG Epochs. (the epochs determine how the size grows)

R9 280 and 280X's are all 3GB. eth and etc won't fit but there's other coins that will, like expanse, ubiq, callisto, and so forth (I asked an admin of wattpool to add the DAG size of the many ethash coins on their pool, and he did so you could check what coins are viable there). There were a few 6GB 7970's which the 280X is based on.
the R9 285/380/380X have 4GB models available.

not just linux, it's the AMDGPU-Pro driver improving things. I was testing an old HD 7850, gets 10-13MH/s on windows and FGLRX on linux, but AMDGPU-Pro on SMOS got me 17-17.7MH/s. HiveOS just refused to use the 7850. SMOS and HiveOS both got me about 27-30MH/s for an R9 290, but only 19-20MH/s on windows. My RX cards were the same on linux distros in general. The RX Cards were the same on linux and windows with compute mode enabled*. linux was doing the same thing as the compute-mode by default, I believe AMDGPU-Pro is basically defaulting to compute-mode behavior for all GCN cards while windows only uses that behavior on GCN3 and up.

Thanks for your reply. I will run rx470/580 with windows 10. And I will let r9 run with Ubuntu.