Topic: Can't break 700Kh/s with 280X Dual-X OC (11221-00-20G) - page 2. (Read 8647 times)

No one can upload the bios of stilt for 280x dualx? I can't see the dropbox link, all are empty, (((

Use SGMiner 4.0 instead of 4.1 for the optimized .CL and .BIN files to work.

this is my sgminer: https://imageshack.com/i/5hbihap
my r9 280 x dual x shappire: https://imageshack.com/i/0nfzyg (work well!)
second equal card, but with 100 khs less: https://imageshack.com/i/0zpqwag

FLASHED  from 0.41 and 0.40(0.40 work well) to 0.39
undervolt only the card that work bad for try something and now with vdcc 1.175 and engine 896 have only 625khs!

i can't see the dropbox bios optimized from stilt! (((
one can upload bios on mega?

I've been following this thread for a while, and I'd like to chip in with a problem of mine if you don't mind. I've got 3 VTX3D 280x cards, and I had them running at 700 KH/s each for a few days, on cgminer. I tried mining Darkcoin with sgminer but I switched back to cgminer on other coins after a few hours. Ever since that, I can't get my cards to go beyond 670 KH/s each; the worst thing however is that the WU dropped from 650 to ~500 on each card, and I can't increase it at all. Is it possible that sgminer somehow messed my cards up? Should I just switch back to sgminer? I'm on Windows 7 x64, and this was my config when everything was running perfectly:

-g 2 -w 256 -I 13 --lookup-gap 2 --thread-concurrency 8191 --gpu-engine 1028 --gpu-memclock 1500 --gpu-fan 85

Higher clocks than 1030 result in highly decreased hashrates btw. They're undervolted @ 1.075mV which worked perfectly fine for me. Powered by a Corsair RM1000 PSU, it's an open rig

Also Mezza  I tried again using the .cl file and .bin file and it turns out sgminer ignores both and makes a new .bin file using the ckolivas.cl as it shows here

http://prntscr.com/2u1jq4

what bin files are in your miner's folder?

I might try specify the kernel again, though i didn't have a lot of luck last time with making sgminer work at all doing that

This is what I seem to be getting as it is

http://prntscr.com/2u1kb2

What do you think of the khs and WU figures

This is using following

--lookup-gap 2 --thread-concurrency 8192,8192,8192 -g 2 -I 13 -w 256 --auto-fan --temp-cutoff 90,90,90 --temp-overheat 85,85,85 --temp-target 68,68,68 --gpu-memclock 1500,1500,1500 --gpu-engine 1065,1065,1065 --expiry 1 --scan-time 1 --queue 0

I need to play with GPU 0 still,

Mezza I would also like to know how you used that 'optomized' BIOS file cos atiflash doesn't recognise it when I try.  Maybe because its a .bin file not a .rom?  I followed crytobadger's instructions like when I unvervolted my card and it still won't work

Also thinking about it again I'm not sure you have been using the 'optomised' kernel file... because when I have multiple .cl files int he kernel folder, like you said you did sgminer just picks one for me...I'd be really interested to see whatyou .bin file is once you've been hashing

Yeah me too. Proving harder than I had hoped too.  The .bin file that you'll find in the sgminer folder has the name of a kernel file at the begining of it. Mine always seems to pick ckolivas.cl.  I had the scrypt130511.cl file in the kernel folder, but it doesn't pick it and if i delete the others it doesn't work and I can't make the kernal argument work either... so far.

What does you .BIN file start with, I wonder what kernel your actually using then?

sgminer has serveral kernel files but i don't know why it picks a particuar kernel

FYI... required reading in this one. "hashrate isn't everything":
https://litecointalk.org/index.php?topic=2310.45

I did some math and I'm getting the following WU %'s for cards 0-3: 90%, 88%, 86%, 88%. That's 640, 624, 622 & 618 WU/s. Not to shabby, but also very interesting... because my GPU 0 is outperforming my GPU 2, even though GPU 2 has 17Kh/s more than GPU 0.

Also... this http://www.overclock.net/t/1461916/tutorial-ultimate-guide-to-tweaking-amd-cgminer-3-7-2-kalroth-edition-sgminer-instructions-included-aswell-get-the-most-hash-out-of-your-gpu

Time to tweak again.

WU/s = Work Utility / second. My understanding is that it's the actual usable data that your card is producing. You may be hashing at 740Kh/s but if your WU/s is 600, your card is doing 140Kh of nothing (except adding to the ambient temperature of your room and making your electric bill higher). But that's as much as I think I know (still a little fuzzy). Also I found this (explanation of cgminer interface), which is very helpful: https://doges.org/index.php?topic=5101.0

As far as the kernel goes, I simply renamed my existing kernel file "scrypt130511.cl" to "scrypt130511.cl.backup", and then copied the modified "scrypt130511.cl" file in to the same location as the backup. I didn't specify anything in my arguments, the miner just picks it up. Incidentally, how can you tell which .cl file the miner is using?

CGWatcher just gives you a nicer way to look at your stats, but I prefer to just look in the miner window since it's gives you immediate stats on what it's doing right that second. I primarily use CGWatcher because it can reboot the miner or even your computer if things crash. But it has a ton of other cool features like e-mail updates (which I just discovered yesterday, pretty cool). I notice that my hash rates jump up and down a lot in the first 15 or so minutes that the miner starts, so any time you make a change to your settings give it at least that long before determining if what you've changed helps or not. After it's been running for several hours the average rates should hardly change much at all. If it continues to jump around for a long time it may mean you're getting errors or lots of rejects (so I've heard).

That's as much as I know right now. Feel like I'll never be an expert at this though since there seem to be no end to the variables that can influence performance. Wish I were an engineer, alas I'm just a lay person trying to make a buck.

 

See I'm even more a newbie than you. what does WU/s do or mean.  I've reduced the VDDC on my 3 cards to 1100/1120/1100 ( the middle card was prevolted at 1225 the 1st and 3rd at 1175 stock which might explain why its needs a high VDDC to remain stable.

What kernel file do you use?  I downloaded the 'optomised' .cl from your earlier post and put it in the kernel folder, but it always seems to use the "ckolivas.cl", if I only leave the optomised .cl file it won't work and putting in "kernel : scrypt130302" into my arguments means sgminer won't work.. so not sure what i'm doing there.

TC 8191 did seem to raise my hash rates up  707K/710kh/s  its the blue bit that i'm taking that from.  My numbers still seem to jump up and down alot though,  does this watcher program help give you an average so you can tell if it really goes up?

Quick update- I did some more tests this weekend.

• tried Powertune +20, this did pretty much nothing for me (but I'm not ruling it out, it may be configuration dependent. Will experiment with this again later)
• tried --thread-concurrency 11200: Not really any improvement. I keep trying this because I read in a tutorial a few weeks ago that to find your optimal TC number, you should start your miner with no TC argument and then see what it defaults to (found in the file name of the BIN file that's generated). When I do this it always uses 22400. So- since I'm using -g 2 (two threads rather than one), it *should* be 11200 for me. But, apparently it's still not. (may try again later... or maybe tri 11199??)
• tried --thread-concurrency 8191 again: BAM! Hashrates jumped up. Now running a pretty stable 705,710,720,700 avg. hash rates across GPU's 0,1,2,3. Has been running stable for 24+ hrs now.
• 8191 required that I could no longer use the pre-generated 8192 BIN file that I mentioned in previous posts. Seems to have made little or no difference, which makes me wonder if it was really helping initially or not.

https://i.imgur.com/govhIrK.jpg?1

Current arguments are:

Up next, going to experiment with...

• Lowering the voltage via TRIXX to see if I'm dealing with an overheating and therefore throttling VRM issue.
• Ways to increase my WU/s and see if I'm actually getting my money's worth out of my hashrate (if hashrate is doing awesome but WU/s is low, what's the point??)
• Ways to reduce rejected shares (what's the point of having an awesome hashrate if the pool is rejecting your work...??)

Comments always welcome. Thanks for taking the journey with me!

Thanks CryptoMer, this will be my next line of inquiry with my on-going tests.

I have never specified the kernel just start the script / bat with  ./sgminer under linux or just sgminer under windoze . Rest looks ok assuming you run three cards. Try 8191 for TC. worked for me and got me into the 720's

I found it but  now i have another error

Can someone please tell me where I'm getting a problem with this .bat file using sgminer if it matters

color 02
timeout /t 30
setx GPU_MAX_ALLOC_PERCENT 100
setx GPU_USE_SYNC_OBJECTS 1
sgminer kernel : scrypt130302 -I 13 --thread-concurrency 8192,8192,8192 --lookup-gap 2 -g 2 -w 256 --gpu-engine 1065,1065,1065 --gpu-memclock 1500,1500,1500 --temp-cutoff 90,90,90 --temp-overheat 85,85,85 --temp-target 68,68,68 --auto-fan --expiry 1 --scan-time 1 --queue 0 -o stratum+tcp://eu.pickaxe.cc:3330 -u Fortuneseeker.warflow1 -p kensei79

I've taken the settings from this guy

https://litecointalk.org/index.php?topic=12369.0;prev_next=next#new < ---this is this thread here I've been lazy and copied this post from anther thread

though i have no idea what the expiry/scantime/queue bit do. but he has the same cards exactly as i do and is getting 700 + hash.  anyway I get the "unexpected extra commandline arguments" error.  I took out the expiry/scan-time/queue parts and got the same thing. I noticed that each card has its own setting and though he has different ones for each card for gpu engine as i'm starting i was keeping mine the same.

any help please

try this one. Has scrypt-jane support patched in as well:

https://bitcointalksearch.org/topic/m.5114275

Where did you get sgminer from?  I keep reading about 'building' it and feel out of my depth.  I think i've downloaded it but it doesn't have as many files in the directoy as cgminer does making me think i've missed something, even though the program opens at least. Also did the .cl files just get copied into the sgminer directory?

finding this very helpful if a little over whelming for such a newbie!

I have a BUNCH of Elpida 280x. They're Gigabyte, not Sapphire, tho. We have them hashing at 740-755kh/s. Granted the Hynix ones do 775+, but you should be able to do more than an even 700.

I've got some Sapphire 280X Dual-X cards as well, though I don't recall if they're from the 20G batch or 25...

In any event, as you're already on Windows, download GPU-Z and check the VRM1 temperature (it's at the very bottom of the graphs). This run of card has a throttles itself as soon as the VRM hits 110C, which it will do on stock settings, even if it's all alone in an open-air case...

My fix: Undervolt the stupid things to 1085mv, run at 900 mem clock, space them out at least three inches from each other, and push air over them via a box fan. I have NO idea how these cards are going to perform in the summer, but I'm fearful.

My steady KH per card: 730

I think the Elpida RAM is lower quality than Hynix, at least that's what I've read somewhere on some forum. And lucky me (sarcastic) ofc I got the Eplida RAM in my GPU.

Quick update, I noticed today that my high performing card (GPU 2) has SK-Hynix based RAM, the other three lower performers use Elpida. Not sure what this means for me going forward, but it's a clue. Starting to feel like a detective over here Input welcomed!