Not all GPU's are equal. You might be able to get 745 core using 0.95v, but only 730 @ 1v on another core. You probably have clocks set too high. Find out which gpu is crashing and reduce clocks by 5 MHz. If it happens again, reduce by another 5 and so on. If you get to the point of a really low clock, like 670, try increasing voltage to 1v and core to 750 and repeat the process. It's what I do. It -may- cause you to lose a little mining time, but at least you can find out the max your GPU's can tolerate.
Also optimum worksize for 5xxx cards (assuming it uses phatk kernel) is usually dependant on memory speed. 128 worksize is optimum with 150-220 memory (depending on core speed. Higher core = higher memory. 700 core is optimum with about 150 memory, 1050 core is optimum with about 220 memory). 256 worksize has optimum memory ranges around 300-380 (also reliant on core speeds. lower core = lower memory, higher core = higher memory). It takes some experimenting to find out the optimum combo.
Generally speaking, I found that hashrate is directly proportional to core speed * shaders. I get 228.7 mhash on a 5830 @ 700 MHz. 5830 has 1120 shaders, so I get 0.0002917 mhash/sec per shader per MHz. Using that math, you should get about 338.4 mhash on a single 5970 core @ 725MHz
|
|
|
I'm not even sure that 4xxx cards can even mine effectively. I mean you can play around with settings and whatnot, but I don't know...
Also are you -sure- your 4850 is under load? It might've crashed and the driver had to reset. Usually when this happens, the card will still show a full load on it, but it won't actually be doing anything until you re-apply a load and then remove it.
Double check that you aren't actually cpu mining too.
|
|
|
It looks like you set the clocks of the wrong performance level. There are 3 performance levels (low medium high) depending on the load of the gpu. It's not smart enough to pick the highest frequency out of the 3, it just sets to perf levels and obeys the clocks of each one.
You probably changed clocks of your low or medium perf level.
Also not sure on this one, but your bat file is pretty long.. why not use a config file? It would be a lot more neater and human-readable.
|
|
|
Your screenshot links to a thumbnail. Please fix to regular screenshot?
|
|
|
Title. I have a fan that is making loud noises. Was told I need to replace the fan and thermal paste on the GPU.
Thank you.
Not necessary to replace the fan. I just bought a 5970 for $200 shipped that had a bad fan making loud noises too. Do what I do and take the sticker off the face of the fan and drill a TINY hole about 3-4 mm offcenter. Stop immediately when you feel the drill fall through. Put in about 5-7 drops of 3-in-1 multipurpose oil, wipe up the excess really good, replace sticker (or put a small piece of packing/invisible tape on the hole if there is no sticker or the sticker is ruined). Hand spin it a couple dozen times. Bam, you have a fan that performs as if it were new. This works for reference blower type fans, or the regular fans on non-reference cards (or any other DC computer fan really). Done this to about 6 different fans so far, all work like a charm. Ohhh, this is EXACTLY what I do. Did you read it on one of my posts?. I replace the sticker with duct tape, and then fan is good for at least 6 more months!. Quite possible. I was scared first time I did it but I'm pretty damn good at it now . I did see it on the forums somewhere. Whoever suggested prying the fan off of the base of the fan has cost me 3 fans, because I broke the fan each and every time; they did NOT want to come apart with a reasonable amount of force. The "drill a hole" way is much safer imo, and doesn't even require taking the card apart! Personally I wouldn't use duct tape though, unless you got it centered really good. Wouldn't want to throw off the weight distribution.
|
|
|
I have $95 MTGoxUSD. I want dwolla.
|
|
|
Title. I have a fan that is making loud noises. Was told I need to replace the fan and thermal paste on the GPU.
Thank you.
Not necessary to replace the fan. I just bought a 5970 for $200 shipped that had a bad fan making loud noises too. Do what I do and take the sticker off the face of the fan and drill a TINY hole about 3-4 mm offcenter. Stop immediately when you feel the drill fall through. Put in about 5-7 drops of 3-in-1 multipurpose oil, wipe up the excess really good, replace sticker (or put a small piece of packing/invisible tape on the hole if there is no sticker or the sticker is ruined). Hand spin it a couple dozen times. Bam, you have a fan that performs as if it were new. This works for reference blower type fans, or the regular fans on non-reference cards (or any other DC computer fan really). Done this to about 6 different fans so far, all work like a charm.
|
|
|
Wish I could afford the rosewill lightning! I need moar powerz!
Sucks being poor, I guess that's why I'm selling stuff. Also bump Well, I'm not exactly -that- poor. I got about $350 worth of video cards I need to sell (2x gigabyte 5850 non-reference, diamond 5830 non-reference) but I have 2x reference 5870 and 1 non-reference, but voltage fully adjustable, 5850 coming in the mail. If I sold my 3 cards, I'd probably jump on the rosewill. Right now I'm drawing 830-860W AC from a 750W HALE90 (80plus gold) and 900-930W AC from a PC P&C 1KW-SR (only 860W available on 12v. not 80plus at all either), and I'm getting sketchy behavior from my rigs; They don't turn off like I would expect, they just become unresponsive and start drawing only ~300 watts, have to hard reset to fix it. In the meantime, I jerryrigged an OOOLD PSU with 336W available on 12v and powered the weakest cards in each system (5850 @ 0.95v), -seems- to be ok so far.. Plus if I were to get the rosewill, then I could finally RMA my HALE90; I melted one of the molex cords and modular outlets because I didn't know better not to power 2 x Gigabyte 5850s (voltage hardlocked to 1.088v) @ 1000+ MHz across a single molex string. <.< Anyways, I'm rambling. Bottom line is I'm interested in the rosewill 1300w, and money will eventually come.
|
|
|
If you want nvidia video card and amd/ati video card in the same computer, you need to install both drivers.
Both video cards will work happily in the same system. Each card will render only to their own respective attached monitor(s) though. You could run 2 monitors with 1 connected to each video card and play 1 3d game on each monitor and both gpu's would be working on their own respective monitor. No idea what would happen if you dragged a windows 3d game between 2 monitors though.
Of course if your interest is gaming on 570 and mining on 5970, the same will apply. Install both drivers.
|
|
|
Wish I could afford the rosewill lightning! I need moar powerz!
|
|
|
Yes, volterra voltage regulators indeed have steps to their voltage settings. Don't be fooled by software overclocking utilities that say otherwise. You can find out what these steps are in radeon bios editor (RBE) under the clocks tab > GPU registers. For me, my steps are 0.95 > 1.0625 > 1.15 > 1.1625. If you flash a voltage between one of these steps, it will run at the next highest step (if you flash 0.97v, it will actually run at 1.0625v, confirmed by hwinfo).
Anyways, you can change operating voltage with Sapphire Trixx, or you can flash them with RBE. Afterburner will probably work too, but its a little more cumbersome to get it to work the way you need it to.
|
|
|
Damn, I was too slow
|
|
|
Hi. 5.3 ghash switching from eligius to here I've been mining at eligius for something like 8 months now, but the recent bad luck and rather large chunk of "extra credit" that I'm "owed" (technically I'm not owed anything, but if I were paid straight PPS, I'd be about 25 BTC richer. Thats how bad the luck has been) leads me to here, a true PPS pool with 0% fee. Almost seems too good to be true.
|
|
|
...Wow, all this bickering about how I chose the word "barrier". I'm SORRY that I used the wrong word. Sheesh. I corrected my post and deleted the word. Happy?
|
|
|
Wait are those the temps when your clocked at 600?
Because if that is, when your clocked at 900 the temps would be way higher.
He's not clocking to 600 on purpose. When it clocks to 600 core and 900 memory, it is in thermal throttle mode. He needs to fix his cooling/heat issues first by doing 1 or a combination of the solutions I gave earlier.
|
|
|
1 ghash mining bitcoins and 1ghash merged mining will get the exact same amount of bitcoins. with merge mining you will have the opportunity to get namecoins too, but you have to set up proper namecoin address and client and all that stuff. imo merge mining isnt worth the hastle. last I checked, it was maybe 0.5% extra bitcoins if you exchanged your namecoin for bitcoin
|
|
|
Yes, you are being thermal throttled. I have 5870's that did this in the winter (outside rigs, frozen heatpipes = skyrocketing temperatures = throttle). You need to find a way to cool your card better by doing 1 or a combination of these things...
Clean out your fan/heatsink Increase fan speed Run reduced core clocks. Run reduced core voltage.
You can also try re-greasing your heatsink, but you will need to get new thermal pads if it is a reference heatsink because just taking off the heatsink will ruin these pads. You should also already be underclocking your memory if it is a dedicated miner. 150-200 is good for worksize 128, and 300-380 for worksize 256 (slightly faster, but slightly more electricity)
Also I should note: When your card is thermal throttling, your fan speed profile will be changed to one specific to thermal throttling as well, which is about 55%, even if you fix the fan to a different value. You HAVE to cool the card better (or cause less heat by underclocking/undervolting the gpu core) to prevent this from happening in the first place.
Undervolting can do wonders for cooling a card, but the power difference from going from 160mem to 300mem is almost negligible. Hmm, think I will test that right now actually.. Going to note here 900 watts exactly with a 7 Cypress GPU rig and 2.36 ghash @ 150mem across (almost) all cards; 1 is 159 because it clocks really good at 0.95v. Lets see what happens with 256 worksize and adjusted memory... posting back in a bit edit: it'll take a good deal of memory tweaking to get the right memory timings.. Cant only do 2.34 ghash with worksize 256 and memory at 299-309 across low (725ish)/high (810) clocked cards. Power is at 915 watts too.
|
|
|
You should have 2 display devices in your device manager (1 integrated, the other discrete). Also, maybe you need to reinstall your opencl?.. try gpu caps viewer and see what your opencl tab says.
|
|
|
Yes, you are being thermal throttled. I have 5870's that did this in the winter (outside rigs, frozen heatpipes = skyrocketing temperatures = throttle). You need to find a way to cool your card better by doing 1 or a combination of these things...
Clean out your fan/heatsink Increase fan speed Run reduced core clocks. Run reduced core voltage.
You can also try re-greasing your heatsink, but you will need to get new thermal pads if it is a reference heatsink because just taking off the heatsink will ruin these pads. You should also already be underclocking your memory if it is a dedicated miner. 150-200 is good for worksize 128, and 300-380 for worksize 256 (slightly faster, but slightly more electricity)
Also I should note: When your card is thermal throttling, your fan speed profile will be changed to one specific to thermal throttling as well, which is about 55%, even if you fix the fan to a different value. You HAVE to cool the card better (or cause less heat by underclocking/undervolting the gpu core) to prevent this from happening in the first place.
also I only use trixx for software clock/volt management. Yes the auto-fan across multiple gpus was bugged last time I used it, but I run fixed fan speed anyways. I found trixx to be less cumbersome when selecting different cards compared to afterburner, which was a huge plus for me. I recently switched to doing bios flashing instead to set mining clocks without extra software on boot up, so now I only use trixx to find initial max stable core speeds for a gpu before I flash them.
|
|
|
Wow, what a slew of information! Can I tip you? Also went to home depot, and figured maybe I just want to stick with 115v. That way I can continue to measure with my kill-a-watts (I can't find any compatible 230v ones). Just means I have to run 2 circuits instead of one big fat one.
|
|
|
|