atp1916
[H]ard|DCoTM x1
- Joined
- Jun 18, 2004
- Messages
- 5,007
Hurin,
These modded AMD drivers will work with 7 cards, allegedly.
These modded AMD drivers will work with 7 cards, allegedly.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
GTX750Ti... wait a sec...
- Thread starter SunnyD
- Start date
Just a note... When cudaminer gets error vomit, the cards that didn't error are still mining even though the cudaminer cmd box is flailing. Found that out when my hash would drop in wafflestats while I was at work, say from 1230 avg to 900 avg (I'm only running four 750Tis).
A crash on a particular card or cards will reset the OC sliders in precisionX back to 0/0 for the affected ones, and the others will stay where you set them and keep mining. I'd get home and could find what cards crashed by seeing which card/cards had their clocks set back to 0core/0mem, stop cudaminer, set clock for the errored card or cards a bit lower, and fire back up.
. . . for nvidia cards? (or are you suggesting I mod nvidia drivers similarly?)
I'm a bit obsessive. I know I can get six cards working in Windows 7. Going to try Windows 8.1 again today to go for seven cards. I updated my BIOS between Win8.1 issues and things working in Win7. So there's some hope I'll have a better Win8.1 experience this time.
There was a request for pics. . . here's my previously posted open-air rig with all seven cards hung. Though I can currently only get six to function (see above). All but one are unplugged at the riser and power connector right now while I install Windows 8.1 (again).
So far, my favorite part about this rig is that without the video cards running, this Haswell-based system with an i5-4570T in it uses only 40w of power. That's down from the i7-920 system I was using which had a base power draw of about 150w.
So far, my favorite part about this rig is that without the video cards running, this Haswell-based system with an i5-4570T in it uses only 40w of power. That's down from the i7-920 system I was using which had a base power draw of about 150w.
Update. . . Windows 8.1 installed. Had the same behavior of widely varying hashrates between cards and each run of cudaminer resulting in the hashrates changing randomly. That was addressed (duh) by running the x86 version of cudaminer instead of x64. This wasn't necessary on Windows 7.
But, unfortunately, I ran into the same wall with Windows 8.1 where getting the seventh card working is concerned. Everything is peachy up to card six. Once I install the seventh card, it successfully installs the driver, but won't let the card run. I get the exclamation point over it in Device Manager with the notice: "The device cannot find enough free resources that it can use. (Code 12)"
So, I guess I'll just have to settle for six cards.
One question for draksia (please), how much RAM do you have in your system and how much video RAM is on each of your cards? I don't think this should matter. But I'm now grasping at straws.
--H
But, unfortunately, I ran into the same wall with Windows 8.1 where getting the seventh card working is concerned. Everything is peachy up to card six. Once I install the seventh card, it successfully installs the driver, but won't let the card run. I get the exclamation point over it in Device Manager with the notice: "The device cannot find enough free resources that it can use. (Code 12)"
So, I guess I'll just have to settle for six cards.
One question for draksia (please), how much RAM do you have in your system and how much video RAM is on each of your cards? I don't think this should matter. But I'm now grasping at straws.
--H
I ahve 4gb in the system and 2gb on each card but I am only using 6.
I also can't get x64 cudaminer to start I get a missing MSVCP100.dll error even though I installed the visual c++ run time stuff.
Spent a couple of minutes (only) to do some light OCing. +135 core, +500 memory.
Now mining with six cards at those settings.
1655 KH/s
445w at the wall via kill-a-watt reading
EDIT: Now at 1750 KH/s with 450w at the wall.
Very happy with it. Sorta missing that seventh card though. Was hoping for 2 MH/s at only 500w. But I guess ya can't have everything. =)
To contrast, my two R9 290s were giving me about 1650 (1500 WU) for 750w. And they were loud, and served as space heaters. With these GTX 750s, I hardly know they're there. There's only the slightest whisper from the fans. I have the stock/standard/reference EVGA model with no PCI-E 6-pin or 8-pin power.
Now mining with six cards at those settings.
1655 KH/s
445w at the wall via kill-a-watt reading
EDIT: Now at 1750 KH/s with 450w at the wall.
Very happy with it. Sorta missing that seventh card though. Was hoping for 2 MH/s at only 500w. But I guess ya can't have everything. =)
To contrast, my two R9 290s were giving me about 1650 (1500 WU) for 750w. And they were loud, and served as space heaters. With these GTX 750s, I hardly know they're there. There's only the slightest whisper from the fans. I have the stock/standard/reference EVGA model with no PCI-E 6-pin or 8-pin power.
Last edited:
Brackle
Old Timer
- Joined
- Jun 19, 2003
- Messages
- 8,594
Ouch. I get around 1580kh/s with 2 280x's at 550w. Still not worth it if you need 6 cards to get 1655kh/s. Sure you save around 100w and alot less heat, but using up 4 more pci-e slots......that just hurts if you ask me. Still not worth it, now big die maxwell is another story.
Yeah I have tried installing the run times for both but for what ever reason the x64 doesn't work but I don't seem much difference in hash rate that people have posted.
Couple things. . .
You're at 550w at the wall? Whole system? Undervolted? I had two R9 290s and was pulling 750w from the wall for only about 1650KH/s (1510 WU). Granted some of that was the x58 chipset and i920. What base system are you using? You've got some really nice efficiency going there.
While it doesn't make sense in your situation, this switch is saving me about 300w and increasing my hashrate.
Hopefully the hashrate will improve when the PCI-E bandwidth issue with cudaminer is addressed. Those not running on risers tend to see 300+ per card. I'm at more like 275. So in the end I'm hoping for 1750-1800KH/s at around the same wattage.
It's not all due to Maxwell, but going from 750w to 450w is making a huge difference in how profitable altcoin mining is for me. I pay $.32/kwhr. It's rough. So every watt counts.
Been meaning to post the full setup. . .
Intel i5 4570T
MSI Z87-G55
16GB Mushkin RAM (yes, overkill)
six EVGA GTX 750 Ti (reference design, no additional PCI-E power)
These USB Powered Risers
Corsair AX1200i PSU (major overkill, but ready for future Maxwell, etc.)
Open Air Frame Custom-Build (see images earlier in thread)
Obviously, that was not a bargain build. But I'm not fully committed to mining yet and I wanted to have some decent components for potential other uses in the future.
--H
As an eBay Associate, HardForum may earn from qualifying purchases.
This may also a function of what card you have, but from my experimentation bandwidth is definitely an issue. On evga FTWs, which have a +169mhz stock core OC out of the box (1189 core vs 1020 reference core). I've found that increasing memory clock trumps core clock by a large margin in gaining hash. Possibly because of the already significant core OC on these though.
I lowered core OC from +35 to zero on two that I had an additional core OC on over stock, and the total hashrate stayed exactly the same. The two cards that had been running zero additional core OC other than stock all along gained hash equivalent to the small amount of hash lost on the cards whose core OCs I set back to zero, with no changes. Definite sign of bandwidth saturation.
I think because of the same issue, -H cuts my hashrate significantly, probably because of the additional bandwidth it's trying to use communicating with the cpu. After some fooling for long-term stability, the only argument in my batch is T5x24 and core/mem +clocks in precisionX and avg hash are:
0/585 - previously running 35/585 - 325 kh/s in x16 slot
0/585 - previously running 35/585 - 315 kh/s on 1x to 16x powered riser
0/450 - 295 on 1x to 16x powered riser
0/425 - 290 on 1x to 16x powered riser
Each card adds 80W at the wall.
For reference, one card alone in an x16 slot at +35/+700 core/mem was about 335-340 kh/s and added 85W at the wall.
Averages around 1220 kh/s total with CPU usage ~10% and system mem usage ~1.6GB
Pretty sure if I were running six cards as well avg hash per card would be lower. These cards are great from an efficiency standpoint, but unless it's addressed I see the pcie bandwidth issue being a severe bottleneck for Big Maxwell. If it's fixed these cards will fly.
My numbers aren't quite as precise or illuminating. But they are oddly and substantially lower despite clock speeds probably being similar.
According to MSI Afterburner. . . all six cards linked at +135core/+500mem
Core Speed jumps around on each of the cards between 1255 and 1311 MHz
Memory speed for each card is 3198 MHz
Voltage jumps around constantly between 90-120%
Hashrates are between 265 and 280 on each card.
According to MSI Afterburner. . . all six cards linked at +135core/+500mem
Core Speed jumps around on each of the cards between 1255 and 1311 MHz
Memory speed for each card is 3198 MHz
Voltage jumps around constantly between 90-120%
Hashrates are between 265 and 280 on each card.
Well I'll be. . . 
I really didn't put much stock in the whole "run chrome (preferably with a flash video/game running)" tip. Especially because I'm using the Intel Haswell-integrated video for my display.
But, while running my stability test (settings in post above), I decided to give it another try. . .
Instantly, my voltages and clock speeds stopped fluctuating so much. . . and I'm now at 1750 KH/s and each card is now at about 290 KH/s.
Power from the wall is up 2-5w to 450w now.
P.S. The theory behind the chrome trick is that it stops nvidia's power-saving schemes from kicking in because it triggers "3D gaming/application mode" of some kind.
I really didn't put much stock in the whole "run chrome (preferably with a flash video/game running)" tip. Especially because I'm using the Intel Haswell-integrated video for my display.
But, while running my stability test (settings in post above), I decided to give it another try. . .
Instantly, my voltages and clock speeds stopped fluctuating so much. . . and I'm now at 1750 KH/s and each card is now at about 290 KH/s.
Power from the wall is up 2-5w to 450w now.
P.S. The theory behind the chrome trick is that it stops nvidia's power-saving schemes from kicking in because it triggers "3D gaming/application mode" of some kind.
Well I'll be. . .
I really didn't put much stock in the whole "run chrome (preferably with a flash video/game running)" tip. Especially because I'm using the Intel Haswell-integrated video for my display.
But, while running my stability test (settings in post above), I decided to give it another try. . .
Instantly, my voltages and clock speeds stopped fluctuating so much. . . and I'm now at 1750 KH/s and each card is now at about 290 KH/s.
Power from the wall is up 2-5w to 450w now.
P.S. The theory behind the chrome trick is that it stops nvidia's power-saving schemes from kicking in because it triggers "3D gaming/application mode" of some kind.
Yeah I have my batch and crome in the startup folder for windows so it auto loads.
I definitely notice a large improvement with it open.
Well I'll be. . .
I really didn't put much stock in the whole "run chrome (preferably with a flash video/game running)" tip. Especially because I'm using the Intel Haswell-integrated video for my display.
But, while running my stability test (settings in post above), I decided to give it another try. . .
Instantly, my voltages and clock speeds stopped fluctuating so much. . . and I'm now at 1750 KH/s and each card is now at about 290 KH/s.
Power from the wall is up 2-5w to 450w now.
P.S. The theory behind the chrome trick is that it stops nvidia's power-saving schemes from kicking in because it triggers "3D gaming/application mode" of some kind.
Yep it's a definite difference even when not experiencing the fluctuation problems you did. My four cards ran fine without chrome, but I'd ballpark the hash increase due to the Flash content at ~3-5%. A modded bios usually won't let the cards throttle so running chrome does nothing for those folks, but we're unmodded and letting them do what they want so they're going to try to throttle at any opportunity.
On a somewhat similar note, using x86 cudaminer is <1% gain for me over 64-bit so for that I'd say use whichever appears more stable over time. I'm still using x86 only because it hasn't caused my 4-banger any problems.
*edit* Almost forgot in the midst of all the technobabble... Congrats! 1750kh/s for 450w is about the best I've seen. Sounds like you'll have a great rig when you get it stable.
Last edited:
Thanks! I'm getting quite happy about all this now. I can only imagine how happy I'll be when/if (as promised) the bandwidth issues are addressed in cudaminer.
I upped the memory OC to +600 (3300MHz). This makes the 1750 KH/s a bit more reliable (it will dip sometimes).
I BSODed (!) at 610 mem OC. So 600 it is. Not sure if I'll ever have the patience to dial in each card individually.
Couple questions if you don't mind. . .
Do you not see those voltage fluctuations at all? Looking at MSI Afterburner graphs now, I'm seeing some variation at 95%-100%. But it's much better with chrome. Your graphs are just flat lines?
Is the total KH/s reading that cudaminer gives us on a "yay!!!" line essentially the same thing as what cgminer would report as WU? Or should we assume some hashrate is lost and that our actual WU is a bit lower than what we see there?
Best,
H
Vengance_01
Supreme [H]ardness
- Joined
- Dec 23, 2001
- Messages
- 7,238
Keep posted these tips guys! I got 6 750TI cards from TD coming pny brand.
Do you not see those voltage fluctuations at all? Looking at MSI Afterburner graphs now, I'm seeing some variation at 95%-100%. But it's much better with chrome. Your graphs are just flat lines?
PrecisionX doesn't graph volts, just GPU power, but the volt readout here never deviates from 1136mV, nor does the freq from 1332. GPU power doesn't fluctuate more than 1% between 89-90%. If you are actually bumping your 100% load limit, that will cause fluctuation.
Here's a super crappy phone pic...
Having just gotten back into mining after a couple years out the only current mining knowlege I have is nvidia/cudaminer, but I would assume it's more or less the same. I'm on wafflepool and my hash actually works out higher per their numbers - about 1300kh/s avg. Hopefully one of our AMD heavies can jump in and answer this better.
Hope these shaky-jake pics come out readable. The avg accepted rate of 94.5% is a little lower than it usually runs on wafflepool. I typically get about 96%
Apologies for the huge pics, had to make them big enough to see due to bad quality on my part
buttons
2[H]4U
- Joined
- Oct 12, 2011
- Messages
- 2,174
I had six r9 290s. my rig with 2 290x got 1830kh from 805 watts at the wall. that was an X58 system but with a 60 watt xeon six core L5639. I just got my first two 750TI and its drawing 80 watts at idle 235 watts under load. im not super impressed with it. 550kh / 235 watts $320 for the two cards.
I really wanted to buy 8 R7 265's and see what I could achieve with an undervolt but who knows when if they are coming back in stock.,
Edited that post. Now getting 1750 KH/s for 450w at the wall.
Can't you lock this in the nvidia control panel?
Interesting idea. . . I wanted to go in and see if I could make the cudaminer executable a known 3D app to the nvidia control panel, and then specify things like "Power Management Mode". . .
But alas, I'm running my monitor off my integrated Intel video and not one my GTX 750 Ti cards. So nvidia control panel won't open.
Might still be worth investigating though. But for now, I like using the Intel video since the computer remains useable and tweakable while mining.
Parja
[H]F Junkie
- Joined
- Oct 4, 2002
- Messages
- 12,670
It graphs whatever you want. Right click in the graph area and click Properties. Then you can redefine what PrecisionX puts in those three slots (the top 3 things graphed). Or you can just double click on the graph and it'll bring up a pop-out with everything graphed.
Srsly? A cell phone pic of a computer screen? We're going to need you to turn in your [H] card. Try Alt-Print Screen.
I was trying to remember how I got Precision X to do this so I could share. Thanks!
Personally, I have my mining rig semi-quarantined. So getting the screenshot back off of the computer after the printscrn isn't trivial. Maybe he has similar situation. In any case, get ready for more camera pics cuz I just snapped a few shots myself.
I'm on waffle as well. I'm at 93.64% accepted after an overnight run. So that sounds about right. Though less than I was getting when I was running two R9 290s.
BallaTheFeared
n00b
- Joined
- May 4, 2012
- Messages
- 36
I was getting around 96% with my 780 on clevermining, I think the switching pools have slightly higher rejects but overall higher payout assuming you never intended to hold alt coins.
As far as Precision X, it doesn't properly read voltage it just gives VID and it won't actually monitor voltage. MSI AB works best for software derived voltage, because it reads actual voltage with droop and not VID.
For CP when running off iGPU for me I got CP up in this situation (gaming on HD4600, mining on 780). If you can't get it to come up try using a dummy plug on the top card.
As far as Precision X, it doesn't properly read voltage it just gives VID and it won't actually monitor voltage. MSI AB works best for software derived voltage, because it reads actual voltage with droop and not VID.
For CP when running off iGPU for me I got CP up in this situation (gaming on HD4600, mining on 780). If you can't get it to come up try using a dummy plug on the top card.
The developer of cudaminer reports that his risers arrived yesterday. So hopefully he will find time (he does this for free, after all) to look into the PCI-E bandwidth issues soon.
I think I've probably squeezed all I can from these cards and I'm now up against the bandwidth bottleneck. So I can't wait to see what I can get out of this rig when/if that issue is resolved.
I do plan to donate to the developer ASAP. Others enjoying this nvidia mining renaissance might consider doing the same (but I ain't judging! sorry if that sounds preachy).
I think I've probably squeezed all I can from these cards and I'm now up against the bandwidth bottleneck. So I can't wait to see what I can get out of this rig when/if that issue is resolved.
I do plan to donate to the developer ASAP. Others enjoying this nvidia mining renaissance might consider doing the same (but I ain't judging! sorry if that sounds preachy).
One word of caution. . .
Feel your molex/SATA power cables. One of the SATA cables coming off my Corsair 1200W was very hot. Almost too hot to touch. Only then did I realize that I had four of my cards/risers plugged into one cable, and two on another. The one with two was cool to the touch.
Just to be safe, I plugged in a third cable, and now have two cards on each of the three cables. Hopefully that keeps them nice and cool.
Probably wasn't actually a fire hazard. But not ideal either.
Feel your molex/SATA power cables. One of the SATA cables coming off my Corsair 1200W was very hot. Almost too hot to touch. Only then did I realize that I had four of my cards/risers plugged into one cable, and two on another. The one with two was cool to the touch.
Just to be safe, I plugged in a third cable, and now have two cards on each of the three cables. Hopefully that keeps them nice and cool.
Probably wasn't actually a fire hazard. But not ideal either.
Best to avoid heat were possible. I just replaced a 16/2 extension cable with a 12/2 extension cable. 10+ amps through a 100ft 16 awg was getting uncomfortably hot. 12 awg is cool as ice.
That's pretty much all I'm waiting for. I like the 750 rig, but its 24/7 stable now and I pretty much consider it a wrap unless buchner pulls off something major with cudaminer tweaks. It was a good way to get my feet wet with scrypt but won't continue building rigs with cards this small. Trying to wait for big(ger) Maxwell because I like the efficiency, but may get tired of waiting and go red for the next one.
I convinced myself not to go balls deep with this with the current rate of return a few weeks ago, so in the end it'll be whatever I can get on one 15A ckt and possibly part of another.
so I got my 15 cards today, unboxed one of the MSI 3Gaming 750 TI's to see what I could do out of the box, looks like stock it came out around 265 khash, so with a little tweaking was able to get this particular card up to 275-290 khash...I was curious what setting people were using within Afterburner with the card staying stable for 24 hours.....and has anybody modified the bios yet to see if that nets them better results, if so what modified settings.
deathbyburrito2
n00b
- Joined
- Dec 28, 2009
- Messages
- 30
I picked up two of the Gigabyte cards. Using Afterburner just set the core clock +100 and the memory +500. That worked out to about 290 kh/s per card, but it was up against the power limit (38.5). I updated the bios to 65.5 and hash rate jumped to 308 kh/s per card. Lose about 10 kh/s per card if Chrome is not running.
I tried +31 mV, +170 core, +575 memory. That gave about 322 kh/s per card, but I haven't had time to check the stability of those settings. Also haven't hooked it up to the Kill a Watt yet.
I tried +31 mV, +170 core, +575 memory. That gave about 322 kh/s per card, but I haven't had time to check the stability of those settings. Also haven't hooked it up to the Kill a Watt yet.
Ok I did the quick bios mod, and updated to the new 335.23 drivers that unlock the GPU clock....Im running stable right now at 305 khash with 175/550...will up the GPU Clock after 24 hours of stability to see if I can get this MSI 750 TI a really cooking, but so far its handled everything I have thrown at it....will test the EVGA SC later this week.
Thanks to everyone who posted their experiences with these cards!
I just got my EVGA Ti's in yesterday, and because of all of the info I learned from this thread I am hashing at 290 per card!
Windows 8, +150 core, +500 memory, command line arguments -i 0 -l T5x24 -C 1, and Chrome running in the background. At 53 degrees C, the fans are barely audible at 46%.
I just got my EVGA Ti's in yesterday, and because of all of the info I learned from this thread I am hashing at 290 per card!
Windows 8, +150 core, +500 memory, command line arguments -i 0 -l T5x24 -C 1, and Chrome running in the background. At 53 degrees C, the fans are barely audible at 46%.
Those numbers (per card) look exactly like mine. Though I think you'll find those numbers will be sustained with less of an overclock. You're likely up against the PCI-E bandwidth issue (hopefully to be fixed in cudaminer soon).
I see the same numbers at +132 and 500 OC.
I see the same numbers at +132 and 500 OC.