I use the INNO3D 2 slot 4070Ti. I bought it specifically because it's a 2 slot card. The 4070Ti Super is the same. It's a great card, I've had no issues although I'm struggling to get any good results out of it with E@H currently.
There's not many 2 slot cards these days.
Hi Chooka!
I agree with you that there are not many 2-slot cards available these days. I've read your posts with interest, and since you mentioned INNO3D (which I never heard of until you mentioned it) I did some checking on them because I also would love to have 2-slot GPUs in my systems - as long as they perform well.
The 4070 Ti (and other INNO3D GPUs) would appear to be an okay GPU for gaming and for normal usage, but not for over clocking as their VRMs are subpar and tend to breakdown when pushed.
We in GPU Users Group, Tom included, tend to push our GPUs harder than most people. If I may ask:
How long have you owned your INNO3D 4070 Ti ?
Do you over clock your 4070 Ti ?
Do you notice any Coil whine?
Does the 4070 Ti run cool enough for you?
Have you had any problems setting up the 4070 Ti for running BOINC projects?
All-in-all, I'm interested in getting an NVIDIA 40XX series card, and may eventually do so with all three of my computers because of the newer NVIDIA's Ada Lovelace generation of GPU. The power requirements have dropped considerably. Plus, they are a bit faster.
BTW, I run Linux Ubuntu 22.04 and I see that you are on Windows 10 Pro which I felt was pretty good too, until Windows decided to continue doing constant upgrades, etc. It became a hassle for me so I made the switch to Linux at the urging of my teammates and I never want to go back.
I did my homework and INNO3D is the only 40 series in a 2 slot design. Just quickly, I've been testing some O3AS work units and here's the results (with my cpu NOT crunching other work)
1 x wu = 915sec
2 x wu = 585sec
3 x wu = 700sec (although the run times REALLY varied at x3. I noticed that 2 wu's were quite close together as I guess I didn't stagger these so some wu's finished in 440sec/wu and others were 800sec/wu so I just took an average and called it 700sec/wu.
Edit - So staggering the wu's better changed the run times to more like 540sec/wu.
Now trying 4 x wu's.
I've just staggered the wu's at 3 x and I'll see if that smooths out the run times. I'll then try X4 but already getting the feeling that might be pushing it. FYI, running 3 wu = 9335Mb of memory being used so getting close to the 12gig on my card.
To answer your questions -
How long have you owned your INNO3D 4070 Ti ? - I have 2 and I've had both since Jan 24.
Do you over clock your 4070 Ti ? - No In fact I undervolt them if I'm running something intensive like Primegrid. Works great!! Video here - https://youtu.be/lJG7sQPdDIs?si=GQQMjqQrX_n8X8ur
I didn't think there was much headroom for overclocking these newer cards.
FYI - The 4070Ti is MUCH more efficient than my 3070Ti.
Do you notice any Coil whine? - No I don't think so.
Does the 4070 Ti run cool enough for you? - Absolutely. Especially undervolted. Right now doing 3 x wu on O3AS is at 58 degrees C. I've no issues with temps.
Have you had any problems setting up the 4070 Ti for running BOINC projects? - No. I use it on E@H, Primegrid, Numberfields & WCG when I can get work. Runs great.
Have you needed to send your 4070 Ti in for an RMA? - Nope. No issues with either of my cards.
Yeah I stick with Windows becuase Linux isn't user friendly enough for someone like me who's grown up with Windows. I use Linux on my VM sometime when running LHC and I found it a long nightmare to get that up and running. I'm ok with it now but it wasn't much fun with all that sudo stuff.
Now that Milkyway has retired gpu work, I don't do all that much gpu crunching. There's not that many gpu projects that I'm interested in (not that many full stop) Electricity prices here is Australia are killing me so I've had to cut back, hence I haven't crunched E@H during summer.
I'm more inclined to upgrade cpu core in future like a EYPC. (Can't see that happening any time soon though.... cost of living is tough atm)
Hopefully I've helped answer some things for you. Happy to help with any more questions :)
You’re talking about Gaurav. To my knowledge he had regular Radeon VIIs not the Pro version. And the Radeon Pro VII isn’t a dual GPU board. It’s basically the same single GPU as the normal RVII, but with more FP64 capability.
what he had was a “Pro” version of some other older AMD GPU (like Fiji era or something) that was a dual GPU board. (4x cards, BOINC reported 8)
You’re talking about Gaurav. To my knowledge he had regular Radeon VIIs not the Pro version. And the Radeon Pro VII isn’t a dual GPU board. It’s basically the same single GPU as the normal RVII, but with more FP64 capability.
what he had was a “Pro” version of some other older AMD GPU (like Fiji era or something) that was a dual GPU board. (4x cards, BOINC reported 8)
Just wondering if anybody is using those? & how does BOINC sees those multi-GPU chips? ????♂️
people have run them before, as with many other kinds of multi-GPU boards.
As far as BOINC is concerned, it treats each GPU core as it's own GPU individually, since that's how the board is really setup. it's just two (or more) GPUs with their own local VRAM on one board and usually with some kind of PCIe bridge on board to expose all devices over a single PCIe link, they aren't hardware linked to work together on a single task/application. it's only the software that would take advantage to use them together in some way.
The multi-gpu GRID and Tesla cards were designed mainly for cloud streaming and VDI applications, not really for heavy compute.
you can use them for BOINC just fine (if the application supports such old GPUs), but they will be seen and N number of GPUs, not a single GPU.
I have a Gainward 4070TiS, it is a great card. I undervolted it to 0.9V @2550MHz, crunches O3A tasks at around ~100-105W, 30-40 mins / 3 tasks. Completion time depends on how staggered the the tasks are. I run 3 tasks concurrently, CPU is 3950x and I run WCG tasks on the CPU (allowing 75% of the cores to crunch). I will try suspending WCG for a few hours to see how CPU affects performance. BRP5 Merkat tasks takes around 5 min. (1 task runs at a time).
I have a Gainward 4070TiS, it is a great card. I undervolted it to 0.9V @2550MHz, crunches O3A tasks at around ~100-105W, 30-40 mins / 3 tasks. Completion time depends on how staggered the the tasks are. I run 3 tasks concurrently, CPU is 3950x and I run WCG tasks on the CPU (allowing 75% of the cores to crunch). I will try suspending WCG for a few hours to see how CPU affects performance. BRP5 Merkat tasks takes around 5 min. (1 task runs at a time).
This is interesting to read, as I have a 4070Ti paired with a 3050X which is crunching WCG :) Currently I suspended WCG to do this testing (on my 3900X system anyway)
I think using 75% of the threads should be ok but I'd expect slightly longer run times while using a cpu. Got to be a bit of a trade off to use both :)
30-40min for 3 x O3A tasks is about what I'm seeing too but with no cpu work.
Oh as a side note, I tried 4 x O3A on the 4070Ti last night. Again, times vary from 33-35min (525/wu) which is pretty good. Memory used = 11.6gig so I'd say 4 wu's is as far as I can go on a 12gig card. (I made an error when I posted 8 gig used yesterday.)
GWGeorge007 wrote:Chooka
)
Hi GWGEORGE007,
I did my homework and INNO3D is the only 40 series in a 2 slot design. Just quickly, I've been testing some O3AS work units and here's the results (with my cpu NOT crunching other work)
1 x wu = 915sec
2 x wu = 585sec
3 x wu = 700sec (although the run times REALLY varied at x3. I noticed that 2 wu's were quite close together as I guess I didn't stagger these so some wu's finished in 440sec/wu and others were 800sec/wu so I just took an average and called it 700sec/wu.
Edit - So staggering the wu's better changed the run times to more like 540sec/wu.
Now trying 4 x wu's.
I've just staggered the wu's at 3 x and I'll see if that smooths out the run times. I'll then try X4 but already getting the feeling that might be pushing it. FYI, running 3 wu = 9335Mb of memory being used so getting close to the 12gig on my card.
To answer your questions -
I didn't think there was much headroom for overclocking these newer cards.
FYI - The 4070Ti is MUCH more efficient than my 3070Ti.
Yeah I stick with Windows becuase Linux isn't user friendly enough for someone like me who's grown up with Windows. I use Linux on my VM sometime when running LHC and I found it a long nightmare to get that up and running. I'm ok with it now but it wasn't much fun with all that sudo stuff.
Now that Milkyway has retired gpu work, I don't do all that much gpu crunching. There's not that many gpu projects that I'm interested in (not that many full stop) Electricity prices here is Australia are killing me so I've had to cut back, hence I haven't crunched E@H during summer.
I'm more inclined to upgrade cpu core in future like a EYPC. (Can't see that happening any time soon though.... cost of living is tough atm)
Hopefully I've helped answer some things for you. Happy to help with any more questions :)
Ian&Steve C. wrote: You’re
)
7990.
https://www.techpowerup.com/gpu-specs/radeon-hd-7990.c2126
Chooka wrote: Ian&Steve C.
)
no, that's too old and not what I'm referring to.
This system: https://einsteinathome.org/host/12219055
8x Fiji 4GB = likely to be 4x AMD Pro Duo cards: https://www.techpowerup.com/gpu-specs/radeon-pro-duo.c2828
_________________________________________________________________________
KLiK wrote:To be more
)
people have run them before, as with many other kinds of multi-GPU boards.
As far as BOINC is concerned, it treats each GPU core as it's own GPU individually, since that's how the board is really setup. it's just two (or more) GPUs with their own local VRAM on one board and usually with some kind of PCIe bridge on board to expose all devices over a single PCIe link, they aren't hardware linked to work together on a single task/application. it's only the software that would take advantage to use them together in some way.
The multi-gpu GRID and Tesla cards were designed mainly for cloud streaming and VDI applications, not really for heavy compute.
you can use them for BOINC just fine (if the application supports such old GPUs), but they will be seen and N number of GPUs, not a single GPU.
_________________________________________________________________________
Chooka
)
Yes, you've answered all of my questions regarding your 4070 Ti's ! Thank you!
I switched to Linux because I wanted to learn new things and I got tired of Windows shenanigans.
And yes, prices are jumping up more everywhere, for everything, including the electricity prices.
Good luck with your 4070 Ti's !
Proud member of the Old Farts Association
I have a Gainward 4070TiS, it
)
I have a Gainward 4070TiS, it is a great card. I undervolted it to 0.9V @2550MHz, crunches O3A tasks at around ~100-105W, 30-40 mins / 3 tasks. Completion time depends on how staggered the the tasks are. I run 3 tasks concurrently, CPU is 3950x and I run WCG tasks on the CPU (allowing 75% of the cores to crunch). I will try suspending WCG for a few hours to see how CPU affects performance. BRP5 Merkat tasks takes around 5 min. (1 task runs at a time).
stsfred wrote: I have a
)
This is interesting to read, as I have a 4070Ti paired with a 3050X which is crunching WCG :) Currently I suspended WCG to do this testing (on my 3900X system anyway)
I think using 75% of the threads should be ok but I'd expect slightly longer run times while using a cpu. Got to be a bit of a trade off to use both :)
30-40min for 3 x O3A tasks is about what I'm seeing too but with no cpu work.
Oh as a side note, I tried 4 x O3A on the 4070Ti last night. Again, times vary from 33-35min (525/wu) which is pretty good. Memory used = 11.6gig so I'd say 4 wu's is as far as I can go on a 12gig card. (I made an error when I posted 8 gig used yesterday.)
FYI - A 3070Ti cannot do 4
)
FYI - A 3070Ti cannot do 4 wu's at a time! Times blew WAY out. Over 1hr.
3wu's it is. :)
Chooka wrote: FYI - A 3070Ti
)
TY for the guidance.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
tried to run 2x Meerkat tasks
)
tried to run 2x Meerkat tasks at the same time, but there is no benefit. Reverted back to 1.
What I noticed is the GPU consumes 120-130W crunching Meerkat tasks, while 3x O3A tasks consume around 105W. 4070TiS undervolted to 0.9V @ 2550MHz.