What kind of results you guys getting from the 1050ti. I saw these cards on eBay for $28.99. Too good to be real?

HOLY $h!t That seems too good a price to be real. Look at Stephen's host with the dual 1050 Ti's. Not bad at all with the special app.
https://setiathome.berkeley.edu/results.php?hostid=8222433&offset=0&show_names=0&state=2&appid=

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

What kind of results you guys getting from the 1050ti. I saw these cards on eBay for $28.99. Too good to be real?

Almost certainly too good to be true. Are they used/new? Are they shipping from overseas?

If they are new and shipped from overseas they are almost certainly "fake" cards.

Tom

---

A proud member of the OFA (Old Farts Association).

What kind of results you guys getting from the 1050ti. I saw these cards on eBay for $28.99. Too good to be real?

HOLY $h!t That seems too good a price to be real. Look at Stephen's host with the dual 1050 Ti's. Not bad at all with the special app.
https://setiathome.berkeley.edu/results.php?hostid=8222433&offset=0&show_names=0&state=2&appid=

I'm running two 1050Ti's, It is a little over 50K a day.

What kind of results you guys getting from the 1050ti. I saw these cards on eBay for $28.99. Too good to be real?

If it doesn't have a brand name, it is almost certainly a "copy" at best or a "fake" at worst.

I have just brought a pair and looked at e-bay first, all of the cheap ones were just "Nvidia" with no brand name , there was no visible name on the pictures either.

I personally would avoid these.

Very few examples so far with unroll = 2. But what I have observed in the limited data set is there is no difference between unroll = 6 and unroll = 2.

. . I may be wrong, but the results seem to be just a little more consistent across the task types, but the owner is in a better position to judge that than I am. Either way they are somewhat slower than I would have expected, by at least 10 mins.

Stephen

:(

What kind of results you guys getting from the 1050ti. I saw these cards on eBay for $28.99. Too good to be real?

HOLY $h!t That seems too good a price to be real. Look at Stephen's host with the dual 1050 Ti's. Not bad at all with the special app.
https://setiathome.berkeley.edu/results.php?hostid=8222433&offset=0&show_names=0&state=2&appid=

. . I am not trying to scare anyone but that machine only has one GPU. So yeah, not bad at all :) Bernie has a machine with dual 1050ti's.

. . But at that price I would definitely be checking the teeth before buying that horse.

Stephen

? ?

My curiosity would lead me to attempt to build the zi3v code branch as it was intended for x86_64 for CUDA10. If you look at my Tegra X1 gpu on my Nano. I do much better with the zi3v code on CUDA10 than the the GT730 times on CUDA60. I don't think the difference in platforms makes that much difference. I only have 128 CUDA cores on my Tegra X1 compared to 384 on the 2GB GT 730 cards like Loren's.
https://setiathome.berkeley.edu/results.php?hostid=8707387&offset=0&show_names=0&state=4&appid=

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

My curiosity would lead me to attempt to build the zi3v code branch as it was intended for x86_64 for CUDA10. If you look at my Tegra X1 gpu on my Nano. I do much better with the zi3v code on CUDA10 than the the GT730 times on CUDA60. I don't think the difference in platforms makes that much difference. I only have 128 CUDA cores on my Tegra X1 compared to 384 on the 2GB GT 730 cards like Loren's.
https://setiathome.berkeley.edu/results.php?hostid=8707387&offset=0&show_names=0&state=4&appid=

How would I go about building zi3v code to run on CUDA 10? Right now I am using the 390 driver.

You just need to grab the zi3v branch from the Seti svn repository along with the sah_v7_opt and Xbranch branches.
https://setisvn.ssl.berkeley.edu/trac/browser/branches/sah_v7_opt/Xbranch
Then you need to isolate the zi3v code branch and move it into the Xbranch. I have detailed instructions from CyborgSam. You obviously wouldn't be building for ARM64 so your configure would be totally different and you won't have to jump through so many hoops we did for making _autosetup and configure work correctly as they are already set up for x86_64. There are some changes to modules to work better with the low cuda core count cards that make it run better without errors. The zi3v code branch is here:
https://setisvn.ssl.berkeley.edu/trac/browser/branches/sah_v7_opt/Xbranch/client/alpha/PetriR_zi3v

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

What makes you think zi3v will give the correct results with CUDA 10.0 on a Kepler? It was tried with CUDA 9.2, 9.1, 9.0. 8.0, 7.5, and 6.5 on Kepler Titans and it gave the Wrong results, just like your Tegra X1. The Keplers didn't give the Correct Results until zi3v was coded with CUDA 6.0. It does give the correct results, doesn't it? Much better than your Tegra X1 with CUDA 10 which behaves just as the Apple iGPU App which you despise so much. Have you tried the CUDA 6.0 App on the Tegra yet? It might just work. It works on Intels and AMDS, don't really see why it wouldn't work on an ARM64 , and it might even give the correct results.

But the Tegra X1 is a Maxwell with Compute Capability of 5.3. Didn't know that I despise the Apple iGPU. Don't even know what that is. No haven't tried to compile the zi3v code with CUDA 6.0 yet. Guess we should try that.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

...Didn't know that I despise the Apple iGPU. Don't even know what that is....

Right, I'll refresh your memory. This is an Apple iGPU, https://setiathome.berkeley.edu/workunit.php?wuid=3549432846
They usually give around a 90% Inconclusive rate, sorta like your Tegra, which gives you a clue how bad an App can be and still remain on the SETI server.
The Same problem exists with the Apple CUDA 7.5 App, it works fine on Maxwells, works really bad on some Keplers and Fermis. It's still on the SETI server too.
I meant just try the existing CUDA 6.0 App on the Tegra, if that doesn't work then try compiling the Tegra CUDA App with CUDA 8. Maxwell should work fine with CUDA 8, and zi3v was originally built for CUDA 7.5, and 8.

I have a Inconclusive rate of 67% which I admit is really bad and not something I like to see. But it is not 90%. I'm am still waiting on Petri to get back so he can answer some questions about configuration. He and yourself have already provided us with some changes in module code that helped reduce the errors.

What I constantly see is an under reporting of Pulses. The Spikes, Triplets and Gaussians are normally correct. Where the app goes wrong is too low a Pulse count.

If the zi3v code was originally targeted at CUDA8, I think we should try that first. Should only need the download and installation of the CUDA8 package and then change the configuration to point at the CUDA8 directories. I need to speak with Sam about that. We have given up on the 0.97 code branch.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

CyborgSam reminded me we are out of luck apparently with anything other than CUDA10. Nvidia does not offer any earlier CUDA libraries for ARM64 from any of their website download links. I put in a Support request to Nvidia to verify and confirm we can't get CUDA8 for the Tegra X1.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

My curiosity would lead me to attempt to build the zi3v code branch as it was intended for x86_64 for CUDA10. If you look at my Tegra X1 gpu on my Nano. I do much better with the zi3v code on CUDA10 than the the GT730 times on CUDA60. I don't think the difference in platforms makes that much difference. I only have 128 CUDA cores on my Tegra X1 compared to 384 on the 2GB GT 730 cards like Loren's.
https://setiathome.berkeley.edu/results.php?hostid=8707387&offset=0&show_names=0&state=4&appid=

. . OK, I may have misunderstood what you were saying there but surely the difference is that it supports, and you are using, CUDA10, whereas the GT730 does not. It is only CC=3.5 and is limited to CUDA6

Stephen

? ?

If the zi3v code was originally targeted at CUDA8, I think we should try that first. Should only need the download and installation of the CUDA8 package and then change the configuration to point at the CUDA8 directories. I need to speak with Sam about that. We have given up on the 0.97 code branch.

. . 0.97 works fine for me :) It is a pity it doesn't for CUDA60 ...

Stephen

? ?

My curiosity would lead me to attempt to build the zi3v code branch as it was intended for x86_64 for CUDA10. If you look at my Tegra X1 gpu on my Nano. I do much better with the zi3v code on CUDA10 than the the GT730 times on CUDA60. I don't think the difference in platforms makes that much difference. I only have 128 CUDA cores on my Tegra X1 compared to 384 on the 2GB GT 730 cards like Loren's.
https://setiathome.berkeley.edu/results.php?hostid=8707387&offset=0&show_names=0&state=4&appid=

. . OK, I may have misunderstood what you were saying there but surely the difference is that it supports, and you are using, CUDA10, whereas the GT730 does not. It is only CC=3.5 and is limited to CUDA6

Stephen

? ?

No you misunderstand. CUDA runs on all versions of cards because is it backwards compatible. It just depends on what the silicon supports for any of the various CUDA functions. The CUDA compiler figures all that out when you configure for the arch level.

The GeForce GT 730 comes in 2 different flavors, one of which is compute capability 3.5, the other is compute capability 2.1

If you have the cc 2.1 version, cuDNN will not work with that GPU (it requires 3.0 or higher).

Other than that, choose the latest driver for your gpu using the wizard at http://www.nvidia.com/drivers


If you have the cc 3.5 GPU, CUDA 8, 9, or 9.1 will work with that GPU. If you have the 2.1 version, only CUDA 8 or prior will work.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

CyborgSam reminded me we are out of luck apparently with anything other than CUDA10. Nvidia does not offer any earlier CUDA libraries for ARM64 from any of their website download links. I put in a Support request to Nvidia to verify and confirm we can't get CUDA8 for the Tegra X1.

It appears you are SOL. The manual(s) says aarch64 is only supported on android prior to 10.1, 10.1 is the first time Linux is listed as being supported on aarch64, https://docs.nvidia.com/cuda/archive/10.1/cuda-installation-guide-linux/index.html#cross-platform Even 10.0 says Linux is Not supported on aarch64, https://docs.nvidia.com/cuda/archive/10.0/cuda-installation-guide-linux/index.html#cross-installation Short version... don't buy a Tegra X1 and expect it to run the Special App...correctly.
Of course, you could try it with v0.98b1, but, I wouldn't hold my breath.

Actually finally found the asterisked note on the CUDA 6.5 Toolkit page.

https://developer.nvidia.com/cuda-toolkit-65

*

** ONLY to be used for Jetson TK1.

So it works on the Jetson TK1 SOC gpu, NOT the Jetson TX1 SOC gpu. Yes, we are SOL.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

If the zi3v code was originally targeted at CUDA8, I think we should try that first. Should only need the download and installation of the CUDA8 package and then change the configuration to point at the CUDA8 directories. I need to speak with Sam about that. We have given up on the 0.97 code branch.

. . 0.97 works fine for me :) It is a pity it doesn't for CUDA60 ...

Stephen

? ?

Stephen, I am confusing you obviously with carrying on two conversations in the thread. There is no reason why the zi3v app couldn't be compiled for x86_64 with the CUDA 8 toolkit for the GT 730 in Linux x86_64.

https://developer.nvidia.com/cuda-80-ga2-download-archive

I am just screwed that I don't have any choice for a lesser CUDA toolkit other than CUDA10 for my Jetson Nano Tegra X1 ARM64 platform.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)