Ask Petri to post his first 30 lines of his Event Log to show how BOINC reports his Titan V?

---

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

A proud member of the OFA (Old Farts Association)

Ask Petri to post his first 30 lines of his Event Log to show how BOINC reports his Titan V?

Wrong way round. I'm trying to re-write BOINC so that it tells Petri the truth.

yes, 64 cores/SM.

straight from nvidia: https://devblogs.nvidia.com/nvidia-turing-architecture-in-depth/

and confirmation of Volta here: http://images.nvidia.com/content/volta-architecture/pdf/volta-architecture-whitepaper.pdf

Thanks, that's exactly the sort of thing I was looking for, and I've found both references.

But I'm a bit worried by the second one. Table 1, page 10: Tesla P100, GP100 (Pascal) is shown with 64 cores per SM. Observations?

my only guess is that it's because theyre talking about the GP100 (Tesla) card, and not the mainstream geforce gaming cards that we are all used to. You'll notice that the Tesla card also sports HBM2 where the 1080ti has GDDR5X.

even though it's pascal, the GP100 is a different beast.

https://forums.anandtech.com/threads/gp100-and-gp104-are-different-architectures.2473319/

---

Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

Thanks. That does seem definitive. The picture is completed by https://docs.nvidia.com/cuda/pascal-tuning-guide/index.html:

The specific compute capabilities of GP100 and GP104 are 6.0 and 6.1, respectively. The GP102 architecture is similar to GP104.

(footnote 2). So we could detect and use that information, though I'm not sure the BOINC developers will want to bother.

Edit - let me bookmark https://en.wikipedia.org/wiki/CUDA#GPUs_supported as a handy reference list of compute capabilities.

OK, I've submitted BOINC code via #2707 so that the RTX cards should display the correct GFLOPS peak from BOINC version 7.14 onwards. Earlier versions will show double the correct speed for RTX, Titan V, and Tesla P100 - hardware wizards please verify.

This project arose from an email from Ray Hinchliffe, author of SIV. BOINC and SIV use slightly different techniques for calculating peak flops: BOINC can look up the number of SMs (shader multiplexes), but needs to be told how many cores are assigned to each SM (my changes above). Ray has a technique for detecting the total number of cores directly. We compare notes periodically to make sure we're getting the same answer.

If anyone out there has a P100, Titan V, or even RTX, it would be interesting to know whether the current BOINC versions (anything later than 2014) give a different peak flops value from SIV, and (in due course) whether BOINC v7.14 converges them again.

More benchmark tests available, this time concentrating on compute performance of the RTX 2080Ti. The Phoronix.com tests had some good CUDA and OpenCL tests that probably are similar enough to make some conclusions about how the 2080Ti would perform on similar Seti work. Also some interesting performance per dollar charts comparing RTX 2080Ti versus GTX 1080Ti.

NVIDIA GeForce RTX 2080 Ti Shows Very Strong Compute Performance Potential

---

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

A proud member of the OFA (Old Farts Association)

More benchmark tests available, this time concentrating on compute performance of the RTX 2080Ti.

Other than a test here or there, it's often twice as fast as a GTX 1080Ti (and even faster than that in 1 or 2 tests).

And looking at the cost/performance comparison, even with the RTX 2080Ti's obscene pricing it actually compares favourably with the GTX 1080Ti- it's behind for most tests, but dead level for one or 2. And this is running code that isn't optimised in any way for the new architecture.

While you can't really justify the RTX2080Ti price for gaming (other than bragging rights) till there are games that can really make use of it's Ray Tracing feautres, when it comes to compute workloads its improved performance will pay for itself in no time (for heavy use scenarios). And it would do even better with software written specifically for it.
Very impressive piece of hardware.

---

Grant
Darwin NT

I agree and why I wanted to hold off any "is it worth it" comments until I saw compute performance tests since its gaming performance is moot in our use case. I think it will come out pretty good regarding credit per watt once one of these show up in Shaggie's charts. It is only pulling about 20-40 watts more than the 1080Ti in these tests. I would like to see how it compared to the Titan V but it looks like Phoronix never did a compute test on that card.

---

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

A proud member of the OFA (Old Farts Association)

pull the trigger...pull the trigger.... ;)

---

pull the trigger...pull the trigger.... ;)

If I had $2,000 lying around I certainly would, unfortunately...

---

Grant
Darwin NT

I'm going to have to wait on pulling the trigger too. Still want to put a TR 2950X build in front of it. But if and when I do pull the trigger, it will be either a EVGA AIO RTX 2080Ti or a custom block version.

---

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

A proud member of the OFA (Old Farts Association)

First live RTX 2080 has been tested. Current versions of BOINC do report the wrong value for peak flops, but my revision has been accepted and will be included in BOINC v7.14. That code has been tested in the real world too, and gives the correct flops value.

May we ask who has one, so that we can look at the performance of it?

May we ask who has one, so that we can look at the performance of it?

The testing was done by Ray Hinchliffe, as mentioned in message 1956627. He provides BOINC-specific extensions to SIV, but I don't know if he runs SETI.

OK, thanks.

Did anyone knows if there is any RTX2080 actualy crunching SETI? So we could follow the performance in the real world.

---

We're still waiting on a Turing card to show up here. Supposed to be one next week on Einstein by Archae86. In that thread Mikey says someone already has one running on PrimeGrid. Says it is faster but not OMG faster. Same generational improvement seen on every new card family release. Someone else did an analysis of the efficiency of the Turing CUDA core compared to the Pascal core on PrimeGrid tasks and came up with 46 -50% faster.

---

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

A proud member of the OFA (Old Farts Association)

Let's wait.

---

First live RTX 2080 has been tested. Current versions of BOINC do report the wrong value for peak flops, but my revision has been accepted and will be included in BOINC v7.14. That code has been tested in the real world too, and gives the correct flops value.

That's good, but the price is about to go up and it has nothing to do with Nvidia or AMD, Intel, Asus, Gigabyte, MSI, PNY, etc, etc, etc.

Buying that used Asus R3E motherboard a while back that needs a new cpu socket is looking mighty good now. :)

---

The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's