I am wondering are there already users here with a Radeon VII?
In theory it has a very high computing performance, even surpassing the 2080ti in some compute benchmarks.
Currently i have a 1080ti but would be willing to switch over to the Radeon VII if the performance is better than on the 1080ti with SoG support.
Is there anyone out there with some information?

I haven't seen any posts yet here at Seti with users with the Radeon VII. Seen posts on other projects like MilkyWay and Einstein with the VII. They seem to be happy with the performance.

---

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

A proud member of the OFA (Old Farts Association)

I am curious how well the Radeon VII performs. I would think that with the memory capacity and bandwidth someone could optimize a killer app. Its just a guess; I am not fluent in GPU architecture.

Still, I would like to see stats with how it does in BOINC.

---

Seti@home classic: 1,456 results, 1.613 years CPU time

I knew I had seen a Radeon VII already crunching somewhere. It was at MilkyWay which needs cards capable of double-precision floating point. It is about 3 times faster than a Titan V.

Radeon VII - https://milkyway.cs.rpi.edu/milkyway/results.php?hostid=797216&offset=0&show_names=0&state=4&appid=
Titan V - http://milkyway.cs.rpi.edu/milkyway/results.php?hostid=776231&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)

That's pretty cool. How is it that they have crunched only one task?

---

Seti@home classic: 1,456 results, 1.613 years CPU time

If I remember right, he is waiting on the EK water block for the card. He stated that the card is beastly loud and hot. I assume he crunched a few tasks to sort out the drivers and then pulled the card or the host offline for the installation of the water block.

[Edit] He has actually done two tasks, another is in Inconclusive waiting for a wingman.

---

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

A proud member of the OFA (Old Farts Association)

Seeing the review on Gamers Nexus and other sites I can understand how the cooling is an issue and one would want to have an alternate means of cooling.

I am curious how well the Radeon VII performs. I would think that with the memory capacity and bandwidth someone could optimize a killer app. Its just a guess; I am not fluent in GPU architecture.

Still, I would like to see stats with how it does in BOINC.

In BOINC it can't do anything. BONC is the manager for projects, it's the projects applications that do the processing.

As Keith posted, if the application can make use of Double Precision Floating Point, it's a beast. If it's only Single Precision (which is what Seti is) then it's on par with the GTX 1080Ti (leads in some, trails in others), although it uses a power station and a half more power to do so (it's better than the Vega 64, but it's still not good).

---

Grant
Darwin NT

...although it uses a power station and a half more power to do so

250W vs 300W, not much between them. 600W vs 750W PSU requirement, also not much between them. Especially when lots of people have 750W as the bare minimum these days, but more often than not over 1,200W because they feel they need to run multiple cards and everything RGB on the world.

And while an RTX 2080 Ti FE says it's rated for 260W, it will draw a lot more! Just to give an idea that YMMV is correct. And really, comparing the VII to a 1080ti is like comparing an RTX 2080ti to a Radeon RX 480 (cos the RX 580 was only a rebrand).

---

Okay, true, it is the application itself, not BOINC. So if Seti uses floating point, is it possible to write a new app to process the data in double precision? I don't know enough to know if that is feasible or not. If it is feasible, it probably not practical right now. I thought I heard only about 5,000 Radeon VII cards were going to be made, so the amount of those using S@H would be a fraction of that.

---

Seti@home classic: 1,456 results, 1.613 years CPU time

No - because the results would not necessarily be compatible with the millions of results that have already been run.
Also there is no time advantage in running double over single where the enhanced precision is not required.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Got it, thank you!

---

Seti@home classic: 1,456 results, 1.613 years CPU time

Ok i decided to be the guinea pig for this after all.
The first few work units seem to run twice as fast as on my old GTX 1080ti.
I need to run more workunits to be sure.

Ok i decided to be the guinea pig for this after all.
The first few work units seem to run twice as fast as on my old GTX 1080ti.
I need to run more workunits to be sure.

You should consult Mike to see what he can come up with for settings to squeeze the most out of it! Congratz for getting a VII card.

---

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group

Your times are already very impressive.
If you want to speed up a little bit look for a file called mb_cmdline_win_x86_SSE2_OpenCL_ATi_HD5.txt.
Its located in your projects folder.

Add the following line

-sbs 2048 -period_iterations_num 10 -high_perf -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64 -cpu_lock

Save as text.
On next WU start it will use the optimized values.

Mike

---

With each crime and every kindness we birth our future.

Thx a lot Mike :)
Do i have to use the stock application for that to take effect, as i am using lunatics custom binaries?
Or is this something that works universal?

Thx a lot Mike :)
Do i have to use the stock application for that to take effect, as i am using lunatics custom binaries?
Or is this something that works universal?

This always works.
Just the name of the cmdline.txt is different.

---

With each crime and every kindness we birth our future.

Thx a lot Mike :)
Do i have to use the stock application for that to take effect, as i am using lunatics custom binaries?
Or is this something that works universal?

Universal, however the command line file for the Lunatics installation may have a slightly different name than for the stock one (that is the case for the NVidia command line file).

---

Grant
Darwin NT

This seemed to have changed something. :)
Before the added command line the GPU seemed to make a lot of pauses during the vlar tasks.
Now its almost running continuously through these tasks which did reduce calculation times.
Still lots of testing ahead like OC and undervolting to maybe get even better results.

250W vs 300W, not much between them.

For me, 20% is significant.
Even more so when it is on a much more power efficient manufacturing node.

And really, comparing the VII to a 1080ti is like comparing an RTX 2080ti to a Radeon RX 480 (cos the RX 580 was only a rebrand).

The comparison is the one most people are using, because well, that's roughly where it sits.
In actual fact it should be compared to the Titan V, because they are actually more similar products- a cut down professional workstation card.
Yeah, you can use them for gaming, but they really are workstation class products.

We'll need to see a Navi card to have an accurate comparison to the RTX series cards (and even then things will get messy due to Raytracing & non-Raytracing abilities).

---

Grant
Darwin NT