One of the best video cards today is the AMD Radeon VII. It DP is better than most cards except the newest Titan.
For the cost I can get three VIIs and blow the RTX Titan away with just 2 of them.
It is good for Milky-way, Einstein, UniverseAtHome, colatzconjecture, seti with the right helper app.
For DP do not count out the R9 280x ot the HD 7970 and the Hd 7990, Titan 6gb, Titan black and the Titan Z
And I believe PG largest GFN still use DP too, it is no slouch at folding at home.
In Setti with the right helper app, my two 1080 Ti were out producing 2 2080Ti and two 1080Ti together not running the right app.
A lot of this what is best has a lot to do with how you set things up to run, what your goals are and how much $$ are you willing to spend
Prime numbers are important, Math is a science in its own right.
Like a secure computer, like data encryption? You like large prime numbers then.
 

I seriously doubt that any AMD is faster than even a regular RTX 2080; and according to the benchmarks it's more in line with a 2070 Super in terms of performance.
A 2080, 2080 Super, and 2080 Ti still should be faster than a VII or Vega 64.
 
But that aside, they do have a TDP of 295W, vs the 215W on the 2070 Super.
I guess it'll depend from project to project how much each project taxes the GPU;
However in cases like Folding@Home, and some Einstein@home GPU tasks, that fully utilize the GPU, an RTX 2070 can be dialed down to 140W, and with a little overclock, has nearly the same performance as stock; probably the same performance as a VII or Vega 64 (but at 100W lower).

Thanks Bill
I do still have them PowerColor LCS Radeon R9 290X DirectX 11.2 AXR9 290X 4GBD5-WMDH/OC 4GB 512-Bit GDDR5 PCI Express 3.0
 
Doing this now, I have never gone over 3 before. "Hey Bill, fire up some milkyway with a 2080ti and I will run some too on a few different GPUs Titans and AMD"
I do not have my Titan installed but I remember from our testing back from then. I will run MilkyWay tomorrow night on this new AMD Card.
 
I Need More CPU Cores For Non-DP <gpu_usage>0.3</gpu_usage> is the max I use.
https://milkyway.cs.rpi.edu/milkyway/results.php?hostid=785575
<app_config>
   <app>
     <name>milkyway</name>
     <gpu_versions>
         <gpu_usage>0.2</gpu_usage>
         <cpu_usage>0.1</cpu_usage>
     </gpu_versions>
   </app>
</app_config>
This gives me 5 Tasks on one card and 4 tasks on the other card.
5 1/2 to 6 1/2 Minutes Per Task.
But as we know most are coming back as  · Validation inconclusive (11) 
Going to test this next but I think I have to Enabled HT
<app_config>
   <app>
     <name>milkyway</name>
     <gpu_versions>
         <gpu_usage>0.1</gpu_usage>
         <cpu_usage>0.1</cpu_usage>
     </gpu_versions>
   </app>
</app_config>
10 1/2 to 13 1/2 Minutes Per Task  Validation inconclusive (38) It is not worth running so many Tasks Per GPU without DP.
This is what I run on my Titan Cards <gpu_usage>0.12</gpu_usage> 8 Tasks and 0.10 10 Tasks
10 on GPU 0 and 7 on GPU 1 Now I have enough task to get 10 on GPU 0 and 10 on GPU 1
But all are coming back · Validation inconclusive (27) because the RTX 2080 Ti Cards have no DP.
This is also running Milkyway@home Separation 
This is my normal running Host 705518 for this Project <gpu_usage>0.5</gpu_usage> on Air
https://milkyway.cs.rpi.edu/milkyway/results.php?hostid=705518
 

Getting any ARP in WCG yet?
 

For folding, 1M ppd is about on par with a 2060.
The 2080 does 1,5M PPD, the 2080 Ti does 2.2 to 2.5M PPD in Linux.
The problem with AMD is that not a lot of projects use dual precision. Not even folding at home uses it a lot. Most projects do single precision, and some projects on boinc do half precision.
So measuring a GPUs performance on DP, when its hardly ever used, is pretty useless.
 
Single precision Nvidia is far superior, which is for most of the projects.
Can't say if GPU grid uses DP or not... I haven't done a lot of work for them yet.

I will run some folding in a little while. 1m was off the tops of my head. It should so what a 2080 does or close.
This is about BOINC. BOINC is a totally different animal than folding. I will fire some up though.
I think AMD looks better gaming, the color is more vivid, or it pops better, but that is just my opinion.
 
OK when Bill gets his 2080ti numbers in.
 
VII  50 seconds to do 5 tasks                 =10 seconds per task
R9280x 130sec to do 4 tasks                  =32.5 sec per task
Titan 6bg 290 sec to do 8 tasks             =36.2 per task
HD7990 dual GPU 145 sec to do 6 tasks = 24 sec per task
S9000 155 sec to do 4 tasks,                 = 39 sec per task
1080Ti 175 sec to do 3 tasks                 = 58sec per task
 
There can be a spread between tasks, I took what appeared to be an average 150 to 200 would be 175
I did not add every one up and divide. Not worth that effort.

bcavnaugh
RTX 2080 Ti = Run time 4 min 59 sec - 5 min 21 sec 3 Tasks Per GPU Stock Clocks
GTX 1080 = Run time 2 min 42 sec - 3 min 35 sec 2 Tasks Per GPU Stock Clocks
Normal OC Settings
RTX 2080 Ti = Run time
 
 

2080ti is 103 sec per task
1080 is 95sec per task
 
 
I started folding on the VII
1.159,022 PPD
14196 wu
2min 5sec a frame  10% done
 
Not the best folder, but the WU is not far along. To me over 1m ppd is good enough.
Nvidia has been better at folding for a few years has not always been better.

I undervolted it some and downclosked it a tiny bit. It is running at 48c in a 77f room

As far as folding, I don't care so much for folding anymore
Started folding 13 years ago. Ever since they told me and a bunch of other folders to go pound sand.
They said "We do not care about you or your hardware, we never asked you to spend thousands on hardware to fold with. All we want is your spare cycles from you desktop PC"
Then they made the multisocket servers a whole bunch of people purchased obsolete and ended the BIGADV SMP CPU project.
Stanford, still to this very day treats their donors like crap when you go to their official support forums.
That was the last straw for many people on many teams. Stanford and their support people can kiss my grits.
 

bcavnaugh

bill1024

bcavnaugh
RTX 2080 Ti = Run time 4 min 59 sec - 5 min 21 sec 3 Tasks Per GPU Stock Clocks
GTX 1080 = Run time 2 min 42 sec - 3 min 35 sec 2 Tasks Per GPU Stock Clocks
Normal OC Settings
RTX 2080 Ti = Run time

2080ti is 103 sec per task
1080 is 95sec per task
 
I started folding on the VII
1.159,022 PPD
14196 wu
2min 5sec a frame  10% done
 
Not the best folder, but the WU is not far along. To me over 1m ppd is good enough.
Nvidia has been better at folding for a few years has not always been better.

Normal OC Settings Running 3 Tasks Per GPU.
RTX 2080 Ti = Run time 70 Sec per task.

You said 5min to do 3 tasks
5x60=300sec divided by 3 is 100. 
 
I added a little more cause you said 5min to 5min 21 sec   at 5:21 that is 107 er task
 

bill1024
So A VII is up to 10x faster than a 2080Ti in Milkyway  Maybe 7 if my math is wrong, but I think I am correct, maybe LOL
 
And no matter what project you like or dont like, there are people who only do Milky way , or Einstein, or primegrid.
That is the great thing about BOINC, there is something for everyone. What ever floats your boat.

And whatever Pop Your Bubble

bill1024
Sorry
But those 2080ti you have do so much better in PG.

Indeed They Do

Bill Can you run some Einstein tasks with your 2080Ti please, and will see how it does vs the VII

folding today on the VII
wu14276 core 21
1,777,606 ppd to 1,850,783  Goes up and down
180w in the AMD software 
 
1min 3sec a frame
 
Not too bad
53c in a 77f room on air still. Not put the water block on yet

bill1024
folding today on the VII
wu14276 core 21
1,777,606 ppd to 1,850,783  Goes up and down
180w in the AMD software 
 
1min 3sec a frame
 
Not too bad
53c in a 77f room on air still. Not put the water block on yet

Not bad,
Although I'm more saying it about the power consumption being lower than I thought.
Can you do something about lowering the Wattage even further?


A similar performing 2080 or 2070 Super, with a proper Overclock, is only running at 125-140W, 47-55C GPU core temps at ~75-85F ambient; open case with the stock cooling solution's fans at 100%; getting 1,3-1,7M PPD of folding performance in Linux.
Should my ambient temps drop by 10F, my GPU temps drop by about 5C, and performance increases by about ~5% due to the Nvidia driver engaging higher boost clocks.
I'm comparing it to folding, because folding at home actually taxes the GPU 100% all the time.
Boinc doesn't, and often runs my Nvidia GPUs at 25-50% load, with occasional 80-100% load; which might be a reason why they are doing so bad in Boinc.
To solve this, I shared Boinc with FAH, to assure 100% GPU utilization all the time.
Boinc and FAH seem to be fine sharing 50/50 the GPU resources, however it comes at a 40% performance drop for FAH.

Would love to see if you can drop the wattage even further down for AMD cards, and see how much the wattage readout is, to make a much closer PPD/Watt rating comparison!
There is an efficiency curve, at which for every X-Watts you reduce on the GPU, your performance drops by a Y amount.
At some point, this curve becomes too steep to be feasible (eg: where you drop 10%watts (eg: 20Watts out of 200W, or 15W out of 150W), and lose 10% performance).
Try to find the card's sweet spot;

I'm interested to see how an optimized AMD performs in Linux.
There seems to be a program called 'WattmanGTK' that can do some basic overclocking and power limiting...
 
Perhaps I'll buy a cheap AMD card soon, to test...