Overclock.net banner

1 - 11 of 11 Posts

·
CYRUS + KEF + LG OLED....
Joined
·
1,423 Posts
Discussion Starter #1
So anybody know if these so-called RTX GPUs are going to leave all our old GPUs in the museum ?

I just grabbed a cheap GTX 1070 .... LoL

Only to find it's just a wee bit faster than my GTX 970s.

So should we all start saving up for an RTX 2080 (or two !) ? ?

dunx
 

·
Not New to OCN
Joined
·
1,745 Posts
We won't know until benchmarks are out.
 

·
Premium Member
Joined
·
5,973 Posts
So anybody know if these so-called RTX GPUs are going to leave all our old GPUs in the museum ?

I just grabbed a cheap GTX 1070 .... LoL

Only to find it's just a wee bit faster than my GTX 970s.

So should we all start saving up for an RTX 2080 (or two !) ? ?

dunx
In FP32 a 1070 should be close to a 980Ti, maybe just shy. Pascal has weaker FP64 though so pascal is not a great performer in MW if that's what you put it on.

Someone was using a Volta card on MW and it was crunching very fast but the tasks were so quick he couldn't load up the GPU and ran out of vram. Hopefully some FP64 performance comes back after pascal.

It would be nice to get upgrade the 970 and 980Ti I still have running. More performance for same power.
 

·
Not new to Overclock.net
Joined
·
5,109 Posts
I had a 2080Ti on preorder but canceled it last night. Going to wait for the EVGA FTW3 like I did for my 1080Ti... Although I am also interested to see results before I do pull the trigger again.
 

·
BOINC Cruncher
Joined
·
1,864 Posts
I'm hoping for a large boost with the new cards since they have Tensor cores.
 

·
Premium Member
Joined
·
4,788 Posts
Still waiting to see benchmarks, but the raw fp32 numbers that have been posted for the 2080 and the 2080ti are nice, but not a giant leap in performance. The jump in single precision compute is less than what we saw from Maxwell->Pascal.
 

·
Premium Member
Joined
·
5,973 Posts
Oh, I missed that. Pascal also got a huge clock increase that came with the node shrink.

Heck of a transistor count increase from the 1080, 7.2B to 18.6B. It'd be interesting if a crunching app could be setup to use some of the Tensor core capability.

And of course I hope it sucks for mining.
 

·
Premium Member
Joined
·
4,788 Posts
Oh, I missed that. Pascal also got a huge clock increase that came with the node shrink.

Heck of a transistor count increase from the 1080, 7.2B to 18.6B. It'd be interesting if a crunching app could be setup to use some of the Tensor core capability.

And of course I hope it sucks for mining.

Something like DHEP would be perfect for fp16 cores. Now they just need to make a GPU app. CSG is running a CNN (convolutional neural network) right now (CPU only), and the project admins have been very happy with the results.

Half-precision/mixed-precision cores could definitely be leveraged, but the traditional types of projects in the BOINC world need the precision of fp32 or fp64. With the advent of consumer grade GPUs, from both AMD and NVIDIA, that have great fp16 performance, hopefully we'll see some new and interesting BOINC projects in the future.
 

·
CYRUS + KEF + LG OLED....
Joined
·
1,423 Posts
Discussion Starter #9
That was sort of my point.... that until everyone "catches up" with tensor cores functionality, then we won't see any decent increase in performance. But once coded we might see all our "old" tech made redundant.

I have retired several cards now due to high power consumption for little reward.

dunx
 

·
⤷ αC
Joined
·
11,240 Posts

Attachments

·
Premium Member
Joined
·
5,973 Posts
One user is having an issue running their 2080 on some [email protected] tasks. There was a data set that were failing. Some tasks ran faster for awhile with a certain [email protected] data set and they were called 'high pay' in thread. Since then [email protected] switched data set files and it runs those ok. They've been able to reproduce it running a task 'offline' outside of BOINC.

Just the one card atm. Not sure if its all 2080s, that one, that model.

https://einsteinathome.org/content/pascal-again-available-turing-may-be-coming-soon
 
1 - 11 of 11 Posts
Top