GTX is dead long live RTX !!

Message boards : Number crunching : GTX is dead long live RTX !!
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 . . . 12 · Next

AuthorMessage
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1954358 - Posted: 8 Sep 2018, 4:09:20 UTC - in response to Message 1954356.  

I found links both at wcctech and videocardz for the Hydro Copper EVGA cards plus the blower card.
https://videocardz.com/77837/evga-unveils-hydro-copper-and-hybrid-geforce-rtx-models
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1954358 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1956310 - Posted: 19 Sep 2018, 14:55:31 UTC

Reviews are in....

Bottom line....

Don't buy it unless you have no choice. Price increase for increased performance isn't worth it , at least for video games....

We need to see how scientific computing does on those before spending any $$$ on them.....
ID: 1956310 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22158
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1956322 - Posted: 19 Sep 2018, 15:40:23 UTC

As Zalster says - hold fire until the scientific reviews are in, or you want to be an ice-breaker and provide us with a review.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1956322 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1956331 - Posted: 19 Sep 2018, 16:06:03 UTC - in response to Message 1956322.  
Last modified: 19 Sep 2018, 16:24:12 UTC

AnandTech has good scientific benchmarks in their reviews. Hasn't posted yet though.
[Edit] And don't forget Phoronix.com for Linux performance reviews on RTX. Says it should be available later today, IF Nvidia releases any Linux drivers for RTX.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1956331 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 1956356 - Posted: 19 Sep 2018, 19:11:13 UTC

i do think that computational scientific loads will fare better than the gaming reviews so far. I'm not up to snuff on the exact calculation types performed on SETI and other projects, but can the Tensor cores be leveraged at all? i know petri has a titan V, and that card has tensor cores, but im not sure if he's utilizing them, or if he even can. im fairly certain the RT cores will be useless for scientific projects.

BUT. still not worth it. the price far outweighs the speed increase and they use even more power than the 10-series cards, so you won't be helping your electric bill at all.

it's a blood bath in the youtube reviews. a resounding "don't buy it" (at this price)
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 1956356 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1956363 - Posted: 19 Sep 2018, 20:12:31 UTC
Last modified: 19 Sep 2018, 20:13:23 UTC

No, the RT cores aren't and haven't been able to be used for SETI, or any distributed computing project that I know of. The benefit that the Titan V card has so far is just the greater number of CUDA cores in that product compared to 1080Ti. Same will hold true for Turing 2080Ti I assume.

I still want to see the comparison of compute performance on the 1080Ti and 2080 cards. The change in architecture for asynchronous compute of Turing might have a greater effect on compute performance since the difference in CUDA core count between the 2080 and 1080 is minimal.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1956363 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 1956366 - Posted: 19 Sep 2018, 20:38:40 UTC - in response to Message 1956363.  

No, the RT cores aren't and haven't been able to be used for SETI, or any distributed computing project that I know of.


i was asking about the Tensor cores. not RT cores. i know RT cores cannot be used. a Titan V does not have RT cores, but does have Tensor cores. Tensor cores are the "AI/DL" cores, and projected to be used for things like DLSS on the new cards. and of course things like AI/DL.

1080ti - 3584 CUDA cores

Titan V - 5120 CUDA cores, 640 Tensor cores

2080ti - 4352 CUDA cores, 544 Tensor cores, 68 RT cores

but i think i still get your point that the tensor cores can't be used for the types of calculations used in SETI and other scientific projects.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 1956366 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1956367 - Posted: 19 Sep 2018, 20:43:00 UTC - in response to Message 1956366.  

OK, I misnomered. I meant the Tensor cores. Nobody in distributed computing is using the Tensor cores. Certainly not Seti.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1956367 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1956369 - Posted: 19 Sep 2018, 20:50:56 UTC

The only benefit so far in Seti computing using the special app is the the Titan V has 80 SM compute units versus the 28 SM compute units of the 1080Ti. The 2080 Ti has 68 SM compute units. The 2080 has 46 SM compute units. The 2070 will have 36 SM compute units. So the benefit to Seti compute is the scaling of SM units in the newer architectures. Don't know yet whether a Turing SM unit is comparable to a Pascal SM unit with regard to compute. Still looking for the compute performance benchmarks.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1956369 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 1956377 - Posted: 19 Sep 2018, 21:33:08 UTC

there will definitely be different scaling across the different architectures.

look at petri's titan V. almost 3x the number of SMs compared to a 1080ti, but only about 2x faster

will be interesting for someone to get their hands on a 2080 or 2080ti to see how well it does.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 1956377 · Report as offensive
juan BFP Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 9786
Credit: 572,710,851
RAC: 3,799
Panama
Message 1956383 - Posted: 19 Sep 2018, 21:48:03 UTC
Last modified: 19 Sep 2018, 21:48:15 UTC

ID: 1956383 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1956390 - Posted: 19 Sep 2018, 22:01:25 UTC - in response to Message 1956383.  

That's what I have been waiting for. So, for N-body simulation, likely comparable to the MilkyWay@Home N-body simulation tasks, 50% faster for the 2080 FE over the 1080 FE. For Folding@Home tasks, 30% faster for the 2080 FE over the 1080 FE. For the 2080 Ti over the 1080 Ti, mostly the same 30-50% faster.

On the Geekbench 4 Compute benchmark, the new Turing cards are 200% faster than the Pascal cards. The Geekbench 4 benchmark are normally run with OpenCL primitives so directly comparable to the SoG app. They also can be run with CUDA primitives but I haven't read the article fully yet to see if they ran both the OpenCL and CUDA primitives and combined the scores or if they were just with the default OpenCL.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1956390 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1956397 - Posted: 19 Sep 2018, 22:34:46 UTC

Looks like Phoronix came up with about the same performance numbers compared to Pascal. About 30% improvement in games at 4K for the 2080 Ti compared to the 1080 Ti. I like their frames per second per watt and frames per second per dollar charts. Shows the 2080 Ti slightly better on the frames per second per watt metric but losing soundly on the frames per second per dollar metric to the 1080Ti.

So the new 2080 card will likely help your power bill compared to the the 1080Ti but you will have the same power bill for the 2080Ti as if you were running the Vega64 card. The Turing FE cards run a few degrees cooler than the Pascal FE cards. The AIB board offerings should improve on the thermals of the FE cards as expected by quite a bit.

Phoronix didn't run their compute benchmark suite unfortunately.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1956397 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1956398 - Posted: 19 Sep 2018, 22:34:50 UTC - in response to Message 1956390.  
Last modified: 19 Sep 2018, 22:45:25 UTC

That's what I have been waiting for. So, for N-body simulation, likely comparable to the MilkyWay@Home N-body simulation tasks, 50% faster for the 2080 FE over the 1080 FE. For Folding@Home tasks, 30% faster for the 2080 FE over the 1080 FE. For the 2080 Ti over the 1080 Ti, mostly the same 30-50% faster.

On the Geekbench 4 Compute benchmark, the new Turing cards are 200% faster than the Pascal cards. The Geekbench 4 benchmark are normally run with OpenCL primitives so directly comparable to the SoG app. They also can be run with CUDA primitives but I haven't read the article fully yet to see if they ran both the OpenCL and CUDA primitives and combined the scores or if they were just with the default OpenCL.


So if' it's only a 30% increase for the 2080Ti then it falls in line with what JayzTwoCents says about price to performance. Not a recommendation at this time.

Edit..

Sorry, skipped over that OpenCl section...hmm. Well if you are running SoG might be a reason. That would put them near Petri's Special app times.
ID: 1956398 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1956400 - Posted: 19 Sep 2018, 22:40:44 UTC
Last modified: 19 Sep 2018, 22:42:23 UTC

I would say if you have the budget for the 2080 or the 2070 and are running something like the 1060 or 1070, there may be a case to upgrade. Absolutely no case for the 2080Ti at its price point.
[Edit] It's a complete sideways move from anyone already running a 1080 Ti to a 2080. No point.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1956400 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1956401 - Posted: 19 Sep 2018, 22:46:18 UTC - in response to Message 1956400.  

I would say if you have the budget for the 2080 or the 2070 and are running something like the 1060 or 1070, there may be a case to upgrade. Absolutely no case for the 2080Ti at its price point.
[Edit] It's a complete sideways move from anyone already running a 1080 Ti to a 2080. No point.



Agree...
ID: 1956401 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 65709
Credit: 55,293,173
RAC: 49
United States
Message 1956409 - Posted: 19 Sep 2018, 23:46:32 UTC

One day I hope to get a 1080 card, like a PNY 1080 blower or an MSI 1080 Aero or a 1080 FE or an Asus 1080 Turbo, or even a 1080Ti version of these. Like maybe in late 2019 or early 2020.

Right now I can wait, I'm doing repairs right now.
The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's
ID: 1956409 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1956529 - Posted: 20 Sep 2018, 20:19:45 UTC

Some cards seem to be available for collection in the UK today (well, it would be tomorrow now), but I'm not paying over £850 for one...

So, a question - and again, I need hard, verifiable, facts only.

What's the hardware architecture for Turing (CC 7.5) cards - and for Volta (CC 7.0) while we're at it?

I'm seeing evidence that both 7.0 and 7.5 have 64 cores per SM - in which case, BOINC will overstate the Peak_Flops two-fold for both TITAN and RTX.

[BOINC is only coded up to CC 5, Maxwell, which has 128 cores per SM. Fortunately, so does CC 6, Pascal, so the current default works there too]
ID: 1956529 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 65709
Credit: 55,293,173
RAC: 49
United States
Message 1956533 - Posted: 20 Sep 2018, 20:36:52 UTC - in response to Message 1956529.  

ID: 1956533 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 1956535 - Posted: 20 Sep 2018, 20:43:45 UTC - in response to Message 1956529.  
Last modified: 20 Sep 2018, 20:49:24 UTC

Some cards seem to be available for collection in the UK today (well, it would be tomorrow now), but I'm not paying over £850 for one...

So, a question - and again, I need hard, verifiable, facts only.

What's the hardware architecture for Turing (CC 7.5) cards - and for Volta (CC 7.0) while we're at it?

I'm seeing evidence that both 7.0 and 7.5 have 64 cores per SM - in which case, BOINC will overstate the Peak_Flops two-fold for both TITAN and RTX.

[BOINC is only coded up to CC 5, Maxwell, which has 128 cores per SM. Fortunately, so does CC 6, Pascal, so the current default works there too]


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


84 Volta SMs
Each SM has:
●64 FP32 cores
●64 INT32 cores
●32 FP64 cores
●8 Tensor Cores
●Four texture units

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

ID: 1956535 · Report as offensive
Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 . . . 12 · Next

Message boards : Number crunching : GTX is dead long live RTX !!


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.