Bastiaan_NL wrote:On my cards I see the following with P13428 on Windows:
40 seconds tpf on a 3090 with 6.4 million ppd
52 seconds tpf on a 2080ti with 4.5 million ppd
64 seconds tpf on a 2080 super with 3.3 million ppd
So if your 3060ti is at 3.1 million ppd with 190W, that sounds pretty good to me considering that it's brand new (drivers etc).
I now have 30 relevant finished WUs on Linux with the 3060 Ti since Saturday afternoon, all from p13428 (filtering out the first WU and the ones which were paused for circa 15 min by the daily backup) :
Average TPF is 1'06", average PPD 3,120k, with a standard dev of 5K PPD (0,16%)
Average GPU Power draw is 189W (16,5k PPD/W)
No failed units
So, it's rock stable, and as you say, it should improve over time with optimized drivers and perhaps a bit of power limiting. For the moment, I think this GPU offers great value for money, compared with the previous gen at least !
Nvidia RTX 3060 Ti & GTX 1660 Super - AMD Ryzen 7 5800X - MSI MEG X570 Unify - 16 GB RAM - Ubuntu 20.04.2 LTS - Nvidia drivers 460.56
I dreamed for so long about being able to purchase my first PC. And it finally it happened. This kind of feeding of the egregor might be useful here
I also want to change my old 1070 to RTX3k series
HaloJones wrote:Now if only I could actually buy one!
Yesterday my local computer supplier finally secured a single MSI 3080 card but at a 35% premium over MSRP, so I declined. Hopefully pricing will stabilize in 2021 once there is an adequate supply.
I follow the PCMR reddit a bit, and there have been a few posts from people who have managed to get one of these for MSRP. So it is possible, and I wouldn't shop at any retailer that does charge a premium over that price anyways.
iMac 2.8 i7 12 GB smp8, Mac Pro 2.8 quad 12 GB smp6
MacBook Pro 2.9 i7 8 GB smp3
It sometimes doesn't depend on the retailer, but the add-in-board partner. For example, if a board partner like Asus, Gigabyte or MSI receive GPUs which can be overclocked, then they can sell those as a 3080 OC (factory overclocked) version, and earn more money on it, rather than selling it at MSRP without the factory overclock. In practice, a non-"OC" card might boost just as high as the OC card, but that's one way for the board partners to make more money without violating MSRP agreements with the GPU designer on the base model.
Some retailers mark the cards up even further than the AIB partner's MSRP, and frankly as long as supply is so low, that is perhaps preferable to the the scalper lottery at launch.
Given that the price per performance was one of the main selling points of this line of cards, the above-MSRP cards sort of negate the card's value.
Online: GTX 1660 Super + occasional CPU folding in the cold.
Offline: Radeon HD 7770, GTX 1050 Ti 4G OC, RX580
Small atom count wu's can massively underperform on rtx30 series and even more so on the top end of those - would is ongoing in testing and re assignments to try and make sure those that really don't ... if you want a feel for how it could work try setting Alz for the cause - I believe this currently has some projects fairly large atom count wu's in play - whether they are big enough to really work a 3090Ti I am not sure.
There's still no way to set more than 1 project running on a GPU in the client?
The 3080 and up have roughly 4 times more shaders, and 2 to 2,5x the TMU, ROPs, SM count, RT cores, and Cache per SM.
The only thing that's roughly the same (less than 15% extra) on the 2060 and 3080, is the tensor cores.
MeeLee wrote: ↑Sat Apr 09, 2022 2:52 pm
There's still no way to set more than 1 project running on a GPU in the client?
That will require new features in the folding client software, possibly new features in the folding core or a new core, and possibly even changes to the server code. Right now they are concentrating on creating a new version 8 client, whether that will include the necessary features remains to be seen.
iMac 2.8 i7 12 GB smp8, Mac Pro 2.8 quad 12 GB smp6
MacBook Pro 2.9 i7 8 GB smp3