A list that gives you an overview about the latest Nvidia cards and theire €/Credit rating. I pick prices of the version with less RAM if more than one because it doesnt affect crunching. The credit entity will be GFLOPS from Wikipedia and prices come from www.idealo.de.

Model - - - - - - - GFLOPS -- € -- GFLOPS/€
Geforce 8800 GS -- 396 -- 166€ -- 2.39
Geforce 8800 GT -- 504 -- 120€ -- 4.20
Geforce 8800 GTS -- 346 -- 97€ -- 3.57
Geforce 8800 GTS(512) -- 624 -- 118€ -- 5.29 yay
Geforce 8800 GTX -- 518 -- 134€ -- 3.87

Geforce 9800 GT -- 508 -- 91€ -- 5.58 yay
Geforce 9800 GTX -- 648 -- 139€ -- 4.66
Geforce 9800 GTX+ -- 705 -- 119€ -- 5.92 yay
Geforce 9800 GX2 -- 1052 -- 424€ -- 2.48

Geforce GTS 150 -- 705 -- not found

Geforce GTS 250 -- 705 -- 113€ -- 6.24 YAY
Geforce GTX 260 -- 715,4 -- 142€ -- 5.03 yay
Geforce GTX 260(216SPs) -- 804,8 -- 160€ -- 5.03 yay
Geforce GTX 280 -- 933,1 -- 289€ -- 3.23
Geforce GTX 285 -- 1062,7 -- 288€ -- 3.69
Geforce GTX 295 -- 1788,4 -- 409€ -- 4.37

Nvidia Tesla

C870 GPU Computing Processor -- 519 -- 946€ -- 0.55
D870 Deskside Supercomputer -- 1037 -- 4116€ -- 0.25
S870 GPU Computing Server -- 2074 -- 4509€ -- 0.46
C1060 Computing Processor -- 1936 -- 1502€ -- 1.29
S1070 1U Computing Server -- 4320 -- 6834€ -- 0.63

Someone has to explain the Tesla-Series to me ...

Interesting... The GTX 295 looks better and better ...

As time goes on the GFLOPS/cost ratio will improve ...

Geforce GTX 260 has G200 chips which would be more future safe, as the chip has several improvements. We also optimize for G200, so you would be slightly faster at the same gflops (10-15%).

gdf

If you stuck in 3 x EVGA GTS 250 at 135€ each, you would theoretically get 2115 GFLOPS for the price of 1 x GTX 295.

With the 15% optimisation improvement of the G200 architecture, you could be getting 2056 GFLOPS.

It would be interesting to see which solution had the lower power consumption too.

Rob

If you stuck in 3 x EVGA GTS 250 at 135€ each, you would theoretically get 2115 GFLOPS for the price of 1 x GTX 295

Perhaps, but I would prefer to stick in 3 GTX295s so I could get 6 tasks running ... :)

The other question is power. The draw of the 295 card is roughly the same as my 280 card but does about twice the work. The power I pull is going to outweigh the costs of the cards eventually.

Yes, buying cheaper cards to fill all the slots you have is a great way to start. And then to slowly replace them with faster cards as money comes free is certainly what I intend to do. I mean, I am running a 9800GT card and that is going to be the first one replaced. But, the likely way this is going to happen is that I will build a new i7 system with slots for 3 GPU cards. I will put in a new card bought with the system (likely a GTX295 or similar), the GTX280 and the 9800GT ... the system that is running with the GTX280 right now will get the HD4870 I am using in one of the systems I will retire ... and if I upgrade the MB in that system will likely get something along the lines of an HD4870(2) assuming that MW has solved the work issue issue ... :)

Geforce GTX 260 has G200 chips which would be more future safe, as the chip has several improvements. We also optimize for G200, so you would be slightly faster at the same gflops (10-15%).

gdf

It looks to be more significant than that. I just had (and lost) a debate on our team forum on this subject. A gtx260/216 completes a 3098 claimed credit WU in 19.8k seconds. A gts260 completes a 3098 claimed credit WU in 43.1k seconds. The gtx260 has 15% more gflops but does twice the work (or does the same work in half the time). It is looking like that either the architecture or the memory interface is giving the G200 a 50% advantage over the G92.

Could the first post still be edited? There are 2 very important points to add, which people might miss if they don't read the entire thread and just go for the first post:

1. The speedup of G200:
- GDF just said 10 - 15%
- based on fractals numbers it's ~90%
- when the G200 optimizations were introduced I estimated >30% performance advantage for G200 at the same theoretical FLOPS
- I can confirm 39.3s for these WUs with a slightly overclocked 9800GTX+
- fractal, I guess you meant "GTS250" instead of "GTS260"? [oh, how I hate this name..)
- it may well be that the G200-advantage is not constant and scales with WU sizes, which may explain why we're seeing a much higher advantage now than in the past (with smaller WUs)
- my 9800GTX+ was good for ~6400 RAC, whereas GTX 260 made 10 - 11k RAC prior to the recent credit adjustments

-> concluding these observations I can say that the advantage of G200-based cards is >50% at the same theoretical FLOPS. This screws your table entirely.

2. The prices:
- the last time I checked (around christmas 2008) there were large differences in the prices of the cards relative to each other (compared Germany and USA). E.g. in Germany the G92 cards were much cheaper, but the GTX 260 was more expensive.

-> check your local prices, don't just look and the values and decide flat out, which card to buy

- also consider power consumption.. getting hard numbers for GPU-Grid is difficult, but several smaller cards will generally be less power efficient than a single fat card.

- G200 is more future proof than G92. Both are DX 10.0, but G200 has additional features which may or may not be needed for future CUDA clients.

Edit: never mind Tesla. Their credits/€ are totally irrelevant, as (hopefully) noone is going to buy these just for BOINC.

MrS

---

Scanning for our furry friends since Jan 2002

Edit: never mind Tesla. Their credits/€ are totally irrelevant, as (hopefully) noone is going to buy these just for BOINC.

Well, I had been GOING to ... but the tables seem to make it clear that it would be an un-good investment. Better to get more GTX295 cards ...

But there is no rush, not till Einstein comes out with their application and I need more power to keep GPU Grid about where it is and put some effort into EaH ...

BOINC is not just a hobby ... it is an addiction ... :)

Glad you didn't do it, there wouldn't have been any benefit, except for nVidia. the TESLAs use the same chips and just add more memory and more robust cooling, possibly along with higher clock speeds.

MrS

---

Scanning for our furry friends since Jan 2002

- fractal, I guess you meant "GTS250" instead of "GTS260"? [oh, how I hate this name..)

Yes, specifically, the one mentioned in this thread and designated as "stock".

I have posted a revised version of the list. this thread can be closed/deleted

I'll lock the thread because we have a better list out there.

MrS

---

Scanning for our furry friends since Jan 2002