blastingcap
Diamond Member
- Sep 16, 2010
- 6,654
- 5
- 76
Ya I read that section and I don't even remotely agree with that part of the review.
Can you please explain what compute advantage the Titan has exactly for its $1000 price? AT included the ElcomSoft's password cracking. AMD cards would destroy the Titan there in 2 seconds.
"If we look at Wireless Security Auditor, ElcomSoft's most popular tool the situation changes slightly, as a single K20 delivers 85,000 passwords per second, compared to the 65,000 on the GTX 690. Then again, Nvidia still lags behind AMD, as the three year old Radeon HD 5970 handles 103,000 passwords per second, and HD 6990 increased that to 129,000. In that aspect, not even the K20 can reach performance achieved by a single consumer AMD card. This is also the reason why a sea of secy agencies went forward and acquired AMD Radeon HD 5990 and HD 6990 cards, instead of going professional with the Tesla and FirePro cards." ~ Source
Bitcoin mining = fail on NV
OpenCL compute = fail on NV
Double Precision = For $500 Asus Matrix Platinum @ 1300mhz gives you 1.33 Tflops. That's half the price for a similar level of DP.
The only way you can justify the compute advantage here is if you use very specific compute programs that rely on CUDA. If you do and it really matters, you are probably a professional.
Besides RayTracing, the Titan's compute isn't looking so hot:
http://www.computerbase.de/artikel/grafikkarten/2013/test-nvidia-geforce-gtx-titan/12/
Ask Ryan Smith. Seriously.
Cherrypicking is easy. BTC mining is due to an instruction reason and MilkyWay@Home etc. for similar reasons.
Yet you didn't mention one of the hugest projects out there, Folding@Home. For that and other such projects, Kepler has the advantage architecturally and no amount of OpenCL can make up for it.
Then there is the CUDA ecosystem NV is trying to build and the support it has today. NV's professional support team is far ahead of AMD, and frankly if I were a business or agency I would think twice about going down the AMD path today--who do I call if my render farm or supercomputer runs into a bug or whatever? NV will be around in one, two, even five years. I guarantee that. AMD is bleeding money and has an uncertain future. To be fair though those guys are buying Teslas, not Titans, anyway.
Even as a nonprofessional, a Tesla Jr., aka Titan may be worthwhile if you already know how to work with CUDA, or you have something that runs better on Kepler architecture or something else. We'll see. I think NV is using this as a sort of experiment, as SirPauly and I were discussing a page or two ago in this thread. If they see this as a viable market, I wouldn't be surprised to see a Titan II, III, IV, etc.
OpenCL driver issues can be fixed, much like how some launches have screwed up SLI or CF drivers that get fixed later on. Does seem like a rush job to launch Titan before the bugs were worked out in OpenCL though.
But I know if I start talking about protein folding and CUDA you will just respond with the same stuff as above and we'll go on for pages so let's just cut to the chase and get Ryan to pop in here.
Ask Ryan Smith.
Last edited:
