Come em get the titan's

I am guessing newegg marketing anticipated demand correctly by having all three cards available for online sales.
 
Pocatello said:
Why are they special?

Why do I need one?
Click to expand...

The new Titans while expensive and out of reach price-wise for many are basically a consumer grade of the flagship Tesla K20X that went into the Titan Supercomputer at Oak Ridge national Laboratories. Now I'm certainly not suggesting that they will be the best crunching card for all workloads (i.e. all distributed computing projects) but this card has massive amounts of compute power. If nothing else it should be fun to see how it stacks up to other Nvidia cards (and AMD cards) in compute preformance (CUDA and OpenCL). :)

Say what you will about other review sites and testing methods but AnandTech has some good background on the new GPU core and compute performance in their two part review if anyone is interested:
http://www.anandtech.com/show/6760/nvidias-geforce-gtx-titan-part-1
http://www.anandtech.com/show/6774/nvidias-geforce-gtx-titan-part-2-titans-performance-unveiled
 
It is, and always has been about software.

Exactly how many consumer/entertainment applications have optimizations for a video chip that has only been put into non-video service? Zero.

There has always been killer hardware that never gets into the consumer market. Will photo editing, or movie editing, or computer animation applications for the consumer ever get written for Tesla? This will determine it's fate.

But right now, today, for the consumer market, it's nothing special. nVidia would have to push/help software companies to optimize for it, or it probably isn't going anywhere.
 
Qinsp said:
It is, and always has been about software.

Exactly how many consumer/entertainment applications have optimizations for a video chip that has only been put into non-video service? Zero.

There has always been killer hardware that never gets into the consumer market. Will photo editing, or movie editing, or computer animation applications for the consumer ever get written for Tesla? This will determine it's fate.

But right now, today, for the consumer market, it's nothing special. nVidia would have to push/help software companies to optimize for it, or it probably isn't going anywhere.
Click to expand...

If something is written in CUDA or OpenCl, tesla will run it. There are already a lot of consumer applications that can fully utilize tesla (audio, video, photo editing, engineering programs, animation software...). At launch OpenCl support is broken for Titan, but that has already been fixed in a driver update.

There isn't anything magical about writing for GK110 (other than there are a few more options just because it supports a more expansive version of CUDA).
 
It's those optimization switches in the compiler that make all the difference.

I remember when the math coprocessor on a PC was optional ($$$) chip. But if your compiler did not have a switch to code for it, it was useless.

In the case of the MC, flipping that switch in the compiler cut processing time 90% on certain tasks.

Not well versed in Tesla options, but from reading that article, it has stunning double+ integer (password cracking), float, and double float performance. Is this something the software is aware of automatically and takes full advantage of?

In any case, distributing computing isn't enough alone to keep a product on the market. More common software would have to see a huge bump in performance to make it a sales success.

IMO that is.
 
Everything that you say is true but I think this could be a popular card in research settings. There are plenty of universities and researchers that would like to get their hands on one (myself included :D).

Not enough to make it a mainstream seller but great for people who need the compute power.

Looks like the project administrator over at GPUGrid, GDF, is trying to get one and thinks it could yield decent gains.

We are now trying to get hand on one titan for optimizing the application on it.

As usual it's very difficult to find them over here, if anyone is willing to donate one please contact us.

In terms of performance, we are expecting a speed-up of 50% over a gtx680 for normal wu. For large jobs, this could be close to 100% faster. ACEMD reduces the speed when the molecular system is large but not on a titan due to the 6GB of memory.

Extra registers could also provide a further boost, but we don't know yet. Stay tuned.

gdf
Click to expand...

Should be interesting to see! :)
 
Don't get me wrong, this is the Z06 of the video cards. Cheap muscle, stunningly powerful.

However, the market is for SUV's right now.

So they just have to make a 8 passenger Vette ... :D
 
metallicafan said:
Everything that you say is true but I think this could be a popular card in research settings. There are plenty of universities and researchers that would like to get their hands on one (myself included :D).
Click to expand...

I honestly think this is who NVidia made the Titan for. I think this market is rapidly growing, and while CUDA has a huge advantage in established user-base, for people that are buying a single card (whether academics, engineers, financial analysts, etc.), the price/performance ratio had firmly shifted to AMD/OpenCl. People don't like writing things twice, which has been Nvidia's biggest advantage. Because CUDA was (essentially) the only game in town for so long, they not only have a near monopoly on compute hardware, but a huge advantage in coding know how. If they start losing the low end compute market, they not only lose those sales, but suddenly you have an expanding user base that not only uses OpenCl, but more importantly does not use CUDA, and has no interest in learning CUDA. Those people eventually give input into large purchasing decisions, like what cards do we want in our next supercomputer.

I have no doubt that if Nvidia wanted, they could release an OpenCl driver that would put their hardware on equal footing with AMD. But that would be accepting competition on even ground. The more people they can keep tied to CUDA, the less they have to complete (at least directly).

They will no doubt sell a lot of cards to gamers. For every 100 people who rage on the internet about how stupid their pricing is, there are a few people that are more than willing to pay it. But, I think the market for 'low' cost compute hardware will provide enough demand to soak up as many cards as they decide to make.
 
metallicafan said:
Everything that you say is true but I think this could be a popular card in research settings. There are plenty of universities and researchers that would like to get their hands on one (myself included :D).

Not enough to make it a mainstream seller but great for people who need the compute power.

Looks like the project administrator over at GPUGrid, GDF, is trying to get one and thinks it could yield decent gains.



Should be interesting to see! :)
Click to expand...

I'm surprised they're willing to spend time optimizing for what will most likely be a very rare card. I'm sure FAH will never be titan optimized.
 
jebo_4jc said:
I'm surprised they're willing to spend time optimizing for what will most likely be a very rare card. I'm sure FAH will never be titan optimized.
Click to expand...

Actually, Titan optimized = CUDA optimized. 6xx optimized is also CUDA optimized, with the added restriction of optimizing memory usage (packet size matters since memory bandwidth is much lower). So, if they optimize for the 6xx series, they will be slightly gimping Titan (relatively, since it could handle larger packet size). If they optimize for CUDA, 6xx will be slightly gimped (when forced to deal with packets that are bandwidth limited...i.e. compute units can do more work than memory bandwidth can provide). If they go to OpenCl they'll both be gimped (but the 680 more so), until NVidia decides to un-gimp their OpenCl driver.
 
jebo_4jc said:
I'm surprised they're willing to spend time optimizing for what will most likely be a very rare card. I'm sure FAH will never be titan optimized.
Click to expand...

It might be as simple as a compiler flag. Dunno.

I imagine since Tesla compute machines have been around awhile, that there is a compiler for it.
 
Note, that if this card becomes popular, they might recompile games and other consumer software to take advantage of it.

Benchmark testing is only as good as the optimizations. AMD and Intel have fought constantly over this. You can have the same source code, but use different compiler flags to give one arch an edge over another.
 
You must log in or register to reply here.
Back
Top