OTOY Forums

Okay, I know availability and pricing are going to be factors in real world, but as a theoretic exercise, which would you rather have:
(1) 3x GTX Titan X - 12 GB Vram w/ 9,216 Cuda Cores; or
(2) 2x GTX Titan Z - 6 GB VRAm w/ 11,520 Cuda Cores

This is assuming you had $3,000 US and Titan X is $1,000 and Titan Z is $1,500.
Also assume your platform can handle it operably.

A) 6GB additional VRam (with X), or 2,304 more cores (with Z) ?
B) What if it was for projects on a deadline, vs your own artistic endeavors?
C) What if power and heating were an issue, vs not an issue?

Oh well...Thought it might be fun to see what people think

Add choice 3: X&Z, that's what I would go for with your assumations
And even then I would go for 2xX and 1 Z spending 500$ more when being able to stick one Z in bottom slot and two X above with slots space between th cards

Definitely 3X

without a single doubt: 3 TitanX spaced out with single slot in between =) why?

400points on OctaneBech easily with decent acoustics 12GB of vram - that's all to say =)
if You wish to add one more X..full loop advised to get the best out of that configuration.

glimpse wrote:without a single doubt: 3 TitanX spaced out with single slot in between =) why?

400points on OctaneBech easily with decent acoustics 12GB of vram - that's all to say =)
if You wish to add one more X..full loop advised to get the best out of that configuration.

If cost was not a concern, I would totally do the 4 Titan X in a watercooled loop. I don't believe in future-proofing, but that would probably be as close as one could reasonably get for the near future anyway. Based on the original post: 3 Titan X with plan of expansion to above later.

Do the Titans (regular, Black, Z and X) all have the waterblock with same place for bridge or mutli links?
If so, you can set them in one watercooled config.
Then I would go for mix of Z and X in one loop.. Maximum power from Z and maximum vram if needed by the X.
If not.. Too bad, they should

rappet wrote:Do the Titans (regular, Black, Z and X) all have the waterblock with same place for bridge or mutli links?
If so, you can set them in one watercooled config.
Then I would go for mix of Z and X in one loop.. Maximum power from Z and maximum vram if needed by the X.
If not.. Too bad, they should

Can't speak for Xs, but the stock bridges for my regular Titan Hydro, Titan Black Hydro and Titan Z Hydro water blocks are all located at different (some slight, some way off) positions. If one installed the blocks him/herself, he/she can likely find one block that fits the regular Titan and the Black and maybe even the X, and possibly save some money, but not installation time, in the process. I couldn't find one for the Z that lined up perfectly with any of the others, but the stock regular Titan Hydro's block came closest to lining up with my Titan Z Hydros. Different manufacture's hydro cards and after-market water-coolers could be different, however.

Tutor remark is worth gold! I go even further:
NEVER PLAN TO PUT/BUY DIFFERENT CARDS IF YOU THINK ABOUT WATERTCOOLING.

Waterblocks inlets and outlets are almost surely not aligned identically so joining them will cause only problems. Always look for same graphics cards, always get the same waterblocks.

BTW: Waterblocks for Z are very hard to get (EK versions). There as some Bitspower or Swiftech left.

smicha wrote:Tutor remark is worth gold! I go even further:
NEVER PLAN TO PUT/BUY DIFFERENT CARDS IF YOU THINK ABOUT WATERTCOOLING.

Waterblocks inlets and outlets are almost surely not aligned identically so joining them will cause only problems. Always look for same graphics cards, always get the same waterblocks.

BTW: Waterblocks for Z are very hard to get (EK versions). There as some Bitspower or Swiftech left.

Smicha,

Your remarks are worth gold. I'm re-iterating your caution, " NEVER PLAN TO PUT/BUY DIFFERENT CARDS IF YOU THINK ABOUT WATERCOOLING," with an additional one. NEVER PLAN TO PUT/BUY DIFFERENT CARDS IF YOU THINK ABOUT CREATING A MASSIVELY PARALLEL PROCESSING SYSTEM, UNLESS YOU ARE SURE THAT THE MIX OF CARDS WILL NOT CAUSE A RESOURCE/IO SPACE ISSUE. I've yet to receive all GTX 780 6Gs necessary for my Amfeltec conversion of my quad CPU Supermicro. New Egg is currently selling EVGA refurbish models for between $350 - $400 (USD) with a one year warranty from EVGA. So far in my conversion, one thing that I've found is this: employing different GTX cards can ruin having the maximum amount of cards recognized by your OS (Windows in this case) because employing different GTX cards may create a situation where the system's IO space is maxed out sooner than would otherwise be the case if all GPUs were exactly the same models. As I've pointed out elsewhere [ http://render.otoy.com/forum/viewtopic. ... 96#p230696 ], different GPU cards, such as the GTX 780 TI ACX SC OC, have different IO space needs. In this particular case, however, my having installed a single slot EVGA GT640 in my Supermicro has made me more aware of the scope of the IO space issue. Having that GT640 installed was preventing me from reaching my system's GPU potential limit on the number of GPU cards that it may ultimately support, with the GTX 780 6G cards that I already have. With that GT640 installed directly in the Supermicro, the SupermIcro system recognized as working GPUs only 2 of my GTX 780 6Gs connected by the Amfeltec splitters. The other cards weren't active because they had IO space/lack of resource issues. But as soon as I removed that GT640 and replaced it with a 780 6G, the system recognized, without issue, all five of the GTX 780 6Gs that I'd installed. One of the five GTX 780 6Gs was then replacing the GT640 in the Supermicro and the other four GTX 780 6Gs were connected by the splitter . When my other GTX 780 6Gs arrive later this week, I hope to get that system to utilize at least 7-8 GTX 780 6Gs.

Thanks for sharing this, Tutor. This piece is of great importance. Hope to hear good news about your 7x 780 6Gb soon.