LinkBack URL
About LinkBacksRead more.EVGA, MSI and ZOTAC are the first to announce their cards today.
Read more.EVGA, MSI and ZOTAC are the first to announce their cards today.
My brain hurts. Today is the 28th May, how could you have learned of a delay tomorrow? And tomorrow is when it was expected to be released, but there was a delay so they're releasing them today?Originally Posted by hexus
yes, I should have written 29th April, thanks!
A pointless card if ive ever seen one o and way way way overpriced for its power
pointless... not in my opinion, cuda based gpu rendering/code will love these, especially those with double precision floating point calculations as the titans support this.
overpriced... as a gaming card yes but as a cheap alternative to a tesla it's not. It's basically a cut down tesla in the same way a titan is, you're just not getting 'cream of the crop' (arguable there) chip and the same customer support from nvidia.
This is in essence 2x Tesla k40's (the titan z is just one) with half the ram and they sell at £4400 a piece....so you're going to save around £6000 assuming you can fit your work in the 6GB of memory.
I'd seriously consider having one myself at £2000 for a mitx 3d render node, but £2400 is a little too high for me. Also temps would likely be an issue in a mitx (even well ventilated ones) so maxwell would be a better fit
What did you expect from nvidia?
I'm hoping that this is a sign that manufacturing problems are finished and we can now see a full range of maxwell cards soon !
pointless... not in my opinion, cuda based gpu rendering/code will love these, especially those with double precision floating point calculations as the titans support this.
overpriced... as a gaming card yes but as a cheap alternative to a tesla it's not. It's basically a cut down tesla in the same way a titan is, you're just not getting 'cream of the crop' (arguable there) chip and the same customer support from nvidia.
This is in essence 2x Tesla k40's (the titan z is just one) with half the ram and they sell at £4400 a piece....so you're going to save around £6000 assuming you can fit your work in the 6GB of memory.
I'd seriously consider having one myself at £2000 for a mitx 3d render node, but £2400 is a little too high for me. Also temps would likely be an issue in a mitx (even well ventilated ones) so maxwell would be a better fit
you do ofc realise that when you buy a professional card your also buying a service package - something these horribly overpriced cards do not have? also quadro cards have features that are laser shut in `gamers` cards - many years ago you could softmod the original GeForce 2 for quadro , and suddenly it could render pro software better and faster.
no this TitanZ is NOT a cheap alternative - those who buy CUDA parts pay for quadro for a reason and will never touch this.
DirectX 12? Is that spec even finalised?
Most existing cards are DX12 compatible too.
And you are just the sort of person Nvidia are looking for.... seriously I work with quadro cards, 5xx series (note no 6/7xx series) and titan cards in 3D applications that the quadro are designed specifically for... honestly there is NO DIFFERENCE in viewport performance these days and do you want to know why... 90% of the main programs they were aimed at have shifted to direct x over opengl and give significantly better performance in dx9+ than in opengl.
Cuda performance is linear based on cuda cores and clock speed, the key differentiator between geforce 6/7xx series and titan/quadro/tesla is based around memory (quadro/tesla go up to 12gb) and double precision floating point calculations (a key element of my post you appear to have missed), the standard geforce cards has theirs crippled (unlike the 5xx series). The titan has it's double precision floating point fully enabled in the same way as the quadro/tesla meaning it's performance is the same (allowing for usual variations of cores/clock etc) and the only difference is memory capacity being less.
And if you'd like to read what I posted again you will see that I mentioned about the customer support and the fact I was specifically referring to the titan and TESLA parts (which is basically a quadro without the video outputs so it works for them too).
I wouldn't say what I did without knowing about the topic
Major institutions with bucket loads of money may pick the quadro/tesla over these due to the support package but for people who freelance/work from home or even smaller businesses these ARE cheap alternatives to tesla and quadro's....
evidence backing up double point floating performance information - http://arxiv.org/pdf/1311.0590.pdf - note it's a pdf, you want page 2 and the row called DP GFLOPS, note the tesla and titan being similar while the geforce 780 is a lot less
Last edited by LSG501; 28-05-2014 at 08:40 PM.
Even for those in the very limited market this is supposedly aimed for (who need DP Cuda support without professional drivers or ECC memory), this doesn't make any sense whatsoever as it's a triple slot design which requires one slot of extra clearance (so in effect 4 slots). So compared to two Titan Blacks (combined price 2/3 of this) in SLI you are paying 50% extra ($2000 vs $3000) and get a card runs 20% slower (706MHz vs 889MH).
Now some may argue that this Titan Z fits into a smaller case than a pair of Titan Blacks but the only platform these cards would ever be paired with is LGA2011 or full Xeon workstation and those make no sense in small cases.
No, the only reason this was released (or paper released) is because the Nvidia CEO promised it in their investor statement and unlike the Ferni woodscrew demo it really is not a good idea to mislead shareholders. If Computerbase.de's review of the Titan Black is any indication:
http://www.computerbase.de/2014-03/n...-black-test/3/
this will most likely throttle like mad even if run at Max mode (which BTW they found to be 61.0 dB(A) compared to the R9-290X at 61.5 dB(A), but for some reason Nvidia didn't get the bad publicity which AMD got for the R9-290X).
SLI doesn't make any difference for cuda, most professional apps I work with don't support sli either (yes it's annoying) or need ecc for that matter. All cuda cards throttle thermally when doing gpu rendering in my experience, they're working hard after all.
I know I said mitx as an option but to be fair I'm a rare case in that I like small render nodes, most normal users would use this in a full atx/e-atx case and if you're buying it from some places like say scan (they do titan based 3d systems already) then there's nothing to say you couldn't have 3/4 of these watercooled in a case, they're still normal size. Now it wouldn't be cheap but it would be roughly twice the performance than the a tesla (or titan blacks) option while still being a relatively cheaper option
I'd also argue you don't need to pair with a xeon or 2011 system either, some apps don't actually need the cpu while doing the 'rendering' with it just be done on the gpu(s) instead.
Agree 100% with LSG501.
Agree also that Titan Z is not cost effective and best solution to CUDA rendering will soon be pairs of 780GTX\6GB GPUs in multiple PCs running as render slaves. If 4GB RAM is enough then 770GTX cards are probably going to offer the best performance for the money.
Well people that buy these can expect a card that can also manage real time design rendering. It's much much more than just a card for gaming, hence the price.
There are currently 1 users browsing this thread. (0 members and 1 guests)
Posting Permissions
Reply With Quote
