+ Reply to Thread
Page 1 of 3 1 2 3 LastLast
Results 1 to 10 of 25

Thread: Gtx 970 & 980

Hybrid View

  1. #1
    Member okokoo's Avatar
    Join Date
    Feb 2012
    Location
    Stockholm, Sweden
    Posts
    52

    Gtx 970 & 980

    Since we're kind of moving into a new architecture it would be nice to hear if someone have picked up one of these cards? From what I've read from tests using GPGPU render in Blender (980) the speed has increased with 18% compared to 780Ti. The 970 is on par with 780Ti. Don't know how the case will be in Bunkspeed but it would be great if this new generation (last one with 28nm) would actually get a better transition than the Fermi-Kepler one did, where there was decreasing efficiency for a long time.

    Anyone got some buzz?

  2. #2
    Senior Member Aselert's Avatar
    Join Date
    Nov 2011
    Location
    France
    Posts
    447
    Good question! I'm curious to see it in the Benchspeed :P

    At this time, and what I know is that the drivers are not fully optimized to have a clear vision of performances. By example, on Octane it looks like there is a decline of performances. Not like in Arion, or Cycles (Blender) as you said. Surprise on Bunkspeed! I plan, nevertheless, a small increase of perfs here

  3. #3
    Senior Member haknslash's Avatar
    Join Date
    Nov 2011
    Posts
    293
    I'm going to hold out for Pascal cards (if they can deliver what they are saying). We've pretty much hit the ceiling with current cards and manufacturing but I do like some of the features of the 980 for gaming , just not so much for rendering performance unless looking at power savings (which is impressive for sure). A decent read on the 980 http://www.anandtech.com/show/8526/n...gtx-980-review
    Kenny Wilson
    900D ASUS Maximus VI Extreme 4770k 3x GTX 780 Ti's AX1200i 32GB RAM H100i a lot of SP120 fans
    Bunkspeed Shot Pro 2014

  4. #4
    Administrator david.randle's Avatar
    Join Date
    Apr 2011
    Location
    San Diego, CA
    Posts
    1,252
    All - With the jump from Kepler to Pre-Maxwell (K5000 (Kepler) to K5200 (Pre-Maxwell) We are seeing a double in performance for the same pricepoint card. Very impressive. I imagine this will go up with full maxwell chips and further even beyond. Remember, with GPU's, we are at an inflection point in the performance curve.
    David Randle / General Manager / Bunkspeed

  5. #5
    Senior Member blitz's Avatar
    Join Date
    Nov 2011
    Location
    Waterloo, Canada
    Posts
    175
    Quote Originally Posted by david.randle View Post
    All - With the jump from Kepler to Pre-Maxwell (K5000 (Kepler) to K5200 (Pre-Maxwell) We are seeing a double in performance for the same pricepoint card. Very impressive. I imagine this will go up with full maxwell chips and further even beyond. Remember, with GPU's, we are at an inflection point in the performance curve.
    I'll believe it when I see it from other users. No offense David, but I got burned when I purchased 2x680's under the advice of your staff that I would see improved rendering times compared to my prior 2x570's. The 570's still out-performed the expensive 680's (at the time) by about 25%.

    I'm going to wait for non-biased results from actual users before upgrading my hardware yet again. CUDA cores on the 980's are less than the 780Ti's, and your double performance claim sounds ridiculous.

    Why don't you show some graphs or statistics or render times on the benchmark scene to back up your claim?
    Last edited by blitz; 09-29-2014 at 01:02 AM. Reason: Toned down the words a bit...

  6. #6
    Administrator david.randle's Avatar
    Join Date
    Apr 2011
    Location
    San Diego, CA
    Posts
    1,252

    K5200 vs. K5000

    Quote Originally Posted by blitz View Post
    I'll believe it when I see it from other users. No offense David, but I got burned when I purchased 2x680's under the advice of your staff that I would see improved rendering times compared to my prior 2x570's. The 570's still out-performed the expensive 680's (at the time) by about 25%.

    I'm going to wait for non-biased results from actual users before upgrading my hardware yet again. CUDA cores on the 980's are less than the 780Ti's, and your double performance claim sounds ridiculous.

    Why don't you show some graphs or statistics or render times on the benchmark scene to back up your claim?
    Hi Blitz - Attached is a comparison in render time on the publicly available benchmark file.

    K5200 v K5000.jpg
    K5200 v K5000 2.jpg

    column 5 is render time and column 6 is FPS

    You can see the render time is less than half when comparing the K5200 and K5000 and FPS is almost double. It looks like this performance trend will continue with subsequent chipsets.

    As for your GTX situation, i personally apologize if a recommendation was made with certain claims by a Bunkspeed employee. This would have been done so inadvertently. We never test nor benchmark GTX cards and never will so any advice we could possibly give is based on predictions, assumptions or publicly available data. We will not be held responsible for claims against performance on GTX series cards.
    David Randle / General Manager / Bunkspeed

  7. #7
    Junior Member tushar's Avatar
    Join Date
    Jul 2012
    Posts
    7
    I just installed a GTX 970 with bunkspeed Pro 2014.5.1017.20683 and it does not work. I can render with CPU alright but with GPU the render queue gets stuck at "Initializing Renderer", background rendering quits with "Connection with the background rendering was unexpectedly lost" and accurate mode gives the error "Renderer argument exception".
    I formatted the computer and re-installed windows before posting on this thread. The computer is a Mac pro tower with 2xquadcore xeons and 64bit windows 7 SP1.
    Is GTX 970 not supported by bunkspeed yet or am I doing something wrong? Please help.

  8. #8
    Administrator david.randle's Avatar
    Join Date
    Apr 2011
    Location
    San Diego, CA
    Posts
    1,252
    Quote Originally Posted by tushar View Post
    I just installed a GTX 970 with bunkspeed Pro 2014.5.1017.20683 and it does not work. I can render with CPU alright but with GPU the render queue gets stuck at "Initializing Renderer", background rendering quits with "Connection with the background rendering was unexpectedly lost" and accurate mode gives the error "Renderer argument exception".
    I formatted the computer and re-installed windows before posting on this thread. The computer is a Mac pro tower with 2xquadcore xeons and 64bit windows 7 SP1.
    Is GTX 970 not supported by bunkspeed yet or am I doing something wrong? Please help.
    Tushar - Sorry to hear about this but it doesn't surprise me. I'll see what i can find out from Nvidia but chances are, a new version of iray will be required to support the 900 series cards.
    David Randle / General Manager / Bunkspeed

  9. #9
    Senior Member JohnG's Avatar
    Join Date
    Nov 2011
    Location
    Ireland
    Posts
    161
    I can understand your scepticism Blitz, I never saw any increase in performance when I switched from a GTX470 to a GTX670 despite having more CUDA cores.

    However, from what I've read about the 980, nVidia are claiming a 2x performance per Watt increase over the 680 with less CUDA cores and it seems they've delivered, which would back up what David is saying.
    Still it would be nice to get some feedback from Bunkspeed users rather than gamers, as most of the reviews seem to be aimed at them.

  10. #10
    Senior Member Aselert's Avatar
    Join Date
    Nov 2011
    Location
    France
    Posts
    447
    I think indeed, there is a term confusion for the general public. Nvidia plays on that for a while...
    I said-I think now to be experimented enough GPU to say- simply that we should not confuse performance and performance/consumption (efficiency).
    I am well aware of it now, especially when Nvidia released the Kepler cards, which compared to Fermi cards are actually more efficient. That is a fact. But are less powerful in pure computing (for many applications in any case). That is a fact too.

    It's like if Porsche is comparing the latest Porsche 918 to an old 917 and saying "The 918 is faster than the 917" = this is wrong. In pure speed terms, the older is faster than the new one. Even if the old one consumes 10x more than the new. So the new is really more efficient than the old = this is true. And for someone like us, who seek pure speed (beyond the agreement of the card), the performance/consumption (efficiency) represents not much. Except for the heat, the electricity bill, and the planet respect!

    So following this, Kepler to Maxwell will not double the pure performance. Maybe double the efficiency, yes!
    In the same idea, I personally not look more the card names, but the chip model, its architecture, the number of cores and their frequency. The rest (naming) is just marketing. I got screwed with the new Kepler architecture, I thought that tripling the number of cores that would triple the performances! I would not do me a second time!

    But another aspect, and it is what David was saying too, is the price: Nvidia have done, it's true, a real effort about the price. And what David was saying is that the K5200 is more or less 2 times faster than the K5000, with roughly the same price. And this is very good deal, yes!
    But when you see, like me, the chips of the 2 cards: you can see that the first is the GK104 and the second the GK110, which are 2 very different chips (like comparing a GTX680 with a Titan). So... logical in a sense to have this performance gap.
    Last edited by Aselert; 10-01-2014 at 03:11 PM.

+ Reply to Thread

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts