David Pirinelli
Forum Replies Created
-
The Anandtech review found the GPU Compute performance to be lacking, below a 570 even.
Haven’t seen a CUDA test yet but so far it looks like there is no “Magic Bullet” here.
-
I am sorry, I am an Art DIrector who writes video card EFI ROMs on the side. I do not know how to use Resolve beyond getting it to recognize my cards. I just how to make cards work for it.
THis is why I went to Jake Blackstone’s place in the past to get some testing done.
-
A quick question for anyone using a 580.
Can you check if it is running at PCIE 1.0 speed or PCIE 2.0 speed?
Easy way to tell is to look in CUDA-Z and see what the “Host to Device” and “Device to Device” “Pinned” speeds are.
If it is running at PCIE 1.0 (2.5 GT/s) , this number will be in the range of 2500-2900. If it is running at PCIE 2.0 (5.0 GT/s), this number will be in the 5000-5900 range.
When I was working this out for GTX470 and GTX480 I found a place in ROM to turn this on, but in 570 and 580 this trick doesn’t work the same and I have noted that some people get one range and some people get the other. If you report this, please also list what slot, what Mac, what other cards, etc.
Also, if you are running a GTX570 or 580, do NOT run the 10.7.4 Beta Update, it will turn your GTX5xx back into a pumpkin. (Can be re-enabled with ATY_Init )
-
David Pirinelli
March 20, 2012 at 1:57 am in reply to: chassis 1: gpu config error no cuda boards availableYou are using a Hackintosh, so that brings a whole slew of issues that can’t be easily gotten through.
If it works on a fresh boot, use it that way.
The AGPM thing only works if your Hack is id’ing as a 4,1 or a 5,1
-
As reported elsewhere, if you run the 10.7.4 seed that has been released, it will break the Nvidia support. The newer Nvidia drivers no longer init Non-EFI cards.
You can still use ATY_INit installed in extensions folder to make most of these cards work however.
Someone also needs to try putting pieces of the “magic” 10.7.3 drivers into a 10.7.4 install to see if it can be pieced together to keep newer cards working.
Be aware that running current 10.7.4 Beta will BREAK GTX570/580 support that came in new 10.7.3 drivers.
As much as we all want to think GTX680 will be working in Mac Pros from Day 1, keep in mind that from release of Fermi cards in March of 2010, we did not get functional OSX drivers until late fall of 2010 when Mac Quadro 4000 was released. It is possible that Nvidia and Apple will burn the midnite oil and release drivers from get go, but I wouldn’t bet on it.
-
If there is anyone in Hollywood doing 4K+ work who is running into GPU memory issues, I would be willing to loan out a test Quadro card. I have 5000 with 2.5 GB, a Tesla C2050 with 3 GB and a Quadro 6000 with 6GB. (I think only 4GB is useable in CUDA). All of these cards have same basic performance as GTX470/570/480 but with larger supply of RAM.
When we tested at Jake Blackstone’s office, the GTX480 was still king of all of these, but I think we were doing 1080P stuff, I’m not actually sure.
I would be willing to find out oif these larger frame buffers have an edge for the extreme resolutions that we know are coming.
(And of course, Dwaine M is welcome to try these out as well if he wishes)
-
AGPM fix not needed on 3,1
-
David Pirinelli
February 26, 2012 at 8:47 pm in reply to: Radeon 4870 + Quadro 4000 configuration in ‘Nehalem’ Mac Pro – is this possible?You run out of power plugs. 5770 and Q4000 use one each, you only have 2 so that works.
5870 needs 2 all by itself.
-
The quick and easy way to tell a Mac Q4000 from a PC one is the number of ports. A Mac card has only 2 due to a limitation of Nvidia drivers. There is one DVI port and a single DisplayPort. The PC version has 3 ports from which you can still only output 2 separate images. The PC card has 1 @ DVI and 2 @ DisplayPort.
RUn CUDA-Z to see if card is functioning in CUDA.
-
Quadro 4000 is hobbled by Memory Bandwidth
The 4800 is a slightly downclocked GTX285 and is almost as good as one in terms of memory bandwidth