GTX 680 or Quadro 4000 for Premier and AE CS 6

[LWRCJohn]

the 680 has well over 1500 cuda cores, while the Quadro has 256 per.  Im not very hip to the new video in computers but do understand that the purposes of these cards are very different.

Would it be better to buy a Quadro or a gtx 680? going in to a  SR-X mobo w/48 gb ram and  4( by 4) HDD arrays on independant sas raid cards.

Im building a new system cuz my SR-2 isnt cutting it, keeps crashing after adding the second raid card to it.

thanks in advance.

[MarkWeiss]

I haven't mentioned the system specs. Even today, it provides respectable editing performance. I've got a couple of grand in SSD drives in it, ten HDDs in all, some 7200rpm spinners, some SSD. It's clocked @ 3.51GHz and handles 4 camera multicam very smoothly, even if one of the cams has Unsharp Mask applied to it.

From what my research indicates, PCI-E 1.0 will only see a 10% performance hit over 2.0 and 3.0 gains nothing with the current technology and is supported by few vendors at the moment.

It's $8K vs. $250. Even if I gain the ability to use more filters and effects without slowing down, it's well worth it. But I don't expect enough business in the coming year to warrant a whole new system.

[Bill_Engeler]

What the others are trying to communicate is that a GTX680 is basically overkill for your system, and you are better off with something cheaper. I think you will see some improvement in export times with any CUDA enabled video card. You could probably find a used GTX 470 for less than a hundred dollars, and there will be no difference for you between that and a 680.

If your current system works for you, great, but you won't get any more joy from a 680 than something cheaper.

[Bill_Gehrke]

If you were thinking $250 why not get a brand new $200 GTX 960 which will use less power than the GTX680, be HDMI 2.0 compatible and have something that a longer life time.

[RjL190365]

Mark,

Bill Engeler reiterated what I and a few others have been trying to state: The GTX 680 is way overkill for your particular system. In fact, even the fastest LGA 775 CPU in combination with the fastest DDR2 system RAM will result in even the fastest such system perform slower than most of the cheap dual-core i3 CPU-based systems of today, let alone a quad-core i7-based PC. In fact, when I stated that "you'd be lucky to attain even 67% of the performance" for GPU-intensive tasks, it's not just the PCI-e 1.1 that would be restricting the performance, it's also the DDR2 system RAM that would. You see, Premiere Pro is very sensitive to the system RAM throughput to begin with (especially in MPE GPU-accelerated mode), and even the fastest DDR2 RAM has significantly lower total memory throughput than a mediocre DDR3 RAM-equipped PC. This is because DDR2 memory controllers on LGA 775 platforms are not on the CPU itself, but offboard on the motherboard (where there are far greater latencies than in an on-CPU memory controller). As a result, the practical (sustainable) memory throughput of an LGA 775 system is typically less than 60% of the memory's maximum theoretical (advertised) throughput due to all those latencies (whereas current DDR3 memory controllers deliver throughput that's much closer to the RAM's advertised specs).

As a matter of fact, even a $600 pre-built PC (if equipped with a cheap $80 discrete GPU) would outperform your current system (in terms of overall system performance, not just selected benchmark tests) even if your current PC is equipped with the very fastest CPU available for your motherboard and an $8,000 GPU. (This is exactly what I was trying to explain.)

And as I reiterate here, $250 is way too much money to waste on such an old PC for a "new" GPU. (And yes, you refer to the original cost of your current PC plus the total cost of component upgrades over the years when you were trying to determine the suitability of a given high-end GPU for your PC when in fact I am referring to the current, depreciated market price of another system with the exact components as your current PC.) And you might have paid $8K plus the total cost of upgrades over the past several years for that PC, but after depreciation it is worth no more (in total) than the cost of that used GTX 680 by itself. And putting a GTX 680 in that old PC is like having one spoke on a bicycle wheel 10 to 20 times longer than all of the other spokes on that same wheel.

And I definitely agree with Bill Gehrke when he stated that "you cannot compensate for a weak CPU platform with a super-strong GPU". What you were trying to do would not work efficiently, if at all.

Finally, you might have wasted money big time on your storage (disk) system: The ICH9R that's on your system's motherboard supports only SATA II (3.0 Gb/second). As such, you'd never achieve more than about 270 MB/second in real-world throughput out of any of the motherboard's six SATA ports. That would severely limit the performance of any newer SATA III (6.0 Gb/second) SSDs that have a real-world throughput of over 500 MB/second. Simply put, your system's SSDs only perform at half the speed that they're capable of, especially on reads.

[MarkWeiss]

Good technical explanation about memory controllers and overall performance.

That said, I have had three different GPUs in my system this week:

GTX580 3GB

GTX680 2GB

GTX680 4GB

The 580 seemed to have some stutter problems playing back. Both of the 680s were smooth as butter with four camera multicam with a adjustment layer containing Fast Color Corrector with a bump in midtons and brights, Sharpen with a setting of 14 and Timecode. CPU use was around 40-55% playing this timeline with these effects on.

Exporting was even more interesting... my XDCam EX footage, when exported to MPEG2 DVD, rendered at more than 2X faster than realtime (2 minute clip rendered in under a minute). GPU utilization around 87%!

However, another identically configured timeline with Apple ProRes 422 footage took 4 minutes to render with the same Adjustment layer effects applied. GPU Utilization around ONLY 22%!

I don't know why that is, as it would seem to be easier to decode intraframe format than long GOP and why the ProRes footage transcode to DVD didn't use the GPU much. Ideas?

Bottom line: the GPU was the best $300 upgrade I ever did to this workstation. My Windows Experience Rating is 7.4 overall, with the GPU and disc being 7.9. CPU and RAM are 7.4. I never could play through a timeline with sharping without dropping frames under CS3. And audio dropouts were something that would come and go throughout the editing session. CS6 with the GTX680 is smooth as glass with 4 camera multicam editor, effects, transitions, and multiple tracks of audio. As long as I'm rendering XDCam masters to DVD, rendering time is less than half the program length. Ie., 48 minute video renders in 22 minutes and the GPU is full tilt boogie.

I'm pleased with this update.

Yeah, I realize I could spend four grand and a few months building a new system, or 30 grand for a real top end system from a vendor (64 cores, Tesla GPUs, 512GB RAM), but I'm not WETA and for the infrequent concert video productions I still do, this makes sense. Plus it transformed my 3D modeling and animation application environment into something out of the future, with realtime production quality renders in the viewport. Best $300 I ever spent.