[Dave Haynie]

Yeah, you can’t directly compare card to card, particularly between the nVidia and AMD, simply because they’re different enough in architecture that the answer is usually “it depends”.

There’s some comparability in naming. nVidia’s GTX5xx, GTX6xx, GTX7xx etc. indicates the main chip generation. The xx digits pretty much indicate the relative performance within that family. And if you get to “90”, you have a two-chip card, so it’s likely close to the double the performance of the next one down… if you can use that. That tends to be a problem for GPGPU… each processor looks like a separate device.

For AMD, the older cards worked similarly, the HD5xxx, HD6xxx, and HD7xxx represent successive generations of product. So it’s more or less the same chip architecture, but even then, not always. For example, the first members of the HD6xxx family were based on the same GPU architecture as the HD5xxx, but it changed in the HD69xx models. But basically, the 1000’s digit represents the generation, the 100’s digit the specific chip used, then it’s relative performance: things like clock and memory speed, how many stream processors are enabled. For example, the HD6790 has 800 stream processors at 840MHz. The HD6850 has 960 stream processors at 775MHz. The HD6870 has 1120 stream processors at 900MHz… not quite as fast as the older 5870, though. The HD6970 has the all-new (at the time) Cayman processor, and 1536 stream processors at 850MHz.

You have to look at what you’re doing. I’ve seen it stated pretty often that the HD6970 is comparable in gaming performance to the GTX570. That makes sense, given that when I bought mine, it was about the same price as the GTX570, and these are priced based on gaming. However, in OpenCL benchmarks, it’s usually pretty close to the GTX690. BUT… and isn’t there always a BUT… is that because the GTX690 is dual-chip (two Kepler GPUs), like the AMD HD6990 or HD7990? Many if not most GPGPU programs — Vegas included — can only use one GPU per program instance. Or is it because the GTX6xx series is using nVidia’s Kepler architecture, and Kepler is reported “not so good” at OpenCL? Here’s some source of OpenCL benchmarks:

https://www.tomshardware.com/reviews/radeon-r9-290-review-benchmark,3659-14.html
https://www.tomshardware.com/reviews/radeon-hd-7970-benchmark-tahiti-gcn,3104-14.html

Anyway, to answer your question… there’s a pretty good parallel. Given that the single-chip GTX570 and HD6970 are comparable, at least for games, it stands to reason that the dual-chip GTX590 and HD6990 would be similarly comparable. In fact, that’s a little easier than sorting out the differences in AMD’s line… particularly because the HD6xxx line was as much a project to get costs and power consumption under control as it was to advance performance. So as mentioned, most of those cards were lower power, die-shrunk versions of essentially the HD5xxx series.. and sometimes further downclocked.

-Dave

[Dave Haynie]

Well, MAYBE you can’t upgrade. It depends on the main board. Rick could upgrade to an LGA2011 six core Sandy Bridge E chip like your (and mine) i7-3930K, one of the other 39xx chips. He’s going from a four core Sandy Bride E to a six core in that case.

So there’s the Sandy Bridge EP version of the LGA2011, the Ivy Bridge EP version, the Ivy Bridge XP version, and the Haswell XP version. The first two are physically compatible, the next two are not, either with the first two or with each other. See here:
https://www.anandtech.com/show/7773/intels-three-versions-of-socket-2011-not-compatible

Now, it’s technically possible for a motherboard to support chips from more than one of these. But particularly for an older main board, don’t count on it. It’s particularly annoying given that all LGA2011 systems use the X79 chipset (so far), so there’s no special reason they couldn’t have made this universal. Or are there different Intel chips all called “X79”.

In short, check with your motherboard maker and see what chips they support in the latest BIOS. That’s your shopping list, if you’re looking to upgrade just the CPU. I checked my Gigabyte GA-X79-UP4 motherboard… if I have Rev 1.0, it supports Sandy Brige E, EP, and Ivy Bridge E chips.. but not Ivy Bridge EP. If I have Rev 1.1, it supports the Ivy Bridge EP chips as well. Presumably, Rev 1.1 added those extra pins. I have no idea right now which I’ve got. The Xeon E5 v2 chips are the Ivy Bridge EP.

The good news is that going from 4 core to 6 core is a pretty nice boost. Going from Sandy Bridge E to Ivy Bridge E, that’s going to be 4-6% on most things, given the same configuration (clocks/cores).

-Dave

[Dave Haynie]

[Rich Kutnick] “Dave, my PC has an i7-3820 processor @3.6GHZ. I assume that this is an older chip, for the PC now is 2.5 years old (UGH!). I don’t even know if I could replace this chip on my motherboard with a newer Ivy Bridge processor. “

That is a Sandy Bridge E series processor, which people sometimes confuse for Ivy Bridge. Anyway, you could almost certainly replace that with an i7-39xx processor, which gets you a 6-core Sandy Bridge E system. That’s going to be a 50% improvement on multithreaded stuff, give or take any clock speed differences. It’s not clear if you could upgrade to an Ivy Bridge E chip, i7-49xx, because that depends on which version of the LGA2011 socket you have. Yup, there are at least three versions. So it’s a main board thing, you have to check with the PCB vendor and see what they support. Later chips in the LGA2011 form factor have more pins!

-Dave

[Dave Haynie]

[Dave Osbun] “The big improvement with i7 processors was when Intel switched from the Sandy Bridge architecture to the Ivy Bridge chips (which are the newer i7’s). If you’re not making that jump the improvement will be negligible.”

It’s actually even more complicated than that, because the i7 LGA2011 chips, the higher end parts, are sort of released in-between Intel’s other releases. So sometimes you hear the i7-39xx called “Sandy Bridge”, sometime “Ivy Bridge”. According to the horse’s mouth, Intel, these are “Sandy Bridge E” parts. The next generation, the i7-49xx, which are actually Ivy Bridge E parts, but sometimes get dubbed Ivy Bridge, sometimes Haswell. In each case, they’re basically a step after the base architecture. But sometimes it’s a pretty good step; the i7-49xx parts are usually only about 5% faster than the i7-39xx series. The i7-26xx are just regular Sandy Bridge, that’s part of the problem.

But anyway, check out this benchmark series:
https://www.pcper.com/reviews/Processors/Haswell-Review-Intel-Core-i7-4770K-Performance-and-Architecture/Clock-Clock-Sandy

They didn’t get the names right, but the progression from i7-26xx to i7-39xx and i7-49xx is steady, but not amazing. And probably not worth spending much on; you’re talking 4-10% improvement, depending on the jump. And you may not be able to update, but I’ll cover that in another note.

-Dave

[Dave Haynie]

[Chas Smith] “And since Premiere does feature a somewhat seemless way to use Raw Video… just makes the pull even easier on my decision to look that way. “

Did they put CinemaDNG back into Premiere CC? Because they did drop support in CS6. And Adobe had really only been supporting it because they invented it. There are only three cameras that use this format, and two of the three were not shipping when Vegas 12 was released. And the Canon hacks came only in 2013.

[Chas Smith] “One of the things that got me using Sony Vegas Pro (starting with SVP6) was the ease of codec import… you’d think with such a huge customer base using T3i, etc….they’d have something in place to exploit that. “

Of course Vegas supports the huge Canon camera base: T3i, T4i, T5i, 60D, 5D, 6D, 7D. But it’s a different story to support the 0.0001% of users playing around with a still-a-work-in-progress unsupported hack 🙂

That said, I did put in a request to Sony for CinemaDNG support. They do already support uncompressed and at least Red R3D raw (not sure about Cineform raw), so it’s not as if Sony’s blind to these things.

Far as “the Betamax disease”, every company making software pretty much expects you to come around to their way of thinking. But don’t forget, when Vegas 2 first shipped, most NLEs required proprietary video; almost no one worked on native video, much less a mix of different formats. Sony still leads on this.

Adobe supported CinemaDNG because it was their project, not out of any great habit of supporting every video format. They dropped it because, basically, no one was using it… maybe ironically just at the time folks started to use it. It’s definitely true that under Sony, I think they’re much less likely to support initiatives like CinemaDNG, things for the whole industry, versus worrying about supporting Sony in particular. But that’s actually a 2way thing… other companies are much less likely to accept an industry standard from Sony, versus an independent like Adobe or some open membership industry committee like MPEG.

-Dave

[Dave Haynie]

[John Rofrano] “I’ve been doing some benchmarking with an old ATI Radeon HD 5870 that I recently acquired while I’m waiting for my NVIDIA Quadro 4000 to come back from PNY. I’ll post the results when I’m done but the difference is night and day with GPU on and off. I created a little render test with a rotating smtpe bars background and composited animated noise texture and without the GPU it plays back at 0.78fps at Best/Full (that’s less than a frame per second) with the CPU’s at 50%. I turn on the ATI Radeon HD 5870 GPU and it plays back at 29.97 full frame rate all day long. The GPU is at 69% while the CPU’s are at 9%. That is impressive GPU performance.”

That’s pretty much the kind of thing I’ve seen with the HD6970 as well. And again, in my experience with the recent AMD drivers, they are rock solid. Maybe that became a mission after AMD bought ATi, not sure, but it’s really a non-issue these days.

As I mentioned here before, there are issue with Vegas and both AMD and nVidia with newer cards. Main Concept doesn’t support any new cards, not for OpenCL, not for CUDA. Vegas itself does, and I’d expect video get even smoother on newer AMD systems.

There’s a different issue with nVidia. The older Fermi architecture was no problem. But the newer Kepler architecture does not seem to have proper OpenCL support. This isn’t a problem in Vegas, it’s not a matter of Vegas not using new features or anything… OpenCL has basically not changed between Kepler and Fermi (well, there was a new version, OpenCL 2.0 put out last fall or so, it doesn’t impact Vegas). You can find all sorts of discussions of this outside of the scope of Vegas or video.

So the short term wisdom is to get an older AMD HD5xxx or HD6xxx (or the Firepro cards based on the same architecture) or a nVidia Fermi, that’s GeForce GTX5xx, or the “no-K” Quadros. Longer term, it’s going to take new code from nVidia to improve OpenCL performance for the Kepler cards, and a new release of the Main Concept CODEC (which probably means a new version of Vegas) to make AVC rendering fast on the newer cards. Or FASTER… certainly, GPU support in Vegas makes things render faster even when the CODEC doesn’t use the GPU.

As far as speeding up a system, it used to be “buy a faster CPU”, and the GPU was basically just gravy. But since Vegas 12, I don’t really think do. Last summer I wet from a 6-core AMD 1090T to a 6-core Intel i7-3930K. That’s about a 2x performance increase, give or take, depending on the work being done. On that very system, the HD6970 gives me a 4x-6x improvement on edit and render speed, on a complex think the red car demo. You can’t get a CPU that’s anywhere near that much faster… to get a 2x improvement, I’d need the newer 12-core E3 Xeon, which runs around $2500. Just not really an option.

The think is, and it’s correct, that the GPU speeds up only specialized code, the faster CPU speeds up everything. And while that’s true, the alternate view is this: how much of an effect does each have on the things that really NEED speeding up. Clearly, I’ve chosen to do both. But I had the HD6970 there from my last machine. It was still more bang per buck for editing and rendering than any CPU I could have afforded back then.

Even today, that’s a $150 video card (if you can find one) versus a $500 CPU. Or maybe $200… the price of these seems to have gone up in the last year or two, despite being older models (mine is from 2011, it was about $300 new). They apparently can mine bitcoins (or could, before that got switched to FPGAs) 6-8x faster than my i7. And you could put four of these in one box.

-Dave

[Dave Haynie]

Yeah, they did a big push in the 5Ds, that was where they figured out the raw video coding, but it’s been spreading. Even to the 50D, which is interesting in that the 50D didn’t have a conventional video mode of any sort. Here’s a chart:
https://docs.google.com/spreadsheet/ccc?key=0AgQ2MOkAZTFHdFFIcFp1d0R5TzVPTVJXOEVyUndteGc#gid=5

My 60D has similar processor and memory system to the T3i; they’re both limited to a 21MB/s SD card. My 6D has UHS support, though not terribly fast, it’s got a good buffer and 40MB/s. But it also seems to the be last model to the party.

I used Magic Lantern awhile back, but it did work well on large exFAT cards, so I kind of left it behind. Probably worth another look.

The only video standard for raw video is CinemaDNG, which as someone mentioned, Vegas doesn’t support. That’s in a way the only possible raw standard, since after all, raw itself isn’t a standard, it’s just what your camera sensor dumps out, different for every camera. Adobe’s DNG is able to encapsulate most sensor geometries, so it’s really the only standard for this non-standard format. And so making that into a flipbook is a good way to make that work for video.

So there are tools to convert a Canon raw capture to a single CinemaDNG file (it’s actually an MXF file; CinemaDNG also exists as just a directory full of indvidual DNG files, with some standard for metadata applicable to the whole sequence). But you can’t get that into Vegas. So ok, there actually are a few other raw “standard” formats, though even less standard than Adobe’s. Cineform supports a raw version, and yes, I checked, it’s a real raw format, based on a 12-bit Bayer-filtered image. Cineform (well, now actually part of GoPro) has a discussion of there here, over a tool they provideo to convert from CinmeaDNG to Cineform. I don’t know what version(s) of Cineform support raw on Vegas, if any… Cineform got so hostile to Vegas with Vegas 11, I stopped using it.

https://cineform.zendesk.com/entries/22175187-Blackmagic-Cinema-Camera-CinemaDNG-to-CineForm-RAW-Workflow-Windows-

Not sure if this applies to MXF CinemaDNG, or just the discrete frame in a directory version.

It should be possible to convert raw directly to an uncompressed AVI, using already available DNG libraries, but no one seems to have done that yet. One of the issues is that you can’t just convert DNG to most other formats, you need to have a means of color grading it first, since you’re more than likely going from 12 or 14-bit raw to 8 or 10-bit video. So it’s not as trivial as a basic conversion program.

I’m kind of interested in playing with this myself (does it show), but I haven’t found an ideal means of interface with Vegas just yet.

-Dave

[Dave Haynie]

Woops… that would be fall of 2011; when Vegas 11 was new. Sure has been some time since the last update….

-Dave

[Dave Haynie]

My AMD Radeon HD6970 has been nothing but stellar in Vegas. It’s true that some other things out there in the world don’t work on OpenGL, but want CUDA for their GPGPU coding. Which is of course limited to nVidia stuff. But for editing or rendering, the effect can be rather dramatic, depending on the job. For example, with GPU enabled, I get full screen/full quality playback of the old “red car” Vegas GPU benchmark project at 29.97fps. Without the GPU, it’s very jumpy, and sometimes down to single digits.. and that’s on a 6-core i7.

When Vegas 11 first came out, I did a head to head of the HD6970 vs. the nVidia GeForce GTX570 — because they were about $300 each. The AMD was faster at everything, but it wasn’t day and night. Vegas 12 uses the GPU better, and some plug-ins, like the Main Concept AVC CODEC, use it very effectively.

Which also brings up a problem… the GTX570/580, the HD6970, they ain’t exactly new cards. They’re well understood in Vegas, and they’re supported by Main Concept. We’ve recently established here, I believe, that Main Concept is hard-coding chip revisions in their CODEC, and won’t accelerate for any chip they know. So if getting the best performance in an AVC render is important, the HD6970 or maybe the GTX580 (should be faster than the GTX570 I tested) would be good choices. That’s because the HD7xxx, GTX6xx, and newer devices, while supported for Vegas internal GPGPU stuff, isn’t supported in Main Concept’s AVC encoder. If you don’t care about AVC and/or you’re mostly trying to speed up editing, perhaps a newer GPU card would be a better bet. I had thought about upgrading the HD6970, but since Vegas is the primary reason I have a good GPU, I figured I’d wait for the next version, and see how that benchmarks on newer hardware. No sense in just guessing.

I know some people are wary of AMD (ATi) driver issues, and certainly there’s some merit there. I had been an ATi fan years ago, and really left them for a some time based on the drivers just plain not working right… and going months without any attention to the bugs. The HD6970 was where I switched back, and I’ve had no problems at all with drivers. In fact, when I did the benchmarking of the two, I compared OpenGL performance in other things, not just Vegas. There were no problems with the GTX5760 in Vegas, but it failed a few of the OpenCL benchmarks back then, as well as being slower than the AMD. But again, the GTX580 might be faster than my AMD, certainly has to be about the same if not faster, and those driver issues were like fall-of-2012 era issues. Certainly not a lingering problem in current drivers… that wasn’t long after nVidia had officially started supporting OpenCL, not just CUDA.

-Dave

[Dave Haynie]

[Barry Hull] “One of the worst offenders is composites, green screen brings renders to a crawl.”

Yup. I did one animation some years ago, with probably 40-50 animated layers, all in green-screen (I hadn’t figured out how to do alpha stuff in Vegas back then). The two minute video took about 4-5 hours to render on my best PC back then. Ouch!

The other thing to consider is that GPU acceleration can really help here, for compositing and some plug-ins, even when there’s no GPU acceleration in the particular CODEC. Obviously, it’s only helpful with it’s stable on your system. It’s been nothing but stable on mine, even back in Vegas 11. Not so for everyone, sadly. GPGPU is still a pretty new thing, as they say, you can tell the pioneers by the arrows in their backs.

Another thing here, when you’re CPU only… just benchmark, for a minute, how much CPU power your PC uses even just playing back a single AVC stream (CPU-only, of course, no video acceleration). Vegas decoding an AVC stream will take more CPU than that. Per stream. It’s not necessarily a small thing. So you have three video tracks, and you’ve discovered that Vegas takes 50% of your CPU to decode each 53 minutes video just for playback, you can be certain that you’re spending as much as 80 minutes per render JUST ON THE VIDEO DECODE! If it’s just pure cuts, no compositing, it would represent over 27 minutes of your render.

-Dave