[Dave Haynie]

[Norman Black] “[John Rofrano] “That processor is not multi-threaded. So you have 4 cores and 4 threads. The Core i7 is multi-threaded so it has 4 cores with 8 threads. That means a processor upgrade to an i7 should double your processing power with 8 threads vs 4 threads.
”

I disagree. The extra logical threads only give you 0-35% better performance. It varies by application. In can be negative!, but in this situation, typically at most 1 percent slower.”

That’s correct. And yeah, it varies by application. What happens in the i7 is the each core has a second set of CPU registers. Any time there’s a pipeline stall for one CPU context (you’re reading memory, for example, which is much slower than CPU internal stuff), the other set, which defines a second thread, is swapped in. So you’re keeping that CPU core busy pretty close to 100% of the time. Without the hyperthreading, it might have been kept busy only 70-80% of the time. That difference is the speedup you get with the dual threads per core.

Not so much anymore, but certainly in the past, and even today on some problems, hyperthreading can slow things down. It’s possible. Why? Caching. You’re running twice as much code through the same sized on-chip caches. If you don’t exceed their capacity with those extra threads, things go faster. But if one thread is kicking out stuff the other thread needs, then you have cache thrashing and the system runs slower. Today’s caches are large enough that you rarely see this. And if the two threads are doing mostly the same thing, even better chance things go faster… for example, when an MPEG or AVC render is on all cores… same code, different data, good chance it all fits. Also a good chance since after all, Sony’s writing the code for an i7… that’s pretty much a given. No telling if they do much testing on lesser processors, but every company doing pro video work is developing with recent higher-end processors.

-Dave

[Dave Haynie]

[Dave Osbun] “If you can, or if you know somebody that can help you, BUILD YOUR OWN SYSTEM.”

I do conditionally recommend that as well. I have never bought a pre-configured PC for myself. But then again, I’ve done this kind of thing professionally. It’s not for everyone. You will invest some time doing this. And it’s not really for money-saving, but to ensure you’re spending your money where it’ll do you the most good, as well as leaving some room for the future.

If you’re looking for guidance on this, as well as asking around here, you can check out Videoguys DIY systems. They’re definitely building for video users, which isn’t true of many DIY system guides online. The system I built for myself last year isn’t far from their DIY9.5 system (I have more RAM, a larger main SSD, internal RAID10, and the AMD Radeon HD6970 left over from my previous system, which was a 6-core AMD).

-Dave

[Dave Haynie]

I’ve actually had better results with AMD than nVidia over the years. I switched to AMD from Matrox ages ago, because nVidia had crappy analog video. Used nVidia for a few years while AMD struggled with driver issues, but back with AMD since Vegas went to OpenCL, and I found I got more performance for the same money with AMD. The Radeon will certainly deliver higher performance in Vegas itself. Some plug-ins are an issue, though.

For example, the Main Concept AVC CODEC. On the positive side, this CODEC is highly optimized for GPU use; it does nearly everything with the GPU. Using Vegas 12 and my Radeon HD6970 GPU, I see about a 6:1 improvement with everything GPU-accelerated. However, it’s been pretty well confirmed that they’ve coded both OpenCL and CUDA modes to match specific GPU chips. If you’re using anything new, you won’t get the GPU acceleration. Hopefully Sony will address in the next Vegas… even better, hold MainConcept to the same standards as the rest of Vegas.

I don’t know that additional GPU RAM is much of an advantage for GPGPU or video purposes. I chose Windows 7 over Windows 8 last summer, when building my system.

-Dave

[Dave Haynie]

I didn’t realize they had gone to the limited-time-discount, but that was totally predictable. Every pay TV subscription does exactly that to get you hooked. Adobe’s following the well defined template on this… unfortunately, that leads to an inferior product.

-Dave

[Dave Haynie]

Exactly.

Think about your level of satisfaction with the things you own versus those you rent or subscribe to. Are you happier as a home owner or a tenant? Do you get more satisfaction out of what you pay for your cable/satellite or internet service, versus, say, a new $1200-$2000 lens or camera.

Ok, maybe not direct comparisons, but here’s the thing: as John says, they have little incentive to improve your condition. And take a look at subscriptions models like HBO or Comcast or DirecTV. They spend a good deal of money on ads, always looking for new subscribers. But they also lose a bunch of customers every year, people who decide that their service just isn’t worth the money.

And what they do, they optimize the service you’re given to keep that level of attrition at a particular rate. Those are the most expensive people to keep happy, leaving every year. And they would cost substantially more to keep happy than it costs to attract more new viewers.

So they have optimized their business model with the expectation that some people every year will be unhappy enough to leave. But that also guarantees that there are lots of customers almost as unhappy, but not ready to leave yet. Maybe they’re just on the brink — I know I dumped both Dish Network and Hugesnet in 2012, I was just pushed too far at some point. And the cost of leaving is very small, really, it’s just a bother, but it’s surprising that a fairly large number of customers will be unhappy with their service at least part of the time.

Now take that to Adobe, who’s now following the exact same model: you’re going to pay until you’re angry enough to leave. But think of the bar… what if I couldn’t edit PSD files? I’d be leaving behind decades of work…. I’d really have to figure out how important that is (and of course, make sure I had TIFFs of each one). And that’s nothing compared to being deeply invested in the Adobe tools.

It may take time, but there’s an awfully good chance that over time, many CC owners will keep paying but still be very upset with the rate of change. This is good for Adobe, of course.. not just the lack of a need to perform, but the ability to spend their development time where they want to, not where the customer demands. In the past, you didn’t upgrade Photoshop unless it offered enough of a set of new features. That’s a very popular program, and so Adobe had no choice to keep it updated if they wanted to keep making money on it. But there’s not an instant replacement for it, either… sure, other graphics programs, but nothing quite like Photoshop (last I checked). Adobe themselves would be better off improving other tools, those with really strong competition. Not only does that make the product stronger, but it strengthens the chance any given customer drops some other tools and goes all-Adobe.

So this is why I’m at PSCS6 and never changing. That’s also a problem because of Microsoft… they have had a tendency to obsolete old software in various ways. Now sure, no one bought Windows 8, so pro applications are working with Win8 but not moving to it. But eventually, Microsoft could get it right, and if they stick to their old tricks, at some point the “I own it” tools will not run in the new Windows.

Some good could come out of this: there is now a very good reason for a software company to build Adobe-like applications for Adobe users. That really wasn’t true in the past; beginners and hobbyists use scaled down versions of the main apps, while pros have little reason to accept a “clone”.

-Dave

[Dave Haynie]

[Stephen Crye] “Then I tried the Sony AVC encoder and was surprised! using “render using GPU” the GPU utilization went to about 60%, CPU dropped to less than 40%, and it completed the render in 2:04 .

So, clearly the Sony AVC encoder can use Kepler.”

And that’s using Kepler in OpenCL mode. So that also suggests that OpenCL, even if it’s not optimized on Kepler, certainly isn’t broken.

Having played and looked at this after your earlier posts, it was pretty clear to me that the MainConcept CODEC has an enumerated list of known devices it will use with OpenCL, and it’ll ignore all others. My guess was they they did the same thing with CUDA… you pretty much confirmed that here.

-Dave

[Dave Haynie]

I did a 2 minute video with bits of animation (some GIF loops, some green screened pre – rendered stuff, and it took about six hours to render, on my machine of the day.

Your rendering time has very little to do with the time of render in a situation like this.. The CODEC is going to take about as long for 4 minutes of this video as it does for any other. The rest is Vegas itself, its compositing engine, any plugins, etc.

First thing to check: is your CPU at 90% or better? If not, could be your files thrashing a hard drive. Unlikely if you have just the main video and some small GIFs, but worth checking. Is Vegas scaling and interpolating your GIF frames to match the output? You MIGHT get a faster render if you pretender the animations to uncompressed AVIs with transparency. The might part is that this increases the demand on your disc drive while lowering demand on the CPU.

If you don’t already, enable GPU acceleration in Vegas itself, not judt the CODEC. The GPU math is a good match to compositing and some plugins.

-Dave

[Dave Haynie]

Yeah, it has me drooling a bit. Plus, I already have some m43 glass, and having done a bit of upgrading my Canon system last year, I was kind of looking at the m43 stuff anyway. And if the Metabones EF to m43 adaptor is ever released, it gets even more interestng.

So yeah, there’s lots here. HD recording at 200Mb/s. Both TV and cinema 4K formats. And a cure to one of main DSLR pet peeves: there’s no 29’59” cutoff, “because Europe”, to get in the way.

And even the other under $2K 4K camera, the AX100 from Sony, pretty interesting… a 1″ sensor in a handycam.

On the other hand, with NAB this spring and Photokina in the fall, I bet this is an interesting year in 4K. I’m skeptical Canon will push that doen to model that doesn’t cost a bundle and have a “C” in the name. Nikon might jump on it, though, to go after the 5D video market… if Sony will sell them a sensor that’s fast enough. But either way, the 4K game does seem to be on.

-Dave

[Dave Haynie]

There are several different wheels in motion here, so it’s a little involved. In Vegas Video Perferences, there’s a setting to use the GPU. This affects Vegas itself: compositing, some effects that get their GPzu settings directly from Vegas. This always uses OpenCL, not CUDA.

There is a CODEC-specific setting in some renderers. For example, Sony’s AVC renderer can “Use GPU if available” — again, OpenCL only. Mainconcept’s AVC renderer lets you select CUDA, OpenCL, or CPU-only.

To really assess the effect of GPU acceleration on your system, you need to try all combinations, Vegas and CODEC with GPU on and off.

Now, the problems I know about. The one that affects nVidia users is the Fermi vs. Kepler issue. It has been widely reported (eg, outside of Vegas use) that OpenCL performance can suck on the newer Kepler architecture. That seems to be entirely nVidia’s implementation, not a Vegas issue. In fact, OpenCL itself changed very little from the Fermi to Kepler generation, but it’s basically a compiler.. it is necessary for these things to be optimized to the GPU architecture.

Next known problem: the Main Concept CODEC only supports the OpenCL deviced it knew about at release time. So only AMD GPUs, and only through the HD6xxx generation. I have a theory they did the same for CUDA as well. That would explain a newer nVidia not getting a boost even on CUDA mode.

-Dave

[Dave Haynie]

I saw a blurb on that camera. One thing that makes me nervous about it: thst single 1/2.3″ sensor. Relative to HD, your pixels shrink by 75% in area for the same sensor size. And ok, I’ve been a little spoiled by the lowlight chops of DSLRs. But even the AX1’s little brother, the AX100 — just announced at CES — sports a 1″ sensor. That’s as far as I know the largest sensor in any fixed-lens camcorder. I remember the early days of HDV, that first single chip JVC with the weird color matrix and absolutely horrible low light performance. And the Sonys that shortly followed, not much better or compromised on actual resolution.

Not trying to pin any balloons here, just urging caution.

-Dave