HD For Indies

...And the new GeForce 8800 is out too. 128 SIMD processing units, ultra-powerful, yaay! Finally, a GPU fast enough and with high-enough quality to really help us out... when the programmers figure out how to use it, that is.

Also, only PC, so far :(.
 

You make it sound like current GPUs are weak! Sure the GeForce 8800 is uber-powerful but people have built entire GPU based color grading stations off of significantly lower powered cards. The 'old' (as of a few days ago) cards had a lot of power!

Taking advantage of the 8800 should be a lot easier than previous cards which is great! nVidia has developed an entire API that allows the user to write scalable C code for both the GPU and CPU at once! Very cool! The only downside is that it seems to be Windows and Linux only. I'm not sure why that is..
 

Honestly, why does anyone need this? I have a Mac Pro, and even though I edit HD and do Windows development in Parallels, I'm not using even a fraction of what the box can do.
 

...then you're not doing the same kinds of things I do (or some of the other readers here).

My entire career has been based on watching progress bars. I keep finding fancier things to do to slow them down, and the hardware companies try to catch up.

For years I did After Effects for a living, always a sloooooowwww render process. Fire up a big compression, encoding, or rendering job, and you'll want all the help you can get.

However, we need better multiprocessor software, otherwise the hardware is useless/frivolous.

Compressor, for example, can't use more than 200% processor "stuff" as indicated by Activity Monitor.

-mike
 

i have a mac pro here too to work on a HD job, editing in FCP and composit in shake and i often wished the thing would be faster... however, as mike says very few of the operations are able to hog all 400% processor power.. 200% seems to be more common. good thing is that you can continue editing while the shake render is ticking away, the bad thing is that you usually need that render to make editing decisions.

++ chris
ps: oh, and dont try to clean up some big photoshop files while you wait, it really kills multiprocessor performance (actually working in a 500mb multilayer file is pretty painful even if it's the only thin you do, so you wait and wait while the activity monitor reports 350% free cpu cycles)
 

There are some architectural changes that change the game a bit. Some big (such as the whole unified shader, scalar processor thing), some small but still useful. For one thing, the GeForce 8800 can load textures up to 8k x 8k. Previous nVidia and ATI cards' maximum texture size was 2k x 2k. Yes, you can work around this, but still it is a pain if you're trying to do film-res rendering on the GPU, for example, and need higher-res textures for your model. Or if you're trying to grade video > 2k res...

Why is everything Windows and Linux first? Well, that's the way it has always been - back to the days when the high-end Macs shipped with low-end GeForce 2 MX chips and were called "graphics powerhouses"... I think the reason is that the applications that taxed the graphics cards the most were:
- games
- engineering / visualization
- 3D animation software
If you look in all of those markets, the Windows side has a far greater selection of software and a greater user-base.

Hopefully as GPUs become more generally used for editing, After Effects, etc, Apple will catch up.

Bruce
 

Chris - actually, storage makes a HUGE difference there as well. If your PShop scratch disk is where your Shake is reading or writing from, prepare for to suffer. And it is the disk seeks that kill you, so a RAID 0 isn't going to help much.
 

Having owned a quad-core G5 for about a year I have to say that the realistic usage of multi-core processing hasn't matured.

I bought the thing knowing that the extra cores would twiddle their thumbs a lot even with a lot of progress bars going.

There are a lot of underused technologies like 64 bit processing – Photoshop, FCP, AFX and other workhorses don't see more than 4 gigabytes of RAM because they're still 32-bit apps, so you don't get the extra memory benefit of having more than 4 gigs with each of those apps on their own. Conversely you can run FCP and Photoshop at the same time with less scratch disk hits if you have 8 gigs or more of RAM.

And having more RAM accessible pushes the bandwidth and speed potential upward for any large project, like 2K+ footage.

Then you have multi-core versus actual multi-processors which have their own busses instead of squeezing threads to each core on one bus.

And there is disk I/O. That's been a bottleneck for many for years. I've seen folks update to high-end Mac towers but not put any thought into how their physical hard drives are actually dragging the whole system down with large tasks. It makes a difference – I remember hot-rodding my old G4 tower by simply getting a fast hard drive array going. It was like night and day on the same processor and bus configuration.

All this matters with 4 or 8 cores. Speed gets better with more cores but the other factors like disks and software can affect that.

It's kind of fun to update an application and have it use all four cores in my Quad G5 when it didn't before (low-end example: Bryce 6). Then you really appreciate what your machine can do. Cinema 4D eats up my four cores pretty nicely and After Effects/Photoshop doesn't.

I think Leopard will help with this a bit on the OS side since it will use more multiprocessing deep in the guts.

I'm sure Apple will have their 8+ core systems out the door soon enough, but I won't feel totally left out unless I was running 64-bit math-based research or an optimized render farm.
 

To further stress point that power can go to good use... So often we have to accept a trade-off of quality vs. render time. Doing a doc with a lot of SD uprezing, it would be great if we could just use fullest quality scaling + de-interlacing + retiming, but that would take effing forever. We also do a lot of maya rendering and it was taking so long we ended up creating a mini renderfarm from used xeon servers and biy pc's running linux. For us, HD is pushing the limits on our machines and the more processor power, thw more you can do.