3.2G Pentium4 with Hyperthreading, 800 FSB - Twin 1G (2x512) @ 2, 2, 2, 5, 2 IDE (1 reading and 1 writing)
Input: 29min and 45 seconds DV input
Output: Mpeg2 - DVD NTSC template - Added 2 Pass, 10 Bit DC Coefficient, Audio 384 Kbps
Rendering Time: 50 mins and 33 seconds

Try DSE's rendertest.veg to get a rough idea of how well your computer stacks up.

http://www.sundancemediagroup.com/help/thefile.html

Here are my results:
Pentium 3.0ghz "C" - Northwood core (not Prescoot), hyperthreading on, 800FSB (not 533 or 400)
2X256MB RAM (dual channel)
Rendering to and from 160GB WD 8MB cache (Hard drive speed should make no difference on this render. On renders similar to file copies, hard drive speed will make a difference when it is the bottleneck).
Didn't have any other programs that gobble CPU (at least I tried to close all unnecessary processes).
*1:30 (90seconds) at best quality. (In real projects, you'd rarely use best quality. For valid rendertest results, use best.)

Memory timings don't matter (I tried). PAT doesn't matter either.

Render speed seems to scale linearly with CPU clock speed. FSB speed makes a small difference too. HT makes a difference when the load on both processors is about equal. (which is not the case with DSE' rendertest).

Pentium CPUs seem to be the ones to buy as they are faster at MPEG2 encoding. In rendertest, Pentiums and AMD64 processors seem to run neck to neck. I haven't gone through the results thorougly.

Here are more data that somewhat affirms linear scaling with the processor frequency.

My system, running Vegas 4.0e, single pass encoding to DVA NTSC video stream, Win XP SP1

Configuration A
(I use this most of the time becasue CPU runs 10 Degree C cooler)

Athlon XP 1500+ (1.25GHz)
768Mb RAM
120 GB 7200 RPM drive
KT266A Chipset
128KB L1, 256KB L2 Cache
FSB 100 MHz (= 200 MHz in AMD Speak, 400 MHz Intel Speak)
Render test takes 4:34 min (274 s)

Configuration B
(I sometimes use this when I need to render, but not overnight)

Repeat test, but bump FSB to 133 MHz
Athlon XP 2000+ (1.66 GHz)
Render test : 3:28 (208s)

IFF Everything scales to XP3000+ (2.5 GHz)
FSB 200 MHz (800 FSB IntelSpeak)
... Then Render test would be 2:18 (138s).

Not scientific, but there seems to be about 25% performance improvement with the Intel system of glennchan than one would get by just linear scaling of an Athlon system. MainConcept encoder optimized for P4, Dual Channel memory, different pipelines, cache size, ...

If I upgraded my system to glennchan's, then my typical 12 hour overnight render would go to 4 hours.

Just for reference; using a Gigabyte 8KNXP motherboard with a 3.2Ghz 800FSB P4 Northwood and 1 gig Corsair XMS DDR-400 memory, it took 1:23 to finish the rendertest.veg test file.

I would really be interested in hearing the results of the test using a similarly configured machine with a Prescott core P4.

John

John,

I just performed the rendering test. The setup is: Gigabyte 8KNXP, 3.2 gig Prescott, 1 gig Kingston Hyperex 400 DDR, Hyperthreading on. Time was 1:18.

Jan

I have one of the AMD64 3000 chips. It was within 2% of the 3200 that had the extra memory but a LOT cheaper ($299 when it came out). It was also a LOT less expensive than a similar speed Intel (about $599 at the time).

Performance seems to be equivalent to the P4's, but the price was much lower and I have the advantage of having 64 already in the event I choose Windows 64 and if any apps begin to be written for it.

We have also started to use this rendertest in our german spoken vegas-forum - with the small drawback, that it seems to be the better way to treat that as a complete NTSC project.

For questions like HT and overclocking - here the findings seems to be confusing up to now completely. Are you aware that DSE recommends to deactivate HT at all?

http://www.dmnforums.com/cgi-bin/readwholethread.cgi?forum=sonic-foundry_vegas&post=040402001856.htm&toppostid=20735

And that people report that the processor is utliized with 50% only, when HT is activated (when rendering to DV-avi from the timeline)? Only the mainconcept Encoder in Vegas seems to be better here, since mainconcept has optimized the encode for HT - what is in line with the findings here.

And: it seems to make also a difference, if you work with one or two instances - in terms of processor utilization. So, still confusing, because that should not help a lot to render faster, only to be able to continue to edit something while Vegas renders in the background.

"I would really be interested in hearing the results of the test using a similarly configured machine with a Prescott core P4"

Isn't Prescott better at rendering but more sensible to overheating?
I did a lil' search and found this... http://www.cdrlabs.com/phpBB/viewtopic.php?t=17178"

Since I'm buying in a few days when Intel lowers the prices, I'd be interested to know... : )

Prescott cores do run hot - very hot. I expressed my opinion on Prescott cores in a previous thread.

http://mediasoftware.sonypictures.com/forums/ShowMessage.asp?Forum=4&MessageID=299203

My opinion hasn't changed since then. What possible reason could there be to install a furnace in a case full of heat-sensitive equipment?

Jan

Here's my Coles notes version of how hyperthreading works with Vegas.

The work done in Vegas is assigned to two threads.
Thread 1- Video rendering
Thread 2- DV encoding, audio processing

For video editing, usually thread 1 has a lot more work to do than thread 2. On renders like rendertest.veg, thread 2 has very little work to do. On real world projects you may see the workload between both threads is more balanced.

Hyperthreading is Intel's 'party trick' to get one CPU to do the work of two. This can increase performance since not all of the CPU's execution pipelines is in use at once. Hyperthreading gets the CPU to try to do the work of 2 CPUs and is seen by the computer as 2 virtual CPUs. Performance gains from HT on versus off varies widely (from -5% to 50% faster), but usually averages around 15-20% when *both* virtual CPUs are working. Each virtual CPU executes a thread at a time.

When hyperthreading is one, both virtual CPUs have to split the CPU's cache. Each gets 256k instead of 512kb on Northwood core processors. More cache adds a slight performance boost. It might be that for rendertest, disabling hyperthreading will give you a very small performance boost for long renders. If you don't run apps that suffer from hyperthreading (i.e. pro audio, Avid, etc.) then you should leave it on.
*There are actually multiple levels of cache, but we don't need to worry about that too much.

The Prescott vs Northwood core Pentiums:
The Prescott has 1MB of cache instead of 512kb. This helps performance. If you look at other rendertest results with the P4 Extreme Edition and the Athlon FX series (both processors line are versions of equivalent clock speed processors with more cache), you see that the increase cache improves render times.
The Prescott has a longer pipeline (31 steps I believe). A longer pipeline lets the processor achieve higher clock speeds. The disadvantage is that sometimes the pipeline will stall and the computer has to wait for the pipeline to finish before further work can be done. On the Prescott 31 clock cycles go unused if a stall happens. How likely a stall is going to happen depends on the program/thread that is running. For games, Prescott processors tend to be about 6% slower than Northwoods because of the longer pipeline (Doom3 is an exception). For things like video rendering, it looks like the Prescott doesn't run into many pipeline stalls.

Prescotts processors also create more heat and consume more electricity. Electricity costs can really add up.

*Pipeline length also explains why AMD processors are the same speed as or faster than Pentium processors of higher clock speed.