Athlon-64 Render Test

Sid_Phillips wrote on 1/7/2004, 6:49 AM
I finally made a little free time to run the render test using the RENDERTEST.VEG file. I ran it both on my new AMD Athlon-64 gaming rig (not optimized for video production) and my Intel P4. Here is how each system is configured:

AMD RIG:
Athlon-64 3200+ CPU
MSI K8T NEO-FIS2R Motherboard
mWave PC3200 400MHz Non-ECC DDR RAM (single stick)
Maxtor 80GB 7200RPM HD with 2MB buffer
ATI Radeon 9800 128MB video card

INTEL RIG:
Intel P4/1.8GHz CPU
IBM M-Pro motherboard (Intel Garilbaldi D850GB)
512MB ECC RDRAM (dual 256MB sticks)
IBM (Hitachi) 18GB Ultra-160 SCSI HD
Maxtor G450 video card

BOTTOM LINE RESULTS:
AMD..... 1:38
INTEL.... 3:04

I tested the AMD system first, using the RENDERTEST.VEG file to render three times to Windows AVI format using the NTSC DV template in Vegas 4.0d. The results were the same all three times, each render taking 1:38. The first render was with everything on the boot drive, the second was with source & target files on the boot drive defragged, the third was with the files on a defragged Maxtor FireWire drive.

The Intel system is constantly defragged so I couldn't run an "unoptimzed". I ran the test three times on it as well. The first test was with everything on the boot drive (Ultra-160 SCSI) which took 3:06. The second test was with the source & target files on a RAID array (Medea iSCSI 60GB) which also took 3:06. The third test was with the source & target files on the same Maxtor FireWire drive as used on the AMD test, which took 3:04.

The preview image on both systems looked and ran about the same, so I didn't see any benefit there. But captured video does run a little smoother and looks a little sharper on the AMD rig. One of the eventual upgrades will be a SATA RAID-0 setup, which might make preview a little smoother as well. But overall I would call it a toss-up between the two, with only a slide edge to the AMD system.

It looks like the difference on the benchmark VEG is around 187% for this particular comparison. Bill Ravens is getting a 1:29 on his tests using a 3.0GHz P4, which would be about a 206% boost in this comparison, and you can definitely get an Intel 3.2GHz rig for the same price as the Athlon-64, so if you're going for the fastest speed right now that's probably the way to go.

My AMD rig is really designed for games more than anything else. The fact that it's 2-3 times faster than my machine at work is just icing on the cake. But if I were buying a system specifically for video editing I would probably go with a P4/3.2EE system, which is undoubtably the fastest processor for video right now.

The only reason I would be tempted by AMD would be for the promise of dramatically increased performance with a 64-bit OS, drivers and apps. But right now I wouldn't bet the farm on it. I might bet one of the cows or a couple of the chickens, but not the farm *grin*. But considering that my "old" home system is an AMD 1200, I am not the least bit disappointed in the new rig. I especially like the flashing skulls on the new ThermalTake case (OK, so I'm just a big kid!)

Seriously though, I do believe the Intel P4 is the best chip for video right now. Anybody out there with a P4/3.2EE they can benchmark for us? Happy new year, ya'll!

Comments

MarkWWW wrote on 1/7/2004, 1:59 PM
I'd also be interested in how the P4/3.2EE performs if anyone has one to try running rendertest.veg on.

Until a few days ago I was still struggling along with a very slow machine by today's standards:

ASUS P2B-S (Intel 440BX chipset)
Intel Celeron 300A (overclocked to 450MHz)
256MB RAM (PC100)
IBM Deskstar 14GXP 10GB Hard Disk

It ran SPOT's rendertest in: 11 minutes 21 seconds

Since I felt I'd really got my money's worth out of this now five-year-old machine (I mainly do audio work and this is still an OK-ish spec for the audio work I used it for) I decided to give myself a treat and buy myself the very fastest thing I could find today. I went for:

ASUS SK8N motherboard (nForce3 chipset)
AMD Athlon64 FX-51
1GB RAM
120GB Maxtor SATA HD (system)
250GB Maxtor SATA HD (Audio, video files)

The new machine runs rendertest.veg in 1 minute 24 seconds which I reckon is probably about the best that can be done today. I'd guess the P4/3.2EE will be pretty much the same in terms of rendertest speed - maybe a touch better, maybe a touch worse. I'd be interested to know for certain - anyone got one to try?

In truth it probably wasn't worth me spending the extra for the FX51 CPU - the speed improvement over your 3200+ is relatively modest, and it's quite a sizeable bit more expensive. Your 3200+ is definitely better value for money. But, it was Xmas and I thought I'd buy myself something really nice - I'd been very good last year, honestly Santa. :-)

Mark
Bill Ravens wrote on 1/7/2004, 5:03 PM
www.anandtech.com has some real interesting comparisons between Intel and Athlon-64. Looks like the AMD really excels at gaming. When it comes to renders, it's practically a draw between the aamd-64 and the Intel 3.2 Gig P4.
Indeed, 64-bit programming will be quite interesting.
TheHappyFriar wrote on 1/7/2004, 7:21 PM
Sid, how much did you spend on your AMD rig? I bet if you built an Intel box for the same $$$ the AMD would outperform it. I don't buy Intel anymore though. After then P3 I feel they've lost their edge. :)
stateofgracie wrote on 1/8/2004, 2:52 AM
64-bit support is not likely to increase performance dramatically. The main advantage of a 64-bit architecture is the addressability of more than 4GB of RAM. This is why 64-bit processors have already been used in servers for years. In desktop processors, this is not a priority yet, which is probably why Intel is not offering a 64-bit desktop processor at the moment. From what I've read about the Athlon-64, the practical performance increase in 64-bit mode can be up to 25%, but in some cases it can be lower than that or even slower than in 32-bit mode. As an aside, due to its architecture, the Power4 (used in Apple's G5) is likely to show no performance increase at all when operated in 64-bit mode.
bcbarnes wrote on 1/8/2004, 4:21 AM
>>The main advantage of a 64-bit architecture is the addressability ...

That's not true in the case of the Amd64 processors (Opteron and Athlon64). While they DO provide access to 64 bit addresses for those that need it - i.e. large database apps, etc. - the big advantage as I see it for most is the addition of extra integer and floating point registers. This will allow compilers to generate code that keeps more information inside the CPU rather than having to dump it to memory. The X86 architecture is well known for being "register starved" and the Amd64 architecture helps this out alot. Microsoft is also moving away from using the X87 for floating point arithmetic in Windows64, and moving more toward using SSE/SSE2, which is much faster.
Bill Ravens wrote on 1/8/2004, 6:42 AM
SGI and SUN have used 64 bit architecture for years, with UNIX OS, of course. Sun machines are de-riguer in intensive graphics applications. UNIX is, of course a 64-bit OS, as are all the apps that run on UNIX.
stateofgracie wrote on 1/8/2004, 8:33 AM
What I meant was that 64-bit software is unlikely to improve on things. The Athlon-64 is a fast processor, as is evident in the render tests. People were making statements about further speed increases when 64-bit software becomes available. It is a common misconception that 64-bit by itself will result in a speed increase. As bcbarnes explains, the speed increase is mostly due to other things which work in 32-bit just as well. While it is true that some 64-bit processors will be slow when executing 32-bit code, this is not the case for the Athlon-64.

Sun and SGI have indeed had 64-bit architectures for a while, and these machines are often (mainly?) used as servers or multi-user machines with huge amounts of memory. Note that for number crunching, the floating point units in current 32-bit processors already work with up to 80 bits data width. However, for most calculations (including video), more than 32-bit accuracy is not required. This is why SSE and Altivec are 32-bit - their speed is due to parallel processing.
pelladon wrote on 1/8/2004, 9:08 AM
The biggest advantage of 64-bit is addressability, accessing more than 4Gigs for processes. That's a reality. Speed isn't the main factor here.

Speed will come from better technologies (PCI Express, DDR2, SATA RAID, HyperTransport, etc.)
TheHappyFriar wrote on 1/8/2004, 9:22 AM
If 64-bit didn't help speed up big number crunching, how could you explain 64-bit SGI O2's @ 180 & 200mhz (with software built for them) out performing duel Pentium 2's (450mhz) with duel proccessors (with the same software optomized for Intel and NT)? I've seen it. Maya 1 on SGI's could outperform duel Wintel boxes. The intel boxes has more memory and more hd space.

If a program only cares about the mhz, then the 64-bit won't help. But if the software is a big cruncher and designed for it, 64-bit will blow by. This has been proven by id software's Quake 3. It cares more about raw mhz then houre power. But Vegas (and other NLE's and 3d programs and most games) will want the hourse power. It's just like bus speed. a higher bus speed will help more then a higher multiplier.

And yes, the added RAM will be good, but most of us won't even get 4gb anytime soon..

The key however is getting the program coded for 64-bit and the OS to 64-bit (windows mainly). If Sony decided to convert Vegas from Windows to UNIX (or Linux) on a PC, I think we'd all be in heaven. Hey, then Vegas could probley work on SGI's, Suns, Mac's.

Also, I check a couple days ago SGI no longer uses IRIX (their version of Unix). They use Linux now. :)


pelladon wrote on 1/8/2004, 9:37 AM
Oh geez,

Anything is better than Intel's FPU. Everyone knows that, that's why everyone is going to SSE/SSE2.

64-bits is mostly moving data. If you have large data sets (hint: database), this can help out. But it also helps to have a 64-bit OS, a 64-bit written app, a system whose bandwidth can handle the load without starving the processor, etc..

Unix-boxes had these things. The PC is finally getting there.

BTW: how's itanium doing?
defucius wrote on 1/8/2004, 10:00 AM
A point that's missing from the previous discussions is that the 64 bit CPUs have extra-wide instructions in their instruction set. These instructions pack 2 or more 32 bit equivalent instructions in one, and saves time in the memory fetch phase of the execution cycle. That's one of the reasons that lower clocked 64-bit CPUs out perform higher clocked 32-bit CPUs.
TheDingo wrote on 1/8/2004, 12:03 PM
>> how much did you spend on your AMD rig?

I just built a new PC for video editing that uses the new AMD64 3000 CPU.

Here's my cost breakdown in Canadian dollars :

$ 315 AMD64 3000 CPU
$ 189 ASUS K8V Deluxe motherboard
$ 218 2 x 512 MB Samsung DDR 3200 RAM
$ 179 Antec 1080 AMG Server case w/ Antec True 430 Watt PS ( very quiet )
$ 132 Seagate 120.0GB w/ 8MB cache ( very quiet too )
$ 66 LiteOn 48X CD-RW/16X DVD-ROM Combo
$ 164 Pioneer DVR-106 DVD+/-RW
-------------------------------------------------------------------------------------------------
$ 1,263 Total without tax

My old PC used an AMD XP 2600 on an ASUS A7N8X Deluxe motherboard, and
takes 2 min 34 sec to complete the rendertest .

This new one I just built finishes the same test in 1 min 32 sec, so I'm really happy
with it. Especially since this is the slowest CPU made for this motherboard!
pelladon wrote on 1/8/2004, 12:04 PM
defucius,

do you have a source for this? Links?
Bill Ravens wrote on 1/8/2004, 12:12 PM
I used to work for a place that did a lot of computerized graphical analysis. In those days, SUN was the only way to fly because of its ability to number crunch. The pipe is bigger with 64 bit, it's not necessarily a CPU speed thing, altho' CPU speed certainly has an effect. It's like water pressure. You can up the water pressure for a given pipe size until, at some point, the pipe chokes and more pressure won't make more flow. At that point, you need to increase the pipe diameter. Enter 64 bit. Now you can start upping the water pressure again...at least until the next pipe size up chokes...then repeat the process.. 128 bit anyone?

BTW, Linux IS UNIX....just Linus Torvalds'interpretation instead of SunOS's.
defucius wrote on 1/8/2004, 2:07 PM
pelladon, you sounded you are really interested in computer architechture topics. I happen to know a couple of good professors at very good universities, you interested in doing a PHD? :)

The benefit of 64-bit architechture really does not stop at addressing more memory. A simple example is reading data from memory -- which is very expensive measured in CPU time. for the same amount of data, 64-bit systems will read in half the time, clock by clock. Same goes for loading instructions for decoding. By reading in instructions faster, you pay less penalty to missed branch prediction, etc. Of course, things are not that straightforward when it comes to real world performance.

There are quite a few article on the web, some specific to the AMD architecture. Just google with keywords "64-bit instruction set." Here is an easy read:
http://bwrc.eecs.berkeley.edu/CIC/otherpr/epic-info.html

A more in depth read might be:
Computer Architecture: A Quantitative Approach (3rd Edition)
this is famed as the computer architecture bible, and the authers have kept it very up to date so far.
pelladon wrote on 1/8/2004, 3:44 PM
The only thing I'm interested in is real-world results.

No one disputes the advantage of 64-bit processing, but the reality shows we have a ways to go. The immediate benefit of greater memory capacity is a given. Apple's G5 can support 8GB of memory. AMD claimes 1TB (terabyte) for their AMD64 processors.

However, the rest of the world needs to catch up. Like drivers need to be rewritten to 64-bits. Having a OS running in 64-bits would help to. And that's going to take a while.

Is anyone using the 64-bit version of Linux?
TheHappyFriar wrote on 1/9/2004, 5:09 AM
I know, but Linux is free (to do whatever you want with) software and unix isn't, am i correct?

I wonder when microsoft will go totaly unix/linux? I even found a commercial app that will alow you to un Windows 95-xp under linux. I forget the web site though (found a link to it on www.linux.org)
Bill Ravens wrote on 1/9/2004, 8:01 AM
UNIX was originally developed at a university in California, either UCSB or UCLA, I can't remember. It was developed as an open source, higher order language, contrary to the Microsoft approach of closed or proprietary source. Many people, including IBM, DEC, Sun, etc. developed their own proprietary OS's based on the open source UNIX.

Bill Gates, in all his wisdom, rejected open source, and you can see the modern day result. To Microsoft's credit, they've created programming standards that the industry couldn't seem to produce on its own. Yet, the downside to it is the stranglehold that M$ has on the marketplace. I doubt, seriously, they'll ever go UNIX, unless there's a mass exodus to UNIX based systems like LINUX.
DataMeister wrote on 1/9/2004, 9:32 AM
Actually UNIX is not an open source operating system. It was developed by Bell Laboratories. LINUX however is open source and was designed to basicly be an open source UNIX. LINUX is what made "open source" popular.

JBJones
Sid_Phillips wrote on 1/10/2004, 11:55 AM
The HappyFriar wrote:

"Sid, how much did you spend on your AMD rig? I bet if you built an Intel box for the same $$$ the AMD would outperform it. I don't buy Intel anymore though. After then P3 I feel they've lost their edge. :)"

The cost of my system was $1690, which included an $80 build fee (I didn't feel like building it myself during the holidays, waaaaay too much going on!) I could have specified an Intel P4/3.2GHz system for about the same price and the rendertest.veg would have probably run 10-15 seconds faster. When it comes to rendering the Intel machines are just better right now.

As I said, though, this particular machihe was really for gaming, not video. And it's a goooooooood machine for games! The fact that it does so well on video is just a nice side-benefit.
TheHappyFriar wrote on 1/10/2004, 12:33 PM
I was wondering because on pricewatch a P3-3.2 is $100 more then the Athlon 64 3200. It's not work the extra 10-15s for me (especialy when i render over night and don't notice the render times!). Also the Athlon 64FX is just under $700 while the Intel P4-3.2 extream is just inder $1k. Wowie. Especialy when the Athlon 64-3400 os just over $400, and probley outperforms then intel 3.2, and compares to the extream.

I miss the day when having 8 megs of RAM and a 486 were a great computer. I was almost "cutting edge" then. :)
zstevek wrote on 1/10/2004, 1:43 PM
I have a P4 3.06 GHz with HT and it took 1.41 to render the video.

AMD's 64 Bit chip is fantastic at 32 bit applications, just imagine what it will be like when the 64 bit windows is released and the programming catches up with it.

I have noticed an improvement in multi-tasking with my P4 HT machine, but more and more I wish I purchased an system with the latest AMD processor in it. There just isn't that big of a speed gain (even though V4 is optimized for HT).

Steve
zstevek wrote on 1/10/2004, 1:55 PM
I also noticed that my CPU is only being used 50 % (+/- a couple) during the rendertest.veg render. Many of the effects you create don't even take advantage of the hyper threading! I know for sure the pan/crop doesn't. This was straight from Sony's tech support. One thing that is nice about the HT is that I can render a video and do other things and it really doesn't seem to slow me down like it did with my old P4 1.7 GHz.


On the other hand if you render a straight AVI to MPEG-2 the full hyper threading capabilities are realized (at least for speeding up the render). I get in the upper 90% range of processor usage and man does it fly through the render.

I can render a 1 hour AVI video to MPEG-2 in a little over an hour.

How long does it take the AMD 64 to render a 1 hour AVI file to MPEG-2 if you don't mind me asking??

Thanks,

Steve