A64 performance

[Speedo]

Hi all! I couldn't think of any better place to ask this question than in these forums.

First, my current rig basically consist of:
Barton 2500+ @2.3GHz running on a Asus A7N8X-Deluxe with 2x512 Gale mem 400Mhz@Cas2. System is running on a 36GB Raptor. GFX is a ATI 9800Pro modded to XT.

Now, I'm currently thinking of upgrading to an s939 A64 3000+ 90nm combined with Asus upcoming nForce4 mobo. Using the Zalman HSF w/120mm fan on the CPU, I'm hoping to get this thing running at up to 2.6Ghz (which I have seen quite some people succeed with).

Anyway, on to my question. I recently built a system for my friend, based on a Sempron 3100+ (1.8GHz) running on Asus K8N S754 with 512MB of cheapo memory (Cas3 etc). GFX is ATI 9600XT. I overclocked it to 2.2GHz, and ran some SuperPI and 3DMark03 CPU scores. Now, the scores compared to my barton system which is clocked just 100MHz higher, was almost identical. I wasn't impressed, but didnt give it that much thought, thinking it was just a "sempron with 256k cache" etc.

However, I just now built another system, almost identical, but with a "real" A64, 3000+. As you know this CPU also runs at 1.8GHz, but it seemed to run perfectly at 2.3GHz, so I left it at that. After everything was installed I thought I'd run it through some benchmarks as well, to get kind of a "preview" of what my coming system would perform like. To my surprise, it didn't score very well either. SuperPI was down to 41s, compared to 44s on my identically clocked barton system. Then I tried the CPU score in 3DMark03 and the A64 scored 840 and my barton system 760. That's 10% higher, not that impressive. Then I did a final test, with UT2004. I set the resolution to 640x480 (to remove the difference in GFX cards as much as possible) and ran to different spots on different maps with the FPS counter running. Comparing the A64 system to my barton system, on identical spots showed an increase of max 7 FPS, and on some spots (which I guess was still gfx card limited?) even 10fps lower FPS.

The thing is that I was really expecting my "new" A64 system to give me around a 40% boost, if I manage to push it to 2.6GHz. But judging by my quick benchmarks and impressions so far, I won't even be close. Does anyone have an idea of why I'm getting so bad performance out of this A64 system? I know the A64 system is only using single channel and cas3, but it shouldn't do that much, should it? Maybe it's not worth upgrading my barton system after all...

I would really like to hear your ideas. (If anyone took the time and read thru my long booring post that is)

 

[klah]

XP-M @ 2.5GHz vs. FX-53(2.4GHz)
http://bit-tech.net/review/309/3

 

[mechBgon]

So you're comparing a system with a 9800XT to a system with a 9600XT?

 

[Speedo]

Originally posted by: mechBgon
So you're comparing a system with a 9800XT to a system with a 9600XT?

Yes, but I was trying to isolate the CPU as much as possible. SuperPI isn't affected by gfx card. Also, CPU score in 3DMark03 is using software rendering, so it shouldnt be much (or at all) affected by the difference in GFX card... In UT I lowered res to 640x480...

klah: Interesting... I will definitely look that article thru...

 

[mechBgon]

If it were me, I'd try the other graphics card for the meaningful benchmark here (UT2004). And I'd run the test on both systems under the conditions I actually use for gaming, since that's what the computer is actually for. I remember DAPUNISHER doing some similar comparisons, and while the 3DMark results were close, in UT Botmatch the A64 pwned the OC'ed Barton, by a significant margin. So much for 3DMark, said I...

 

[Markfw]

The Prescotts' performance was quite dismal, I was expecting rather more from this processor, I must also mention its heat problems; I touched 88'C at one point during ScienceMark! After solving the temperature problems, I managed to get it to sit loaded at a rather modest 77'C - it's quite interesting how this processor is a mainstream processor outputting some serious heat. The FX-53 sat loaded at around 45'C at stock and never crept above 50'C when overclocked to 2520Mhz.

Very interesting how the Barton at 2500 did so well, not to mention the quote above from the conclusion.

 

[Shenkoa]

Comparing 2 systems with 2 different video cards is a waist of time, I dont care what res the machine's are at. Use one card or the other and then post back.

 

[KDKPSJ]

Originally posted by: Shenkoa
Comparing 2 systems with 2 different video cards is a waist of time, I dont care what res the machine's are at. Use one card or the other and then post back.

Agreed, except typo of "waist"

 

[slag]

My 2800+ A64 crunches pi in 36 seconds. This is with 512 mb memory.

This is also at 2520 mhz.

 

[bigal40]

According to the benches in the article posted by KLAH the FX-53 beat out the barton at 2500mhz.
Looking at this article you can see that the 3500/3000 at 2.6ghz beat the FX-53 in almost all benches.
so if you get a winchester running at 2.6 it will be better than the FX-53 which is better than the barton at 2500mhz

 

[uOpt]

I updated from a Barton 2500+ to a newcastle 3400+ socket 754.

The performance gain for my standard benchmarks at the time wasn't really impressive, for video encoding and compilation, and for video encoding 64 bit code didn't do well for Linux' MEncoder.

By now I benchmarked a few more things where the AMD64 shines a little better compared to my P4 2.8C, for php, pythin and CMUCL-compile Lisp code it is about twice as fast, and for mysql even faster if I figured out how to run that stupid benchmark correctly. However, I didn't take these numbrs when I had my Barton.

Netto Doom3 performance didn't improve at all, I still get almost the same framerate out of my GeForce 5900XT in 1024x768 4xAA medium quality, obviously this is where the graphics card shuts off, and the Barton was fast enough. I didn't measure framerates at lower resolutions.

What did improve however was network performance due to the NVidia non-PCI gigabit Ethernet. Totally rocks, and runs under Linux with no trouble if you have a new enough kernel.

 

[Toro 45]

I considered upgrading my Xp-M @2420mhz (210x11.5) a while back but didn't. Not because the A64 wasn't faster but because of the same thing you found. The A64 wasn't a "lot" faster. However I think if you can get an A64 to 2600mhz then the perfromance gain might be worth it. Here's a few mobile Barton scores I got @2420

Super Pi:
1m=42s
2m=1 36s

Sandra CPU - 10012/3811
Sandra Mem - 3296/3099

PCMark:
CPU - 7510
Mem - 6782


When I built my mobile setup the cost of an A64 was a solid $100 more for cpu and mobo now with the price on the A64's coming down, I'd snag up one in a heart beat if I was building.

 

[Speedo]

Thanks a lot guys, I really appreciate your inputs.

klah:
I read that article. Although it felt a little biased towards AMD (even the AMD logo at the end ) it was interesting to see the barton actually performing so well.

Shenkoa & everydae:
I know I could have done that. But actually I was mostly concerned with the SuperPI score, which has nothing to do with the video card.

slag:
Ok... I forgot to mention I used 1MB in SuperPI

MartinCracauer:
Interesting... Yea, well I guess we all expect quite a boost when putting out those $$$ for upgrading the system. I guess its also very important to remember that if expecting to get a boost in games, we must not be GPU limited. About the network boost you mention; don't you have to have the non PCI nVidia LAN in both ends? I actually upgraded my server and two other computers and the switch to gigabit here recently, so a boost in that department is certainly welcome as well

Toro 45: Yea, I would definitely recommend anyone building a new system to go Athlon64. However, in my situation I have a respecable system already and are speculating if its worth the cost of upgrading.

You know how it is; you always try to motivate yourself that an upgrade is really worh it =) And actually it feels that now is a good time to get a "top of the line" system, considering the stall in CPU development the coming year. (The system should feel "fresh" longer).

 

[PwAg]

Here are my specs:

A64 3000 90nm @ 2.4ghz (267x9) - SuperPI 1m = 32s

MSI Neo2/1gb OCZ @ DDR533 1:1

The upgrade has prove "very" considerable performance wise coming from a 2.4C @ 3.0.

 

[fisheye]

CPU marks in 3dmark is definitely affected by more than just the CPU.
I tested my A64 3200+ once with my old 512 pc2100 and ti4200, and again after getting 1024mb pc3200 and a 6800gt.... CPU marks rose 170 pts... about a 20% increase with the exact same CPU at the exact same speed. I think for the CPU test, the rest of the system has to be identical for comparable results.

 

[uOpt]

Originally posted by: Speedo
MartinCracauer:
Interesting... Yea, well I guess we all expect quite a boost when putting out those $$$ for upgrading the system. I guess its also very important to remember that if expecting to get a boost in games, we must not be GPU limited. About the network boost you mention; don't you have to have the non PCI nVidia LAN in both ends? I actually upgraded my server and two other computers and the switch to gigabit here recently, so a boost in that department is certainly welcome as well
[...]
You know how it is; you always try to motivate yourself that an upgrade is really worh it =) And actually it feels that now is a good time to get a "top of the line" system, considering the stall in CPU development the coming year. (The system should feel "fresh" longer).

The other side of the ethernet test is an Intel 875 board with CSA gigabit, so yes, non-PCI on both ends.

Now I only have to figure out why the Fedora kernel gives better NFS server results than the stock/Linus kernel. Strange...

I am getting happier with my new AMD box now that I run more benchmarks. But fact is that the applications where it shines and not shines are opposite of what I need. It runs python and PHP much faster, but who cares, my performance-critical things are written in C or C++ and I certainly don't put anything I want fast into mysql. It only means interactive performance is a little snappier.

Compiling stuff and encoding video or audio is where I sit around twiddling my thumbs, and the P4 Northwood holds up very good there.

The AMD64 upgrade from the Athlon XP was worth it for me because I got rid of the stupid Via chipset, the better Ethernet, the quieter ventilator and I took the opportunity to get 1 GB of ECC RAM. However, I payed $680 for the upgrade and only got $150 out of selling the XP. If I had upgraded for performance that would have been way out of line.

Adding to the cost is he fact that I had to get a new power supply unexpectedly. I had a Sparkle 300 watts on the XP which wouldn't feed the AMD64, although in the end the AMD64 takes less power. Go figure. But I think the AMD64 has more "peak" and powerup demands. So be warned if you try the same trick.

 

[Speedo]

MartinCracauer: Btw, (OT) why is WD banned from your systems? What bad experiences do you have?

 

[uOpt]

Originally posted by: Speedo
MartinCracauer: Btw, (OT) why is WD banned from your systems? What bad experiences do you have?

I first banned WD a few years ago over a firmware screwup they had. People in Linux and FreeBSD were seeing filesystem corruption.

It turned out that they broke soft error correction. A soft error is a recoverable error and normally pretty common and not a big deal. With the offending WD drives, this mechanism was broken. After a soft error, a disk block would be written at the wrong place. I don't have to tell you what it means to overwrite a random block: not only is the file you actually wanted to write incomplete, now you have another random file with junk in the middle or worse, it could hit a directory or allocation table entry. You can even wipe out files on partitions you didn't even mount. WD was kicking and screaming but the facts were undeniable.

These weeks, I decided to throw out the Maxtors I have (2 or the 3 I bought in March went bad, so I don't trust the remaining 2 ones either).

I wanted to make a decision between WD caviar and Seagate barracuda. The WD was more attractive because you can get it with 250 gigs, the Seagate maxes out at 200 right now.

However, I scanned everything I could and found:
- people on newegg gave no negative feedback for the barracudas at all, the WDs had quite a few DOAs and short-lived ones.
- storagereview's database rates these WD drives as better than 5% of all drives, the barracudas are better than 75%. Doh!

Both effects can theoretically be explained by claiming it is based on higher overall sales figures or more idiots buying WDs.

However, what tipped me over are the "RAID edition" drives that WD makes of teh Caviar. Needless to say, there is no need whatsoever for special disk features for RAID. The issue is only reliablity. They are selling the same drives for $60 more just for a better burn-in so that they have a higher chance to survive in an environment where they are actually permanently used. If that isn't a strong sign that the non-RAID drives are not even trusted by WD themself, then I don't know. What jackasses.

My Fedex tracker says the Seagates are at my doorstep. If I start bitching tomorrow you know I have been wrong

 

[carage]

Upgraded from a Barton 2800+ to a Athlon64 3500+, I am very impressed, although I still have my eyes fixed on the FX-55 if I can get one.
Hey, what's a good 2nd hand price for a 2 week old 3500+?

 

[Speedo]

MartinCracauer:
Ok, hehe. I see... Well, good luck with your seagates. Btw, I never had any problems with WD drives. I've seen quite some broken IBM drives though

 

[Speedo]

Its seems to be that the performance advantage with A64 compared to AXP differs much depending on application. I also learned that the cpu score in 3dmark is quite video card dependent after all.


Thanks for all your input guys.