A64 3200+ vs. A64 3000+

[Shimmishim]

Setup
Shuttle AN50R (with vdimm and vcore mods)
LDT = 3x
A64 3200+ (10 x 200)
A64 3000+ (10 x 200)
2 x 256 megs Kingston HyperX PC3200 (bh-5's) 3.26 vdimm volts
2-2-2-6-11-16 TIMINGS
1:1
BBA 9700 PRO (stock speed 324/310)

Sandra Arithmetic
3200+ = 8362 MIPS / 3159/4118 MFLOPS
3000+ = 8209 MIPS / 3159/4087 MFLOPS

Sandra Mem (unbuffered)
3200+ = 1816 / 1844
3000+ = 1853 / 1871

SuperPi 1M
3200+ = 43 seconds
3000+ = 44 seconds

SuperPi 2M
3200+ = 1:40 seconds
3000+ = 1:42 seconds

3dMark2k1
3200+ = 17,919
3000+ = 17,617

Comanche 4
3200+ = 65.06
3000+ = 62.45

PiFast
3200+ = 61.05 seconds
3000+ = 61.81 seconds

PovRay with XP optimization
3200+ = 29:01 minutes
3000+ = 28:58 minutes

I have screen shots of everything but don't want to link all that junk so here's just a summary of everything!

 

[Corporate Thug]

well done SHIMM!!!

i know there are a lot people trying to choose between the 3200+ and the 3000+!

 

[Vette73]

And think when the 939 chips come out I doubt there will be even that MUCH difference between 1mb and 512kb. Then think if having less cache allows greater overclocks when VIA, nivida, sis, etc.. come out wiht their PCI/AGP lock chipsets.

I also went with the 3000+

 

[Shimmishim]

Originally posted by: shady06
well done SHIMM!!!

i know there are a lot people trying to choose between the 3200+ and the 3000+!

thanks!

performance difference doesn't warrant keeping the 3200+ around anymore

so off to the FS/FT forum this 3200+ goes!

 

[Jhatfie]

Ahhh, but how does the 3000+ overclock compared to your 3200+??? That 2.6ghz from your 3200+ is not to shabby. Best my A64 3000+ will do stable with 1.68v is 2450mhz.

 

[Shimmishim]

so far it's up to 2.61 as well...



i think heat is an issue now!

i think i need an r404a mod for my prommy mach 1.

so maybe i should keep my 3200+ and take advantage of the extra 512 cache... hmm...

 

[Jhatfie]

2.61 is very nice. I think I'd stick with the 3200+ since they OC the same. I'm debating doing the vcore volt mod to see if I can do a little better than my 2.45ghz. I've been as high as 2.52 @1.68v, but it is not very stable and I am having some weird SATARaid error issues when playing games I want to hammer out first before I OC more.

 

[Shimmishim]

Originally posted by: Jhatfie
2.61 is very nice. I think I'd stick with the 3200+ since they OC the same. I'm debating doing the vcore volt mod to see if I can do a little better than my 2.45ghz. I've been as high as 2.52 @1.68v, but it is not very stable and I am having some weird SATARaid error issues when playing games I want to hammer out first before I OC more.

good idea!

do you think your problems may be due to lack of pci lock?

 

[stardust]

AHA! the memory myth is true! newcastle's integrated memory controller does yeild better results than the clawhammer's!!

Shimm, if you don't mind, can you please try to overclock that 3000+ to 2.2ghz and compare something like superpi? According to AMD's ratings, 2.2ghz w/ 512cache and SC ram is equivalent to 2.0ghz w/ 1mbcache with SC ram (Maybe a little faster).

I want to see if AMD's rating system really lives up to it's speeds.

 

[DAPUNISHER]

Yep, the results synch with the reviews done around the web comparing them. I keep hearing the extra cache will play a more significant role when 64bit apps start being used but we'll have upgraded to newer gen CPUs by then so it's irrelevant. I doubt the pci lock on the nF3 250 will provide more overclocking headroom since it looks like most the multi locked single channel A64's top out around where Shim is at for 24/7 usage and not just finishing a 3Dmock bench but one never knows....

 

[cowdog]

Ouch. I paid about $255 for an A64 3200, and I would rather have used the extra 50 or so bucks toward a fast SATA drive or a shiny new video card.

Thanks for putting those A64 siblings head to head Shimmishim.

 

[hytek369]

good review

 

[MercenaryForHire]

Anyone with an Athlon 64 2800+ care to chime in to add to this? I'm wondering how badly the 200MHz clock hit affects performance.

- M4H

 

[Dufrane]

I am interested in that comparison also

 

[jdogg707]

thanks for benches!

 

[Jhatfie]

Shimmishim,

I do not think it is the lack of a pci lock on the AN50R. I did a fresh reinstall of XP with everything at default speeds and I get the same errors. It is weird because it is only games that cause the errors resulting in my games locking up. 3dMark can loop for 24hrs, I am Prime95, Memtest stable as well. No other programs including long video encoding sessions (6+ hours) cause the errors. Maybe my onboard Realtek Raid controller is not happy. I've tried several drivers, diffrent bios versions. I am loading up XP with my SATA drives not in Raid to see if that cures my problem. It will be a bummer though since I really enjoyed the extra speed allowed with running Raid0.

 

[Shimmishim]

ahh...

i hear that the pci for the SATA craps out or around 237 mhz.

What people suggested is buying a SATA card with a locked pci frequency of 33 mhz and that'll let you use your drives in RAID configuration at high fsb's!!!

 

[Fricardo]

Originally posted by: MercenaryForHire
Anyone with an Athlon 64 2800+ care to chime in to add to this? I'm wondering how badly the 200MHz clock hit affects performance.

- M4H

Here here!

Considering the identical cache but 200 Mhz clock difference in both situations, I would expect the difference to be about the same between the 3000+ and the 2800+ as it is between the 3400+ and the 3200+. Maybe a little greater difference between the first 2 as the % change in clock is greater.

 

[stardust]

Originally posted by: Shimmishim
ahh...

i hear that the pci for the SATA craps out or around 237 mhz.

What people suggested is buying a SATA card with a locked pci frequency of 33 mhz and that'll let you use your drives in RAID configuration at high fsb's!!!

What I did was just buy a Promise Fastrak TX2Plus that supports 66mhz PCI That way, I can actually benefit from overclocking my PCI clock by getting faster harddrive speeds

 

[mato7742]

i suppose since this thread is a little outdated that adding to it wouldn't really matter since everyone understands the problem here. when this thread was created nobody really knew the technical specs for o/c'ing with pci/agp locked mobo's, howver, now that the testing has be done, overclocking on a pci/agp locked mobo has proven that much higher overclocks are possible. substantially even. on one of the anandtech reviews the editor spoke of overclocks getting as high as 280. that's quite higher than anything one could accomplish without the lock.

so, it's in my opinion a more logical decision to wait for pci/agp lock before you consider half the cache of the 3000+ as a match for the 3200+ because if you can get the speeds of the 3200+ past that of even a 3400+ (maybe even 3700+) the performance improvements will be substantially better.

 

[Shimmishim]

actually...

over at xtreme, a lot of those guys are running an HTT of above 300 w/o the pci lock...

 

[Shimmishim]

Originally posted by: mato7742
i suppose since this thread is a little outdated that adding to it wouldn't really matter since everyone understands the problem here. when this thread was created nobody really knew the technical specs for o/c'ing with pci/agp locked mobo's, howver, now that the testing has be done, overclocking on a pci/agp locked mobo has proven that much higher overclocks are possible. substantially even. on one of the anandtech reviews the editor spoke of overclocks getting as high as 280. that's quite higher than anything one could accomplish without the lock.

so, it's in my opinion a more logical decision to wait for pci/agp lock before you consider half the cache of the 3000+ as a match for the 3200+ because if you can get the speeds of the 3200+ past that of even a 3400+ (maybe even 3700+) the performance improvements will be substantially better.

i dind't run any overclocked tests but the 3000+ would run at 2.6 ghz and the 3200+ would run at 2.6 ghz as well..

and since everyone doesn't overclock i figured stock testing would be more applicable to people here at AT.

btw, i ran my chip at 9 x 289 w/o a pci lock w/o any problems.

 

[kulki]

Did you use the mobile processor or the desktop variant?

 

[Shimmishim]

desktop.

the mobile ones weren't available back then

 

[tji]

I ended up going with a 3200+ CPU rather than the 3000+. For benchmark tests I would expect to see very little difference between the two CPU's, since it's all about the clock speed. (Unless they used a data set > 512KB and < 1MB, then the 3200+ can be made to look very good). But, I think in real world performance, when you're multitasking and dealing with more data, the extra cache will help more. Since I plan to use this board for a Linux server doing a bunch of tasks, I thought the cache might help.

I'm waiting for more socket 939 options to upgrade my HTPC. When I do that one, I'll go for the smaller cache, since it's just streaming huge amounts of data and doing lots of number crunching on the video decompression. In that environment I'll go for the highest clock speed, with less concern for cache.