Athlon 64 3000+ Overclocking Results

[stardust]

Conclusive results of my A64 overclocking experience..

Pic of benchies

Aquamark3

New SiSandra benches

Pics

From my last post, I tried to achieve 250mhz FSB...

Originally posted by: stardust
Since all the other boards have something like this, I figured.. why not?

I waited months for the nforce3 250, but it still is a no show. I managed to earn quite a bit of money from my part time job so I scratched my itch and bought the Gigabyte GA-K8N Pro. I should be getting my system up and running by thursday, but in the meantime I think we should gather our knowledge about the athlon 64 and try to list things we can do to squeeze as much speed out of the cpu as we can.

From months of research, I've come up with a components list of an athlon 64 system using the nforce3 150 chipset that I think will be most successful for overclocking without breaking the bank. This system also happens to be what I am building

Goal:
-250+ fsb
-DDR500
-600+ HTT.

Special Parts:
-CPU: Athlon 64 3000+
-Motherboard: Gigabyte GA-K8N Pro
-Memory: 2x512mb Geil Ultra Platinum PC4000 (For overclocking, i'll start with a single stick of 512mb)
-Storage Controller: Promise FastTrak S150 TX2plus (Supports up to 66mhz PCI frequency)
Sound Card: Audigy 2 ZS (I heard this card can withstand up to 50mhz on the PCI clock)
-Cooling: Dtek Custom Flowmaster XT Water Cooling kit. (whitewater block and hydor30 pump)

Comments?

Issues:
Now I need some advice on what multiplier's I should try...

From what I've seen, many people clocked their FSB to outrageous 300+mhz speeds, but crippled their HTT speed by setting the LDT as low as 2. Also, the motherboard bios is known to only allow a maximum of 225fsb so overclockers are forced to use Clockgen to lower the multipliers and inflate their fsb.

My questions are, what is the point of high fsb if your HTT speed is limited? How does the synchrony of the memory affect system performance? Is there a way of getting high fsb without the use of clockgen? How will removing PCI speed sensitive devices to allow high fsb help improve performance if you are forced to keep HTT speed @ nominal? How high can I raise the Vcore under the conditions of my system and not significantly raise my chances from falling victim to the famous Athlon64-sudden-death syndrome?

After some unfortunate problems, I was forced to RMA my geil PC4000 memory and I replaced it with 1GB OCZ PC3200 EL.

Since the new sticks of memory didn't like anything higher than DDR404 I was forced to keep a 5:6 ratio. That means, the highest overclock possible on the FSB was 242mhz. Surprisingly, I was able to push my fsb exactly to 242mhz and looped prime95 for 9 hours without any errors. Since I wanted a nice whole # I kept my final system spec at 240fsb.

My final specs:

A64 3000+ @ 2.4ghz
OCZ PC3200EL @ DDR400 @ 2-3-2-6 @ 2.7V
FSB: 240mhz
HTT: 600mhz (2.5 LDT)
PCI: 39.7mhz (Audigy stable)
Graphics card: stock

Pic of benchies
Aquamark3
New SiSandra benches

All in all I was very satisfied with my A64 experience and I will be using my new system for some time!
I was especially pleased to have topped 51000 in Aquamark3

 

[exar333]

Nice OC...im thinkin about gettin a simialr system...i already have the OCZ 3500EL...so it seems I would get pretty similar results with some luck

 

[stardust]

Originally posted by: exar333
Nice OC...im thinkin about gettin a simialr system...i already have the OCZ 3500EL...so it seems I would get pretty similar results with some luck

You'll have more headroom than me. I was able to achieve 300+ fsb with super low LDT multipliers and cpu multipliers. The speed was actually lower because of all the low multipliers and I was really stressing my Audigy card..

I suggest 250fsb for you, but get a board that will allow you to reduce CPU multipliers through bios. Like the Shuttle AN50R.

 

[Shimmishim]

Originally posted by: stardust

Originally posted by: exar333
Nice OC...im thinkin about gettin a simialr system...i already have the OCZ 3500EL...so it seems I would get pretty similar results with some luck

You'll have more headroom than me. I was able to achieve 300+ fsb with super low LDT multipliers and cpu multipliers. The speed was actually lower because of all the low multipliers and I was really stressing my Audigy card..

I suggest 250fsb for you, but get a board that will allow you to reduce CPU multipliers through bios. Like the Shuttle AN50R.

lowering mults through the an50r bios isn't a guaranteed thing... there's a new bios that was released on 4/19 but i dunno how that one was...

the best one to date is the 10/30/03 one and that's the one i still use today...

i just use clockgen to get lower mults...

 

[stardust]

I'm waiting eagerly for a new bios update... this current one isn't very complete..

 

[dripgoss]

How are you able to lower mults in ClockGen? Everytime I try, my system freezes (even if I lower everything else too). Here's my usual tweaks (in BIOS)

230 HTT
LDT 2x
Vcore 1.7
DDR Volt +0.3
5:6 Mem ratio (166)
Memory Timings at Auto (3-3-8-3)

 

[Zebo]

what vid card? how high oc?

So far it seems the barton needs exactly 200 more Mhz to keep up to the 512 cached a64. DAPUSHIER verified this. And in UT the picture gets worser. PLease run UT2003.

 

[BlindBartimaeus]

I believe that is 400mhz. I have seen the benchies, a 2800 A64 is about the same as a 3200XP...so that is 1.8 vs. 2.2

 

[DAPUNISHER]

Originally posted by: BlindBartimaeus
I believe that is 400mhz. I have seen the benchies, a 2800 A64 is about the same as a 3200XP...so that is 1.8 vs. 2.2

For gaming the 2800+ actually beats that 3200+ in most games despite the 400mhz speed difference, for everything else I've tested from POVRAY, Sciencemark, media encoding, SETI, ect. the Barton can hold it's own very well and given 400mhz over the A64 will win most of those handily.

As to the overclocking results he's looking good :beer: I get the same SuperPi score but because he doesn't overclock that 5900 my 5800U was faster in
AQ3 despite only running 2.15ghz 1:1 on a K8USA at the time. This 9800PRO doesn't bench well in AQ3 so no comparison to my current setup.


Anywho, I have the same problem with clockgen locking my system everytime so I'm sticking with bios settings of 240mhz 5:6 2.5LTD until I get a 250GB, then I'll play some more

 

[stardust]

Originally posted by: Zebo
what vid card? how high oc?

So far it seems the barton needs exactly 200 more Mhz to keep up to the 512 cached a64. DAPUSHIER verified this. And in UT the picture gets worser. PLease run UT2003.

Will do.. i'll pm you this weekend

 

[stardust]

Originally posted by: DAPUNISHER

Originally posted by: BlindBartimaeus
I believe that is 400mhz. I have seen the benchies, a 2800 A64 is about the same as a 3200XP...so that is 1.8 vs. 2.2

For gaming the 2800+ actually beats that 3200+ in most games despite the 400mhz speed difference, for everything else I've tested from POVRAY, Sciencemark, media encoding, SETI, ect. the Barton can hold it's own very well and given 400mhz over the A64 will win most of those handily.

As to the overclocking results he's looking good :beer: I get the same SuperPi score but because he doesn't overclock that 5900 my 5800U was faster in
AQ3 despite only running 2.15ghz 1:1 on a K8USA at the time. This 9800PRO doesn't bench well in AQ3 so no comparison to my current setup.


Anywho, I have the same problem with clockgen locking my system everytime so I'm sticking with bios settings of 240mhz 5:6 2.5LTD until I get a 250GB, then I'll play some more

The multiplier thing may be motherboard specific issue, clockgen works fine, actually the system is still on the 10x divider because the system is perfectly stable @ 2.4ghz

New SiSandra benches

 

[Jeff7181]

Any numbers to indicate how running the memory async effects performance?

 

[stardust]

Originally posted by: Jeff7181
Any numbers to indicate how running the memory async effects performance?

They do not affect performance, I'll have more benchies up this weekend..

From my past benchies running Sisandra memtest showed no noticable difference (<1%) in running DDR400 1:1 and DDR400 5:6

I guess asynch only affects athlonxp/mp/barton/pm cpu's and a little on intel cpu's

 

[nycxandy]

Originally posted by: Jeff7181
Any numbers to indicate how running the memory async effects performance?

I'm interested as well.

 

[Jeff7181]

Originally posted by: stardust

Originally posted by: Jeff7181
Any numbers to indicate how running the memory async effects performance?

They do not affect performance, I'll have more benchies up this weekend..

From my past benchies running Sisandra memtest showed no noticable difference (<1%) in running DDR400 1:1 and DDR400 5:6

I guess asynch only affects athlonxp/mp/barton/pm cpu's and a little on intel cpu's

I guess that makes sense since now there's no "FSB" for the RAM to run out of sync with

 

[nycxandy]

Originally posted by: Jeff7181

Originally posted by: stardust

Originally posted by: Jeff7181
Any numbers to indicate how running the memory async effects performance?

They do not affect performance, I'll have more benchies up this weekend..

From my past benchies running Sisandra memtest showed no noticable difference (<1%) in running DDR400 1:1 and DDR400 5:6

I guess asynch only affects athlonxp/mp/barton/pm cpu's and a little on intel cpu's

I guess that makes sense since now there's no "FSB" for the RAM to run out of sync with

So there's no need to buy PC4000 ram to overclock an A64?

 

[DAPUNISHER]

Originally posted by: Jeff7181
Any numbers to indicate how running the memory async effects performance?

stradust is right on for most the gaming I've done, but for Sciencemark, SETIbench, PiFast, SuperPi, POVRAY, you get the picture by now I've seen substantial differences in times because of bandwidth. In fact, I've been screaming that fast timing ram is a waste for A64 outside of better overclocking results because of the inconsequential performance diferences in games, but for those othertasks I mentioned timings can improve performance noticeably. I believe it's Joe2004 whom said that for his chess software timings made a really big difference.

 

[nycxandy]

Originally posted by: DAPUNISHER

Originally posted by: Jeff7181
Any numbers to indicate how running the memory async effects performance?

stradust is right on for most the gaming I've done, but for Sciencemark, SETIbench, PiFast, SuperPi, POVRAY, you get the picture by now I've seen substantial differences in times because of bandwidth. In fact, I've been screaming that fast timing ram is a waste for A64 outside of better overclocking results because of the inconsequential performance diferences in games, but for those othertasks I mentioned timings can improve performance noticeably. I believe it's Joe2004 whom said that for his chess software timings made a really big difference.

So running 1:1 does make a difference ?