A64 OC Data

Status
Not open for further replies.
R

relic

[H]ard|DCer of the Month - August 2007
Joined
Mar 30, 2001
Messages
9,318
This sticky is for hard data and useful speculation on A64 overclocking.

No fanboy trash or bragging, let's keep it helpful.
 
I think it would be good to get the fsb and timings of a bunch of Athlon 64s so that we can tell what actual difference the cache level makes on an Athlon 64's ability to run at a higher Clock. So here is some initial information.

My Athlon 64

240 fsb 2:1 DDR 480 10x multiplier 2400mhz CPU Volts 1.57

Athlon 64 3000+ 512k Stepping SH7-CO

Mother Board K8V Deluxe Bios Rev 1005.011

Memory Corsair XMS3500 256meg about 1 year old at DDR480 2.0-4-4-4 CMD1

Running 1- Maxtor Diamond Max 80gig +9 and 1- 30gig +6

Both IDE 40gig C: 10gig D: 60gig E: stripped software raid WIN2k

True Power 430 watt

Voltages According to ASUS PROBE Software

+12V 11.84

+5V 4.999

+3.3V 3.312

4 80mm case fans approx 30 cfm each + 1 system slot blower under Radeon 9800

CPU cooled by Zalman 7000A-Cu Temps between low 30's idle to mid to high 40's under load.

Room Temp between 60 and 75 degrees
 
I have an Athlon64 3000+ running at 2350mhz

Asus K8V Deluxe
AMD Retail Fan&Heatsink
BrandX Basement PC2700 (DDR333)

235mhz FSB
177mhz memory clock

Running Toast for 15 minutes pegs the temperature at 67° C

Boots and runs at 2400mhz but has a tendancy to overheat after a couple of minutes when running at 100% processor usage. Sorry, I didn't grab the numbers off my processor core but it is one the very first retail boxed processors, bought it the day it came out.
 
I actually posted this info in another thread, but I figured it belongs here as well.

Max stable overclock so far with stock AMD cooler: 2470 (174.5x9)
Using clockgen for lowering multiplier and raising FSB.

Motherboard: Shuttle AN50R v1.3 (no mods)
CPU: A64 3000+ (stock cooler) CAAOC 0402TPMW
RAM: 1x512MB Mushkin PC3500 level 1
PSU: Thermaltake 480W
HDD: 2x80GB Hitachi SATA Raid 0
DVD: Pioneer DVR-106BK
CDRW: LiteOn 48x
VC: Geforce2 MX400 (AN50R would not work get past bios 90% of the time with my 9500 PRO installed even at default settings, it would just freeze or lose/corrupt the hard drives. Have a 9800 PRO on order and should be here any day that will hopefully work.)

By using clockgen to lower the multiplier I was able to do the following: 2470mhz at 1.65v
Here are my settings (voltages, divider and timings set in bios, FSB and multi in clockgen):

BIOS Settings:
VDIMM: 2.9v
CPU: 1.65v (CPU gets a bit warm when OC'd and running this voltage. I'll wait for my watercooling to run it here 24/7)
LDT: 1.5v
Chipset: Auto
LDT speed: 2.5x
Memclock: 166 (5:6)
Memory Timings: 2-3-6-3 (My mushkin lvl 1 will not run anything other than cas 2 with the AN50R. It will run almost up to 240 with 2-3-6-3 timings @ 2.9v though. Anything higher and I'll need more voltage, but I have not done a volt mod yet)

Clockgen Settings
FSB: 274.5 (was able to hit 280 or 2520, but crashed during benching even with 1.7v)
Multplier: 9x

There is a good article on overclocking the A64 -
here
Quick screenshot of my fooling with sandra and clockgen.
[screenshot

Hope this helps.
 
Athlon 64 3200+ @ 2.2ghz

Mobo:Chaintech ZNF3-150
CPU: A64 3200+ w/retail hsf
RAM: 1x256MB Corsair XMS PC3500 rev1.1
PSU: EG651P-VE 550wtt
HDD: 2x74GB WDRaptors & 1x180GB Maxtor
DVD: Sony +/-DVDRW
CDRW: LiteOn 52x32x52
VC: built by ATi 9800XT (wont run in combination with Kingston HyperX PC3500 512MB sticks, but a Leadtek 5900U runs fines with the Kingston...with a Leadtek Ti4200 128 card, get random blue screens with the Corsair memory ...odd)

BIOS Settings:
AGP: 66mhz
FSB: 220mhz
all votages and other settings are at default
 
A64s cannot have their multipliers adjusted up, they can only be adjusted down. The FX series are completely multiplier unlocked (one of the reasons why you pay a premium).

I have a question. Have you A64 oc'ers noticed any performance difference running memory dividers? Is it like the Athlon XP where any ratio other than 1:1 completely sucks? I noticed that jhatfie runs a divider.

EDIT: Read that link jhatfie gave...very helpful, I suggest everybody read it.
 
Athlon 64 3000+ @ 2.2ghz CAAMC/0344 stepping
5:4 Ratio
Stock Voltage/Stock Cooler
Asus K8V Deluxe
ATI AIW Radeon 9700 Pro
Antec Truepower 430 Watt
 
OK, so I set my voltage in my bios for the CPU to 1.65 and the RAM voltage to 2.8. I was able to hit around 2100mhz with 2-2-2-5 timings on my RAM. Should I loosen the timings on the RAM and go with higher HTT? I was at 2179mhz, but it wasn't stable through Prime95. The AGP was at like 75 or more and I don't think my 9800 pro liked that.
 
the chaintech nforce3 150 mobos' new bios now includes the ability to set you own memory timings now and thus system is stable with 1gig of Geil pc4000 memory now ...but overclocking has something yet to be desired for (thus far) ..before flashing to new bios , I was running at 230mhz on fsb stable with 256meg stick of v1.1 Corsair XMS pc 3500 with 1.7v on core and 2.9v on mem ..couldnt hit it after flashing new bios with the corsair mem ..but geil was now usable ..

...the expirament continues
 
ClockGen has the ability to set the AGP frequency when running in ghost mode. Would this lock the AGP?
 
I just read today that the K8T800 PRO will allow PCI / AGP lock - frequency setting.
 
No A64 boards right now have a working PCI lock. That may change with the SiS 755 chipset and the upcoming K8T800 Pro chipset. However, the ECS board containing the SiS 755 A chipset doesn't have a working PCI lock. Asus is planning on releasing a board with the SiS 755 B chipset that supposedly will have a PCI lock. Considering Via's track record with PCI/AGP locks, I doubt that the K8T800 Pro will have a PCI/AGP lock. Who knows, Via may suprise us.
 
AMD Athlon 64 3400+
Abit KV8-Max3 Mobo
512MB Kingston DDR466 3-4-4-8
9800Pro R360 256MB DDRII
Custom water cooling:
Dangerden RBX
Eheim 1048
Dangerden z-chip
Dangerden maze4 GPU block
Blackice pro
all 1/2" tubing

running at stock voltage 225x11 110 Degrees F load.
video card running 450/stock ram
havent really overclocked much yet, gonna actually try next weekend.
totally stable for 6 hours f@h. ( that's all I've tried so far).

3DMark01SE: 22000
 
*UPDATE*

OEM Athlon 64 3200+ @ 2.35ghz

Mobo:Chaintech ZNF3-150
CPU: A64 3200+ w/SLK-948U & Swiftech NB cooler
RAM: 2X512meg Geil PC4000
PSU: EG651P-VE 550wtt
HDD: 2x74GB WDRaptors & 1x180GB Maxtor
DVD: Sony +/-DVDRW
CDRW: LiteOn 52x32x52
VC: built by ATi 9800XT

BIOS Settings:
AGP: 66mhz
FSB: 235mhz

1.7v on core
2.9v on vdimm

running memory divider at 166(5:4?)
when set at 200 , system wont boot ..so pc3200 stuff would be more then sufficient

running chaintech on version 1 of bios ..the new bios they have out allows for more memory timing adjustments, but I guess I havnt figured them out yet cause I couldnt get anykind of a decent oc with it.
 
Is it just me or am I hoping for to much with trying to keep the RAM 1:1 with the HTT?

Even with lower the LDT multiplier it just seems like trying to crank my RAM up over 210 just isnt stable.

With the ram divider I can hit 2.3Ghz on 1.65v but my ram is down at like 193

I was thinking that maybe its just the ram, but I heard such great things about this Muskin memory I am not sure.

I might try to lower the timings and see what I can do with a 1:1 oc.
 
Originally posted by Dekar12
Is it just me or am I hoping for to much with trying to keep the RAM 1:1 with the HTT?

Even with lower the LDT multiplier it just seems like trying to crank my RAM up over 210 just isnt stable.

With the ram divider I can hit 2.3Ghz on 1.65v but my ram is down at like 193

I was thinking that maybe its just the ram, but I heard such great things about this Muskin memory I am not sure.

I might try to lower the timings and see what I can do with a 1:1 oc.
Click to expand...

cant do 1:1 (200 in bios settings) on ram and get anykind of oc out of system ..have to run mem at 166(5:4?) ..or no boot at 215 and up ...I heard that that setting might act as 1/6 divider ..
 
Two things... slow the timing down on the memory if your gonna o/c and also try upping the voltage for the memory.

Very important you slow memory latency down if you want to use a higher fsb ;)
 
Dekar12,
That Mushkin lvl 2 should do much more than 210 stable. I noticed that you have 2 sticks of 512 though. I have you tried just using one stick at a time to see if that allows for a better OC? Maybe one is a bad stick? Heck even my Mushkin level 1 could bench at 235+ with 2-3-3-6 timings @ 2.9v (not memtest stable though).
 
Originally posted by jhatfie
Dekar12,
That Mushkin lvl 2 should do much more than 210 stable. I noticed that you have 2 sticks of 512 though. I have you tried just using one stick at a time to see if that allows for a better OC? Maybe one is a bad stick? Heck even my Mushkin level 1 could bench at 235+ with 2-3-3-6 timings @ 2.9v (not memtest stable though).
Click to expand...

hmmmmm

I think I might try that. I was thinking about using memtest, but I have just been dredding the time it takes to test. Plus at this point if one stick was bad, I couldnt even get the lvl 2 stuff back since its out of production.
 
its not his memory thats a problem ...I have same problem where I cant run 1:1 and get a decent oc with Geil pc4000 stuff ..one or 2 sticks ..dont matter
 
It may be the motherboard. It may not like running 2x512MB sticks or maybe it does not like 1:1. But, I do not have any problems running 1:1 at 235 with my AN50R and Mushkin lvl.1 (other than memtest complaining). Only reason I run 5:6 now is because my memory will not go any high enough at 2.9v as my A64 is good for 2450mhz stable.
 
I just got my 64 bit setup built this week.

A64 3200+
SLK-948U and a 92mm Vantec Stealth
Abit KV8-MAX3
256MB Kingston PC2100 ** my 1024MB Corsair PC3200 arrives today by FedEx
60gig Maxtor (which is running out quick!)
GFFX 5200 256MB
Leadtek TV2000 XP Expert
Lite-On 48x24x48x16 CD-RW/DVD player
No name 400w PSU

i have it OCed to 2.2 (220x10) right now. i haven't tried to push it beyond that *yet* :D when i get my RAM i will probably push it some more. i will be getting a new PSU soon too so that will help. I may need to invest in a different fan for my 948 b/c it isn't running super cool since i OCed my proc. it pushes 50C when encoding a DVD. but a stock settings it runs high 30s to low 40s
 
Has anyone gotten over 240mhz fsb with memory at the same frequency. I can get to 2500mhz and 250fsb but only with the 3:2 divider. So my pc is faster at 2400mhz or 240x10 and memory at 240mhz ddr. I have Corsair XMS3500.
 
i got to 240 today but i didn't puch it any higher. i had VCore at 1.65. i will probably play with it some more and see what i can do.
 
0322_k8nhassomekick.gif


K8N Pro has a little bit of 'go' in it, doesnt it :D
 
Supaman, how did you get your PCI/AGP at stock 66/33 with such a high HTT? Did a higher divider kick in after you oc'd to a certain HTT to bring them back to spec? Or is ClockGen just fooling me :( ?
 
Originally posted by DaveX
Supaman, how did you get your PCI/AGP at stock 66/33 with such a high HTT? Did a higher divider kick in after you oc'd to a certain HTT to bring them back to spec? Or is ClockGen just fooling me :( ?
Click to expand...

well its nf3, so AGp is locked at 66. it is lying about the PCI speed though. the normal divider is 1/6, which would suggest 58MHz PCI.. but that cant be, so I would suppose a new divider kicks in at some point
 
Everyone's saying you can't AGP or PCI lock on nF3 150. Actually, I always thought that PCI and AGP were directly related; that whatever the AGP speed is the PCI would be half of that??? Me wrong i guess. That's a Gigabyte board you're talking about rite?
 
I read somewhere (AnandTech?) that the Shuttle AN50R and Gigabyte K8NNXP have working locks; perhaps that means the K8N Pro does as well? Would be pretty sweet, as iirc it's one of the cheapest boards as well.
 
Unfortunately, there's no real way of telling without an oscilliscope whether the lock is working or not. The software ones are all crap and don't always work. It might be just that your particular AGP card can handle a higher AGP speed. Some cards do, and some don't.

Also, the pci cards are worse off. Very few of them can handle any overclock at all. That's why it's a good idea to use as many of the onboard features (LAN, Raid, Sound) as possible. That way you can get away with no PCI cards at all, and hopefully get a better OC if the PCI bus isn't locked.

Note that nobody has 100% confirmed a working AGP/PCI lock on any nForce 3 150 board. The 250 version is yet to have enough boards and reviews to confirm a working lock or not.
 
I read somewhere (AnandTech?) that the Shuttle AN50R and Gigabyte K8NNXP have working locks; perhaps that means the K8N Pro does as well? Would be pretty sweet, as iirc it's one of the cheapest boards as well.
Click to expand...
Yes, and i curse Andandtech for putting out that rumor. Although they tested the shuttle board later with an oscilloscope and pointed out their mistake, many people still beleive the shuttle board has a lock. Nothing about the Gigabyte board thoguh.
 
Originally posted by M4d-K10wN
Yes, and i curse Andandtech for putting out that rumor. Although they tested the shuttle board later with an oscilloscope and pointed out their mistake, many people still beleive the shuttle board has a lock. Nothing about the Gigabyte board thoguh.
Click to expand...

Shuttle AN50R has a working AGP lock. There are many who have tested their boards past 300 HTT and there is no way their gpu could've handled such a high speed. Go to XS or OCforums...they will prove how wrong you are. Anandtech only proved the PCI lock was broken on a lot of nForce3 150 boards...it would work and sometimes it wouldn't. Plus PCI/AGP lock was working in the engineering samples, broken PCI or AGP lock is a result of it not being implemented properly.
 
Shuttle AN50R has a working AGP lock. There are many who have tested their boards past 300 HTT and there is no way their gpu could've handled such a high speed. Go to XS or OCforums...they will prove how wrong you are. Anandtech only proved the PCI lock was broken on a lot of nForce3 150 boards...it would work and sometimes it wouldn't. Plus PCI/AGP lock was working in the engineering samples, broken PCI or AGP lock is a result of it not being implemented properly.
Click to expand...
1) PCI lock was broken?? WTF. PCI = AGP/2. It's just a half divider as simple as that. I don't know what you've been smoking.
2) The board was tested later with an oscilloscope to measure the freqency. It does, in fact change.
3) Some video cards are more toleran than others. "over 300 HTT" will give you anywhere from 99MHz to +InfinityMHz in AGP speeds. Most geForce FX cards can easily take 120 , and some even more, with the exception of the 5800 (or the 5700, forget which).
 
Status
Not open for further replies.
Back
Top