-
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
1.6ghz 1MB L2, mobile Athlon 64 @ newegg...
- Thread starter amdking
- Start date
- Joined
- Feb 4, 2003
- Location
- SF Bay Area
I don't overclock at all from the BIOS. I use a config script at boot to overclock within Windows. But the 3200+ mobile would probably still have the best chance for a high overclock.hitechjb1 said:
Regarding to memory bandwdith efficiency, I saw the 95%+ efficiency when the memory divider is higher than CPU multiplier, or memory bus frequency is lower than the FSB setting. E.g. in the earlier post, CPU multiplier of x10, memory divider of 12.
Have you tested to see whether memory efficiency can be maintained if memory bus is set to run much higher using a lower divider.
E.g.
CPU at 2000 MHz
Memory frequency at 2000 / 8 = 250 MHz
Question is:
Does memory bandwidth efficiency remain consistently constant over a range of memory bus frequency, say from 200 - 250 MHz while CPU frequency remains constant, say 2000 MHz or 2250 MHz or 2500 MHz.
Have you tested to see whether memory efficiency can be maintained if memory bus is set to run much higher using a lower divider.
E.g.
CPU at 2000 MHz
Memory frequency at 2000 / 8 = 250 MHz
Question is:
Does memory bandwidth efficiency remain consistently constant over a range of memory bus frequency, say from 200 - 250 MHz while CPU frequency remains constant, say 2000 MHz or 2250 MHz or 2500 MHz.
- Joined
- Feb 4, 2003
- Location
- SF Bay Area
First data point I've got is 3183 at 209MHz at CPU/12 for 95.1% efficiency. CPU at 250x10. Timings were at 7-2-2-2.
I have 3053 in sync at 200x10 with 7-2-2-2.5(does not allow me to boot at CAS 2). 3053/3200= 95.4% efficiency.
3181 in sync at 209x10 with 7-2-2-2.5 timings, 95.1% efficiency, only a 2 mb/sec difference compared to async, quite negligible.
Anything else in particular I should try out?
I have 3053 in sync at 200x10 with 7-2-2-2.5(does not allow me to boot at CAS 2). 3053/3200= 95.4% efficiency.
3181 in sync at 209x10 with 7-2-2-2.5 timings, 95.1% efficiency, only a 2 mb/sec difference compared to async, quite negligible.
Anything else in particular I should try out?
- Joined
- Feb 4, 2003
- Location
- SF Bay Area
Worth mentioning, the memory cannot be run higher than the HTT.
230x10 with 7-4-4-2.5 yields 3489 mb/sec, 94.8% efficiency. The slight drop is due to the loosening of the timings.
Computer just hung changing multiplier down to 9x
:
It doesn't sit well with rapid speed changes. Coming up in a sec, though.
Edit: Alright 230x8=1845mhz, CPU/8 mem divider yields 3486mb/sec.
As you can see, the memory controller is quite flexible, and delivers quite consistent performance regardless of dividers, CPU speed, timings, etc.
IIRC, Maxvla had about the exact same efficiency running at much higher speeds with looser timings.
230x10 with 7-4-4-2.5 yields 3489 mb/sec, 94.8% efficiency. The slight drop is due to the loosening of the timings.
Computer just hung changing multiplier down to 9x
:It doesn't sit well with rapid speed changes. Coming up in a sec, though.
Edit: Alright 230x8=1845mhz, CPU/8 mem divider yields 3486mb/sec.
As you can see, the memory controller is quite flexible, and delivers quite consistent performance regardless of dividers, CPU speed, timings, etc.
IIRC, Maxvla had about the exact same efficiency running at much higher speeds with looser timings.
Last edited:
Thanks for the tests, ....
Assume eventually the BIOS would provide 350 FSB setting for overclocking, ..., to max out CPU, memory, HT in ASYNC mod, would you still be using Clockgen.
What are the good and bad about using Clockgen comparing to BIOS?
One thing I see is that, without using bios setting, how do you setup for memtest run for high CPU and high memory speed (though I use memtest only occasionally).
Is there an equivalent clockgen for Linux?
Assume eventually the BIOS would provide 350 FSB setting for overclocking, ..., to max out CPU, memory, HT in ASYNC mod, would you still be using Clockgen.
What are the good and bad about using Clockgen comparing to BIOS?
One thing I see is that, without using bios setting, how do you setup for memtest run for high CPU and high memory speed (though I use memtest only occasionally).
Is there an equivalent clockgen for Linux?
Last edited:
- Joined
- Feb 4, 2003
- Location
- SF Bay Area
The DTR complicates my setup. Hopefully the newer boards won't suffer from this(I think that the Shuttle already doesn't), but I cannot overclock from the BIOS. Even at a 201MHz HTT, it fails to POST. It will only boot up at stock speeds. I can manipulate timings, dividers and voltages which suffices to do whatever I need. People first adopted ClockGen over BIOS overclocking in normal A64 setups to avoid BIOS corruption that sometimes accompanied HTT's above 270. I don't have hard evidence of this happening; it's just what I've heard. For Linux, there is no alternative that I know of unfortunately. I'm only keeping my memory at 210, and for notorious BH5, there's not much doubt that its stable, but if I were to go higher, not being able to run Memtest could cause problems.
So, optimally, using only the BIOS would be preferable. Clockgen also seems to have problems with changing the multiplier. It hangs often. It seems like changing the multiplier only works reliably when the HTT is at 200mhz. With that said, you way want to opt for the 3000+; 9x300 should give you plenty of headroom.
So, optimally, using only the BIOS would be preferable. Clockgen also seems to have problems with changing the multiplier. It hangs often. It seems like changing the multiplier only works reliably when the HTT is at 200mhz. With that said, you way want to opt for the 3000+; 9x300 should give you plenty of headroom.
Gautam said:
So back to my very original concern about low default mulitplier of x8 and maybe x9 (for extreme cooling) for the A64 2800+ and 2900+ respectively, it looks like such concern is still valid until the motherboard BIOS can provide high FSB setting to at least 320-350 MHz range, ....
So as price and yield mature, higher multiplier CPU is preferred
- as price goes down to make high multiplier CPU more affortable and less price differential to lower multiplier CPU, and
- as yield maturity implies even higher overclock potential which requires higher multiplier for a given max FSB setting.
So my conclusion,
at least a 3000+ with x9 multiplier for air/water cooling
at least a 3200+ with x10 multiplier for water/extreme cooling
PS: from what you described about all these problems with bios/clockgen overclocking, it well could be the bugs/issues in bios, ... in the first generation of motherboards/hardware .... which I usually try to avoid, and so get a second generation 250 GB or K8T800 Pro.
Last edited:
- Joined
- Apr 9, 2004
- Location
- College Park, Maryland
well I got 1Gb of PC3200 memory with me. I really cannot afford to buy additional memory say PC4000. Now having said that, would it make sense for me to buy this mobile A64 chip. It seems that only way to OC this processor is by increasing the FSB to around 300MHz. I am sure my memory wont run 1:1. Even a 5:4 wont work with PC3200. What possible ratio's can I set so that I can use my PC3200 and still get FSB to 300?
thanks
thanks
this is an interesting read, I have the Chaintech NF3 250 non GB version on its way, zipzoomfly.com had a wicked deal now they are sold out, only $77 compared to $95 at newegg. All user reviews looked real good as well as a review at ocbench.com I plan to add a 3000+ mobile to it, and the ram I have is Hynix PC3200 with BT-D43 chips which has done 277 on a Max 3 already.
Once I have the gear I will be back with plenty of questions
CK
Once I have the gear I will be back with plenty of questions
CK
how about the 35w mobile a64?
http://www.newegg.com/app/ViewProductDesc.asp?description=19-103-448&depa=0
http://www.newegg.com/app/ViewProductDesc.asp?description=19-103-448&depa=0
- Thread Starter
- #33
yazz88 said:
HOLY ***** they keep adding these mobile athlon 64's like bugs on a windshield on a road trip. This is insane haha, they added a new mobile yesterday now one today....and wow!
That 1.2v may be a real shiner in overclocking someone should try it out, someone that can get 350FSB =)
- Joined
- Sep 22, 2003
But its a clawhammer....
IIRC, you should get newcastles if you want to get high overclocks w512 cache
IIRC, you should get newcastles if you want to get high overclocks w512 cache
- Joined
- Jul 29, 2001
yazz88 said:
This one is actually a NewCastle, JFYI
im a bit confused reading through this thread. I understood that memory speeds on a64's were set by dividing core clock speed by a memory divisor (eg. CPU/8), which makes the most sense to me, but people keep on mentioning that it is a ratio with the htt bus (eg.5:4). does someone have incorrect info or does this differ on different mobo's?
Similar threads
