1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.
  2. Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

    Dismiss Notice

Definitive answers to overclocking A64

Discussion in 'CPUs' started by Soul Rider, 7 Mar 2006.

  1. Soul Rider

    Gangster

    Joined: 29 Nov 2005

    Posts: 434

    Location: Milton Keynes

    There are many different questions and theories pertaining to getting the best results from overclocking an A64 chip - should i drop the LT multiplier? Should I use memory dividers? and many other questions.

    This article tells you all exactly how it works and what effects the various aspects of changing LDT multipliers, memory dividers and FSB settings actually gives you.

    If you've ever been curious as to what is important to get an good A64 overclock, this is for you...

    http://www.neoseeker.com/Articles/Hardware/Guides/athlon64oc/

    Apologies if it's already been posted, but i think it is very informative and i haven't seen it here before.
     
  2. Jamie1984

    Gangster

    Joined: 20 Oct 2005

    Posts: 427

    Location: Newcastle Upon Tyne

    agreed, some nice testing going on there and a very informative, good read.
     
  3. Squidge

    Wise Guy

    Joined: 18 Oct 2002

    Posts: 2,185

    It has been posted before, but it's a good informative read, so I don't think anyone will complain about a repost.

    Also, after the number of articles and the like stating how the A64 prefer tight memory timings, I've been spending a few hours benchmarking memory at various timings, and the onlytimings that seem to make a noticable difference is the Trp, which can't be changed much anyway. So I'm now going to relax the timings and max out frequency :)
     
  4. Soul Rider

    Gangster

    Joined: 29 Nov 2005

    Posts: 434

    Location: Milton Keynes

    Keep us updated squidge on how this goes. I'm personnally now rethinking the need for PC4000 RAM. Although there is a benefit to overclocking the RAM, the use of dividers doesn't hold it back too much. I might go for 2GB of cheaper RAM, rather than 1GB of expensive.
     
  5. Dutch Guy

    Capodecina

    Joined: 18 Oct 2002

    Posts: 24,563

    Location: Amsterdam,The Netherlands

    The only thing missing form the article is how memory timings affect the speed in several programs.
     
  6. Soul Rider

    Gangster

    Joined: 29 Nov 2005

    Posts: 434

    Location: Milton Keynes

    True, are you volunteering to test this out Dutch Guy?
     
    Last edited: 8 Mar 2006
  7. Explicit

    Wise Guy

    Joined: 11 Feb 2006

    Posts: 2,186

    Yeah, that's a very informative read, good find.

    As above, the impact of memory timings would have been the finishing touch.
     
  8. Dutch Guy

    Capodecina

    Joined: 18 Oct 2002

    Posts: 24,563

    Location: Amsterdam,The Netherlands

    I wish I had the time, I work fulltime and have a 3 yr old son coupled with a house renovation means little free time :(

    I could do a test at my current 250Mh memoryspeed by testing 2-2-2-10-1T down to 3-4-4-10-2T with a few programs.
     
  9. geeza

    Banned

    Joined: 30 Aug 2004

    Posts: 9,212

    'and the absolute best way to squeeze performance out of your K8 system is to avoid the use of a memory divider, though that requires high-end memory.'

    im on a 150divider at 280x10 at the moment. Ill let you know how my gskillpc4000 holds up on no divider :)
     
  10. Larok

    Gangster

    Joined: 30 Oct 2005

    Posts: 340

    great link, thanks!
     
  11. Squidge

    Wise Guy

    Joined: 18 Oct 2002

    Posts: 2,185

    Well, after hours of fiddling around with memory timings, voltages, etc I found that on my particular system I lost of about 400MB/sec of memory bandwidth by relaxing the memory timings, but I gained over 1000MB/sec by increasing the frequency. So a nice result of 600MB/sec increase.

    Strange thing was, the memory seems to be most stable at 2.4V :D any higher than that and it fails in the BIOS's MemTest86. I always thought overclocking required more voltage than stock, not less...

    The memory bandwidth increase changed my 3dMark06 score from 3852 to 3860, so clearly not of much use there. Several games seem a little quicker though (couple of fps).
     
  12. mj3zero

    Gangster

    Joined: 18 Oct 2002

    Posts: 221

    Location: Behind YOU !


    I am a total nooby where it comes to A64 clocking but I just changed my system from 10*270 with memory at 180/360 to 9*300 with the memory at just under 200/400 (the a8n32 dividers are weird - I thought it would be 200 exact but its not :confused: ). The only thing I've tried so far is SuperPi's 1M bench and it ran nearly 1 second quicker.

    Whilst I guess total clock speed is king, working that with the fastest memory speed looks like the icing on the cake. I think I need some fast memory now to see how far this board and chip will go.

    mj
     
  13. geeza

    Banned

    Joined: 30 Aug 2004

    Posts: 9,212

    going from 200mhz memory to 250mhz i gained 1000extra marks in 3dmark 01 and 1 second in superPI. im sure games feel faster but that might be just in my head
     
  14. Dutch Guy

    Capodecina

    Joined: 18 Oct 2002

    Posts: 24,563

    Location: Amsterdam,The Netherlands

    In CPU/system limited games the difference will be more noticable than in GPU limited games.
     
  15. Soul Rider

    Gangster

    Joined: 29 Nov 2005

    Posts: 434

    Location: Milton Keynes

    Just summing up so far -

    Squidge discovered a loss of 400MB/Sec when he slackened his timings, but managed over a 1000MB/Sec by getting the higher frequency the looser timings allowed, showing that sacrificing timings for the frequency resulted in an overall gain on the memory bandwidth.

    mj3zero discovered that changing his 2.7Ghz clock from 10*270 to 9*300 and thus increasing his memory speed resulted in a SuperPi bench imporovement of 1 sec.

    This is great guys, this extra research added to the tests in the OP really give us an insight into what produces the best overclock possible.
     
  16. geeza

    Banned

    Joined: 30 Aug 2004

    Posts: 9,212

    ^ so its a totally mixed bag all around then!
     
  17. Soul Rider

    Gangster

    Joined: 29 Nov 2005

    Posts: 434

    Location: Milton Keynes

    What?

    They are supporting the same evidence. Loosening timings to improve frequency means better results in squidge's example, and mj3zero found the extra memory bandwidth from the higher bus increased speed.

    So both say increased memory frequency is better for the system, how is that a mixed bag?

    Two different tests, same results..
     
  18. Soul Rider

    Gangster

    Joined: 29 Nov 2005

    Posts: 434

    Location: Milton Keynes

    Sorry for the threadromancy, but it's only a couple of days and i fell this fits in with this topic...

    Lots of people have had problems running their memory at 1t with these processors. Someone in the forums, forgot name and thread :eek: , was unable to run at 1t timings, so had them relaxed to 2t, however....

    The lowered the burst rate from 4 - 8 and when they tried to tighten the memory to run at 1t it was stable.

    Now here's the question i would like to find the answer to...

    Changing the burst rate down to 8 decreases memory performance, increasing to 1t increases performance.

    Is there anyone out there who could try testing this, 8 bursts @ 1t, then 4 bursts @ 2t, then 8 bursts @ 2t. Keeping all other memory timings the same.

    Can we see how much of a decrease in performance is brought on by 8 burst timing, and how much of an increase is brought on by 1t. This way we can see whether this is a worthwhile way of getting memory running @1t.

    I would do it but my current memory can't run @1t.