Odd bus speeds on Athlon64 3000+

Status
Not open for further replies.

dOm1naTOr

Wise Old Owl
Ya now Im at 230 mhz @3-4-3-7 in dual ch.
Now how to test/bench mem bandwidth [dun say sisoft sandra-its fails to open even after many reinstall].
 

geekgod

Journeyman
@aravind_n20
download everest..
there's an option to benchmark your ram..gives your read,write and copy bandwidths and latency..

however..i'd suggest you check you ram for stability using memest86+ at these speds, if you havent already done so.
 
Last edited:

mobilegeek

Ambassador of Buzz
geekgod said:
@aravind_n20
download everest..
there's an option to benchmark your ram..gives your read,write and copy bandwidths and latency..

however..i'd suggest you check you ram for stability using memest86+ at these speds, if you havent already done so.
man memest86+ 3.5 takes ages to end ... how much time did yours took wen you checked ram, did it show errors
 

geekgod

Journeyman
it will take about 20 mins max..

in my machine it took about 17-18 mins for a single pass..i ran at least two passes.

the max stable clock i could get was 220..@2.5-3-3-8.above that it gave errors. might have gone higher if i had loosened them a bit more or pumped a bit more volts, but temps became very high.also it is better not to loosen timings too much on an athlon64 syste..the on die memory controller likes them tight..:)

also check if you are running the rams at 1T or 2T. you may get higher clocks at 2T, but that doesnt help performance.

it is mandatory to run memtest86+ if u r oc'ing ur ram. might very dangerous for your data if you are running ur system with unstable ram.
 

deathvirus_me

Wise Old Owl
Well , 2T helps to gain in performance in these conditions :

(i) If ur using 2 GB or more RAM , specially or AMD system
(ii) If ur running 2 GB or more RAM in dual channel

Also .. 2,3,3,6 always won't perform better than 2,3,3,8 .. coz if ur running 2 GB or more in dual channel , and ur ram has a higher rating like ddr400 , then 2,3,3,8 will give better mem. performance ... and on nForce 4 chipsets , values above 8 may result in even better performance .. for example u can try 2,3,3,10 ...
 

dOm1naTOr

Wise Old Owl
I get the read/write nd latency score like:
5365/1854/62.4ns
and the scores of
cpu qn/cpo photo wrx/cpu zlib/fpu julia/mandel/Sinjulia as:
4594/19100/30167kbps/2852/4342/2627
Is this scores good nd seems normal.
 
Last edited:

geekgod

Journeyman
deathvirus_me said:
Well , 2T helps to gain in performance in these conditions :

(i) If ur using 2 GB or more RAM , specially or AMD system
(ii) If ur running 2 GB or more RAM in dual channel

Also .. 2,3,3,6 always won't perform better than 2,3,3,8 .. coz if ur running 2 GB or more in dual channel , and ur ram has a higher rating like ddr400 , then 2,3,3,8 will give better mem. performance ... and on nForce 4 chipsets , values above 8 may result in even better performance .. for example u can try 2,3,3,10 ...
what u said is not entirely correct..

command rate of 2T doesnt always translate into better performance..but in some cases into better stability. the performance lag..if any.. wont be noticeable if there is a large amount of ram available.
for athlon64 systems..lower command rate is always preferable..simply because of the fact that because of the on die memory controller,the time needed between successive commands is much less.so a 2T timing( 1 command per two clock cycle) is not necessary.its not that it is always bad..just not necessary.
intel on the other hand..locks the command rate to 2T on its motherboards as the system is more stable at that timing.

same for the tRAS timing..if it is too high, it is not that the performance will always be hampered.though in most cases it will be..bt the controller has to wait for more that it needs to. again that wait wont be evident for a large amount of ram.
__________
latency is high..but thats normal as you are using loose timings for the first three values. tighten them all within 2.5-3 and latency is bound to rise within 50 ns.
in my system with timings of 2-3-3-7..latency is 48.1 ns.

as for the other readings, please tell whether they are for read, write or copy. 1854 seems too low.
 
Last edited:

dOm1naTOr

Wise Old Owl
Ya now ive reached 2.5-3-3-6 @230mhz at 1T.
now read/write/lateccy are
5956/2250/49.7
@geekgod
what is the expected score for write for Athlon 64 based system?
 
Last edited:

deathvirus_me

Wise Old Owl
Ya now ive reached 2.5-3-3-6 @230mhz at 1T.
now read/write/lateccy are
5956/2250/49.7

6023 / 4104 / 53.3 ... this is at 3,3,3,8 @ 202 MHz @ 2T .... now don't tell me i've good mem. modules coz i don't ...

same for the tRAS timing..if it is too high, it is not that the performance will always be hampered.though in most cases it will be..bt the controller has to wait for more that it needs to. again that wait wont be evident for a large amount of ram.

I beg to differ a bit .. higher tRAS means that there is a higher probability that the data written is not correct .. this may result in fetching the same data over repeated cycles , which in turn decreases performance .. from 6 -> 8 i got a solid 150 MBps+ increase with my cheap mem. (mem. read / write ) .. and on a pentium rig , even more .. some people even use 10 ..

Also for AMD CPU's .. a lowered HTL can severely affect mem. performance (in a bad way i.e.) ...
 
Last edited:

geekgod

Journeyman
man..by higher tRAS i didnt mean as high as 13-14..i meant near 10.:) with very high values..the mem perormance will obviously be degraded.

as about your ram benchmark..i think 6 is too low for the memory you are using. it will obviously perform much better at 8.my system doesnt boot or memtest shows errors if i go above ddr400 at values lower than 7..even though it is corsair..

my everest readings..

6178 |
5788 |--------->3-3-3-8 at 1T
51.60|


6053 |
4155 |--------->3-3-3-8 at 2T
57.80|

6181 |
5788 |--------->2-3-3-7 at 1T
47.90|

as you see,in my case,changing between 2-3-3-7 and 3-3-3-8 doesnt alter performance by much except by decreasing the latency.but i think if i jhad better memory and could boot stably at 6..there woud have been a difference.
but changing it to 2T affects the overall performance in a bad way.
__________
@akshayt..
for disabling ncq go to device manager---->uncollapse the branch for ide controllers---->double click on srial ata controller.
in the dialog box that opens, go to the tab for the channel in which the drive you are disabling ncq for is attached and there uncheck "command queuing".restart your pc.

by disabling ncq..i got decrease in cpu utilization by about 2%, and gains in burst transfer rate, avg transfer rate and access times by about 20mbps, 6mbps and 5 ns resp..:). not much..but shows that disabling ncq helps.

ncq is useful if you are into heavy multitasking..as all it does is intelligently queue concurrent read/write requests considering the disk transfer rate, access times, process priority etc. it is especially useful if you are running a server. but for linear functions like file copy etc or for most of the works in a single user system..it is not of much help.
 
Last edited:

dOm1naTOr

Wise Old Owl
Hey how did ya get that much high score on everest for mem write??The reference score of even an Athlon x2 4400+ is given as 2268: see the shot:
 

samrulez

Cyborg Agent
THAT'S COZ THE EVEREST VERSION U R USING IS OLD....
SEE THIS

untitledjy1.jpg
 

dOm1naTOr

Wise Old Owl
Now that Ive downloaded 3.01.652 nd here is the scores:
memread/write/copy/latecy: 7254/7386/5898/49.2 @2.5-3-3-6-1T @480MHZ in Dual channel. HTT at 960Mhz and CPU @2.64 Ghz.

And other benchies in the correct order as:
5072/20169/31234/3155/4916/2750
 
Last edited:
Status
Not open for further replies.
Top Bottom