Originally posted by: Zebo
I wish i had a 10 multiplier so i could run 10x265 and the use 166 mem divider giving me 220 which my ram can run 2-2-2 1T for high levels of performance
this is exactly how I run on my 754 3000 which has the 10X
I should have listened to myself above and got the 3200 but $50 was too much a premium. I'm sure the same thoughts are with everyone. cost/benefit swings 3000's way.
would that setting be faster than 10X265 with 1:1 ratio with ram at 2.5-3-3-10 or 2.5-3-3-7 1T?
- Oct 3, 2004
- 5,761
- 12
- 81
the retail heatsink supposebly supports up to 2.4 comfortably...http://www.hexus.net/content/r...JsX3Jldmlld19JRD05MzM=
I dont know how much the memory has to do with it...
i'm running 270fsb on my msi ne02 plat (s939), and 265fsb on my chaintech vnf3 (s754). for some reason i the max i could do on my dfi lanparty ut with the same cpu was 245....
hi
i have teh same mobo (k8n neo 2 paltinum)and cpu (amd 64 3000 939)
I m using corsair twinx3200c2pt...
what would you think my settinsg should be in the corecell and bellow the corecell in the bios...
i m runing at 1.8 right now with a 1.4 bios...3.3.6.3 is my memory setting right now...i also run dual 128bit at 1T
anythoughts would be appreciated....
thanks and
Merry xmas and all that yots
i have teh same mobo (k8n neo 2 paltinum)and cpu (amd 64 3000 939)
I m using corsair twinx3200c2pt...
what would you think my settinsg should be in the corecell and bellow the corecell in the bios...
i m runing at 1.8 right now with a 1.4 bios...3.3.6.3 is my memory setting right now...i also run dual 128bit at 1T
anythoughts would be appreciated....
thanks and
Merry xmas and all that yots
While we are on the subject. Im looking to buy the A64 3200+ myself, but I dont know too much about AMD cpus, is the 130nm or the 90nm better. Ive seen some places say it is slightly cooler and maybe like 4% faster. Is there advantages to having the newer one? Does the motherboards for the newer 90's make any difference? I guess I have no idea which mobo to get yet, which is another problem.
Something to keep in mind is heat produced. From what I understand, and I could be totally wrong on this, processors that are overclocked to a given speed produce more heat than comparable stock chips. For instance, if I take a Pentium 4 2.8Ghz and OC it to 3.4Ghz, it would likely produce more heat than a stock 3.4Ghz chip. Why this is, I don't know and would be interested if anyone has any more of an idea.
when you overclock most need to add more voltage to the cpu to make it stable..more volts=more heat
In many cases a cpu that is overclocked and stable at the default voltage should see little extra temps....today most cpu are multiplier locked at least the higher multipliers and so the fsb is increased which may require more voltage and often will lead to more heat
IMHO for example a A64 3200 (939) that is 1800 at default voltage but say stable at same voltage but clocked at 200 should make little or no increase in the cpu temps....
Also this is very important a retail chip at stock speed and voltage will run different temps on different mobos and system so there is some variaition even in the "normal" situations...high temps are bad..small few degrees...not a big issue
In many cases a cpu that is overclocked and stable at the default voltage should see little extra temps....today most cpu are multiplier locked at least the higher multipliers and so the fsb is increased which may require more voltage and often will lead to more heat
IMHO for example a A64 3200 (939) that is 1800 at default voltage but say stable at same voltage but clocked at 200 should make little or no increase in the cpu temps....
Also this is very important a retail chip at stock speed and voltage will run different temps on different mobos and system so there is some variaition even in the "normal" situations...high temps are bad..small few degrees...not a big issue
Yes, you will be observing the impact of higher memory bandwidth granted to the lower models, BUT as you see even for the 2250Mhz you will need excellent memory paired with the 3000+.
When you reach the 2500Mhz mark -easily attainable by most 0.09 A64 chips-, you will need 250Mhz(DDR500) capable memory for the 3200+ and 278Mhz(DDR556) capable for the 3000+...
As I've said many many times, this is why 1:1 mem is not an otpion for both the 3000+ and the 3200+ in order to OC the CPU to its real potential. Keep in mind that even if you find memory modules of this caliber they will be running in very relaxed timings, so the benefit of higher mem bandwith will be offset by higher latency (how much depends on each OC scenario).
In the end all users have the abillity to max BOTH their CPU freq and Memory bandwidth by using the appropriate CPU multiplier, FSB=HTT and Memory Divider (200, 166, 133, 100).
Example:
Athlon 64 3500+ owner: 11 x 227HTT=2500Mhz, 1:1 mem=227Mhz
Athlon 64 3000+ owner: 9 x 278HTT=2500Mhz, 6:5 mem=227Mhz(11 divider)
GuitarDaddy
Lifer
- Nov 9, 2004
- 11,465
- 1
- 0
Here are some results from my 3200 on an ASUS A8N-SLI with a single 6600gt
Best CPU o/c
270 x 9.5 = 2565mhz
Ram @ 173mhz 2-2-2-5
Aquamark3 Total 60,679 Gfx 8427 CPU 10832
Super PI 1m=35seconds, 512k=17seconds
3dmark05 2982
Best 1:1 o/c
240 x 10 = 2400mhz
Ram @ 240mhz 2.5-3-3-10
Aquamark3 Total 60,876 Gfx 8418 CPU 11149
Super PI 1mb=34seconds, 512k=15seconds
3dmark05 3060
On my rig it seems memory bandwith is the king. At least for these three benchies.
Best CPU o/c
270 x 9.5 = 2565mhz
Ram @ 173mhz 2-2-2-5
Aquamark3 Total 60,679 Gfx 8427 CPU 10832
Super PI 1m=35seconds, 512k=17seconds
3dmark05 2982
Best 1:1 o/c
240 x 10 = 2400mhz
Ram @ 240mhz 2.5-3-3-10
Aquamark3 Total 60,876 Gfx 8418 CPU 11149
Super PI 1mb=34seconds, 512k=15seconds
3dmark05 3060
On my rig it seems memory bandwith is the king. At least for these three benchies.
1. There is no 9.5 multiplier... (So your BIOS uses 9 or 10 and a higher or lower FSB=HTT to compensate) Use CPU-Z to tell us your REAL operating frequencies including the memory frequencies because they are being MISREPORTED in BIOS...
2. If what you post is correct, you have 40% more memory bandwidth and 6% less CPU frequency on the second scenario giving you 2,5% more performance...
3. Why on earth do you keep your mem@173Mhz, go beyond 200Mhz (DDR400) under ANY scenario... (even if you need to relax the timings)
4. Only when you know how to (methodology) compare apples to apples can you reach definite results... So unless you do a proper comparison bandwidth is not king, at least in the sense you put it here...
GuitarDaddy
Lifer
- Nov 9, 2004
- 11,465
- 1
- 0
1. CPU-Z and AI booster both show the 9.5 multi. All of the settings I listed were confirmed with CPU-Z and AI booster and nTune. So if there is no 9.5 multi all of these utilities are fooled in exactly the same way. (I have seen the comparisons showing a performance hit using the 1/2 dividers) Thanks Zebo!
2. My post is correct, and I agree with your calculations.
3. This board doesn't offer manual memory dividers(bummer!!). You can run 1:1 or you can run auto and it applies the dividers automatically and always keeps your memory <= 200mhz
4. What comparison would you suggest? Since you obviously "know how"
The tone of your post indicates that these results disturb you. I myself am disappointed in the overclocking limitations on the A8N-SLI. But overall I love the features, stability, and performance of this board. I realize that to a lot of people CPU mhz bragging rights are important. Personally I don't care wether I run 2ghz or 3ghz, I just want to find the best performance I can get with this particular setup.
I will add that the lack of manual memory dividers on this board is a definate overclocking limitation. If your desire is to reach a maxium CPU overclock and overclock your ram you need to choose another board over the A8N-SLI at this time. I am hoping a future bios revision will address this limitation. I am interested to see if this is a limitation of just this board or all nforce4 boards.
- Oct 3, 2004
- 5,761
- 12
- 81
i would definetely go with the 90 nanometer processor due to low temps and usually better OCing. A good looking motherboard for AMDs 939 is http://www.newegg.com/app/view...=13-152-049&depa=0
I need more info on this. Do you have the latest CPU-Z (I think it is 1.26)??? In theory there are NO HALF dividers supported by AMD A64 Proccessors!!!, so whatever the BIOS and the utilities report there is no such thing... Since you have a very new board we must investigate this!
To check the latency issue in dual compaprisons:
1. Same CPU freq, Same memory freq, different latencies eg: 2-2-5-2 vs 3-3-10-3
2. Higher or lower same CPU freq, Same memory divider as before, different latencies eg: 2-2-5-2 vs 3-3-10-3
The more of these tests the better: This way you identify the latency performance impact and how it scales when frequency scales...
To check the bandwidth issue in dual compaprisons:
1. Same CPU freq, same latencies, different memory bandwidth (different dividers) eg: 200 vs 166
2. Higher or lower same CPU freq, same latencies as before, different memory bandwidth (different dividers) eg: 200 vs 166
The more of these tests the better: This way you identify the mem bandwidth performance impact and how it scales when frequency scales...
This is the general idea, if we assume that 1:1 and ASYNC mem do not incur any serious differences in performace (tested many times by others) and that HT frequency performace hit is negligible (which is valid). Of course there are many more combinations to be tested (mixing the above ones), but most of these will not alter the results significantly...
I do not have this board and I can not comment on what is there and what is not... Maybe if you send me some more detailed info on your BIOS settings I could help you find what is supported and what is not...
PS1 From your previous results we can assume that the mem bandwidth increase in percentage yields between 1/6th and 1/7th in general performance benefit of the the same percenatge increase in CPU frequency (which is a lot previous reports speak of 1/10th). But we can not conclude anything on latencies...(my experience through various comparative benchmarks is 5% maximum general performance benefit of 2-2-5-2 vs 3-3-10-3). To sum it up ignore my tone
and provide more rigorous testing...PS2 From all the above it is obvious that you can run your system with much higher performance than what you have posted and the only PROBABLE limiting factor is the lack!?! of manual memory dividers in your board...
Maybe you guys are talking about diferente things. One talking about the A64 haven't half-multipliers FROM FACTORY, they are all 9, 10, 11, etc... and the other is talking about what he choose while overclocking ?!?!?
My guess...
GuitarDaddy
Lifer
- Nov 9, 2004
- 11,465
- 1
- 0
I am using CPU-Z 1.26. And I think we are talking about the same thing. I was surprized when I saw the 1/2 multi's on this board. Because I have read several post's regarding the lack of 1/2 multi's on A64 chips. And read one article at "Tom's Hardware" I think? where they claimed to prove that A64's didn't support 1/2 multi's and said they had confirmed with an AMD rep. And that the bios actually used a whole multi instead of the 1/2 multi and adjust's the FSB accordingly.
I have also seen an article claiming to debunk this theory? I don't know?
What I do know is that CPU-Z 1.26 and the utilities provided by asus show the 1/2 multi's,
and that I can get a higher FSB with a 9.5 multi than I can with 10.0,
and I can get a higher FSB with a 9.0 than I can with 9.5 according to these utilities.
I don't have a clue how you would prove it one way or the other. Are there any other utilities that I can check the the CPU clock and memory settings with ?
The plot thickens, thanx to GuitarDaddy a much more complex picture surfaces:
An article in Sudhian sparkled the whole mess BUT contrary to what they write, we know less at this point than we did before...
We are dealing with:
1. Undocumented features by AMD (functional A64 half multipliers)
2. BIOSes like the MSI K8N Neo2, which misreport what they actually do: If I understand correctly according to the last Sudhian article MSI does rounding instead of using half multipliers, so in this case the lack of half multipliers is still valid, although for a different reason...
3. Diagnostic software that reports the BIOS changes and not the actual situation on the hardware level...
The above three create a gigantic mess
MUCH MORE ACCURATE AND UPDATED INFO on my later post. Keep reading
An article in Sudhian sparkled the whole mess BUT contrary to what they write, we know less at this point than we did before...
We are dealing with:
1. Undocumented features by AMD (functional A64 half multipliers)
2. BIOSes like the MSI K8N Neo2, which misreport what they actually do: If I understand correctly according to the last Sudhian article MSI does rounding instead of using half multipliers, so in this case the lack of half multipliers is still valid, although for a different reason...
3. Diagnostic software that reports the BIOS changes and not the actual situation on the hardware level...
The above three create a gigantic mess
MUCH MORE ACCURATE AND UPDATED INFO on my later post. Keep reading
Zebo
Elite Member
- Jul 29, 2001
- 39,398
- 19
- 81
When using a half multiplier you are loosing bandwidth because it puts an additional wait state ras to cas and mem knocks down to the next whole multiplier dispite what you bios says. I think this is clearly illustrated here.
http://www20.tomshardware.com/...19/cpu-scaling-05.html
There's a reason AMD only sells these in whole multipliers.
http://www20.tomshardware.com/...19/cpu-scaling-05.html
There's a reason AMD only sells these in whole multipliers.
That was my point in the post above.
Althought AMD only sell cpu's with full multipliers, we can choose to run half-dividers in BIOS ( or with a64 tweaker ) and it still works correctely.
All the tools we have, all say that it is running half-divisors and the benchmarks put them half-way the two near ones.
This is weird if they say that half-multipliers do not existe...
I could check it out more later at home, putting again a 8,5x on my 3000+
Althought AMD only sell cpu's with full multipliers, we can choose to run half-dividers in BIOS ( or with a64 tweaker ) and it still works correctely.
All the tools we have, all say that it is running half-divisors and the benchmarks put them half-way the two near ones.
This is weird if they say that half-multipliers do not existe...
I could check it out more later at home, putting again a 8,5x on my 3000+
Ok, i've read sudhian article, and it makes some good points.
The one I don't get it is how is mobo got so messy on the vdimm voltage output, while there are many hard overclockers out there that monitor everything to the last detail, and haven't seen any of them talking about that problem.
I'll be getting tomorrow a OCZ Booster for my BH-5, since the trick to raise the vdimm on the MSI neo2 platinum envolves some bridge cutting, and them i'll make some testing with the voltimeter.
The one I don't get it is how is mobo got so messy on the vdimm voltage output, while there are many hard overclockers out there that monitor everything to the last detail, and haven't seen any of them talking about that problem.
I'll be getting tomorrow a OCZ Booster for my BH-5, since the trick to raise the vdimm on the MSI neo2 platinum envolves some bridge cutting, and them i'll make some testing with the voltimeter.
HALF A64 CPU MULTIPLIERS (8.5x,9.5x etc) DO EXIST AND ARE FUNCTIONAL EVEN IF UNDOCUMENTED BY AMD.
There is no diagnostic application (including CPU-Z, Core Center, RM-Clock etc) at the moment that reports CORRECTLY the MEMORY DIVIDERS (200, 166, 133, etc) that result from the use of HALF CPU MULTIPLIERS. OCA64 is going to become the first one to do such calculations providing THE REAL Operating Memory Frequencies...
PS1 MSI Core Center does not even recognize HALF CPU multipliers, let alone calculate the resulting MEMORY DIVIDERS misreporting everything...
PS2 CPU-Z and ClockGen misreport the MEMORY DIVIDER and as a result they misreport the OPERATING MEMORY FREQUENCY of an A64 running with a HALF CPU MULTIPLIER...
PS3 Although there are HALF CPU MULTIPLIERS there are no HALF MEMORY DIVIDERS
Zebo I strongly suspect that Toms Hardware was using CPU-Z or a similar tool and failed to identify that the reason for the reduced memory bandwidth observed was the even lower memory frequency that HALF CPU MULTIPLIERS incur due to the MEMORY DIVIDER calculation methodology.
EXAMPLE:
8.5x CPU multi, 1:1 Memory, 248Mhz FSB=HTT, Memory Freq = 234Mhz (divider 9)
Mem Freq is NOT 248 reported by CPU-Z and Clockgen (divider 8.5)
I am 100% positive about this because my memory maxes at 233-234 (3D mark 2003 crash)...
Stay tuned for OCA64 v1.2
There is no diagnostic application (including CPU-Z, Core Center, RM-Clock etc) at the moment that reports CORRECTLY the MEMORY DIVIDERS (200, 166, 133, etc) that result from the use of HALF CPU MULTIPLIERS. OCA64 is going to become the first one to do such calculations providing THE REAL Operating Memory Frequencies...
PS1 MSI Core Center does not even recognize HALF CPU multipliers, let alone calculate the resulting MEMORY DIVIDERS misreporting everything...
PS2 CPU-Z and ClockGen misreport the MEMORY DIVIDER and as a result they misreport the OPERATING MEMORY FREQUENCY of an A64 running with a HALF CPU MULTIPLIER...
PS3 Although there are HALF CPU MULTIPLIERS there are no HALF MEMORY DIVIDERS
Zebo I strongly suspect that Toms Hardware was using CPU-Z or a similar tool and failed to identify that the reason for the reduced memory bandwidth observed was the even lower memory frequency that HALF CPU MULTIPLIERS incur due to the MEMORY DIVIDER calculation methodology.
EXAMPLE:
8.5x CPU multi, 1:1 Memory, 248Mhz FSB=HTT, Memory Freq = 234Mhz (divider 9)
Mem Freq is NOT 248 reported by CPU-Z and Clockgen (divider 8.5)
I am 100% positive about this because my memory maxes at 233-234 (3D mark 2003 crash)...
Stay tuned for OCA64 v1.2
StriderGT
I have DDR500(PC4000 HyperX) and haven't been able to realize anything over 220MHz fsb. I was hoping that I could run the system at 250MHz FSB with 1:1 ratio.
I have DDR500(PC4000 HyperX) and haven't been able to realize anything over 220MHz fsb. I was hoping that I could run the system at 250MHz FSB with 1:1 ratio.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 21K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter