- Joined
- Jul 26, 2004
Just like the title states PPl seam to be getting very nice OCs with these new brisbane's but when ever I see the Pi times im like hmm that doesnt seam right my 939 @ lower clocks is beating it . why is this ?
-
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
Why are AM2 slower in Pi @ the same clock vs 939 ?
- Thread starter ||Console||
- Start date
- Joined
- Jun 16, 2006
Im probly wrong but let me give awnsering this a try.
For super pi memory timeing is a big facter in the outcome. so DDR2 with looser timeings will score less than a 939 rig useing DDR with tighter timeings.
For super pi memory timeing is a big facter in the outcome. so DDR2 with looser timeings will score less than a 939 rig useing DDR with tighter timeings.
- Joined
- Jun 16, 2006
Internet messed up so im sorry about the DBL post. A mod can go ahead and delete this post.
OP
- Joined
- Jul 26, 2004
- Thread Starter
- #4
I get that about the timings but ddr2 is running @ a higher freq so the higher timmigns shouldnt matter that much , afaik
OP
- Joined
- Jul 26, 2004
- Thread Starter
- #6
i dont know if that is right or not I have never used a AM2, but i have seen bw tests in Sandra and the ddr2 always has more bw than the ddr1 .ryboto said:
this is the main thing that buggs me about am2's Why I would take that upgrade path for somthing that is slower .
You need a pretty big speed bump if you're going to compensate for increased latencies.
There are also some design issues with DDR2 as compared to DDR1. Can't remember where I read about that first, but that also had to do with latencies and with being unable to read very small portions of data - it's always four sequential reads whether you need them or not. At least that's what I think I remember.
There are also some design issues with DDR2 as compared to DDR1. Can't remember where I read about that first, but that also had to do with latencies and with being unable to read very small portions of data - it's always four sequential reads whether you need them or not. At least that's what I think I remember.
- Joined
- Apr 6, 2002
I think that since the jump to AM2 was primarily only to add support for DDR2, they didn't really make any significant changes to a) the processors and b) the motherboards. I think that when AMD finally has an architecture change, you will start to see those times drop to rival the sub 10s times at the top of the Intel rankings.
OP
- Joined
- Jul 26, 2004
- Thread Starter
- #9
I hope sosplat said:
I just dont know why amd would release am2 just for ddr2 even though it is slower than the old line (939)
- Joined
- Jun 16, 2001
SuperPi is an interesting benchmark in that its performance depends on a few specific things like memory and cache latency, and other architecture specific quirks.
Latency is very different from bandwith, latency measures how fast data can move from one place to another whereas bandwith measures how much.
Latency would be how long it takes you to get from A to B on a specified road, while bandwith is concerned with how many cars you can fit on said road. You can't really compensate latency with bandwith because they are two entirely different things.
SuperPi only cares about getting 3 dudes to Oakland in an hour, it doesn't help if you can get 5 dudes to Oakland in two.
Also, the way the A64 memory controller is designed, the CPU pretty much has all the bandwith it needs.
SuperPi imho is a fun competitive benchmark, but imho is a poor indicator of overall performance. It seems to favor certain CPU architectures over others more than other benchmarks. For example my C2D laptop running at 2Ghz beats my X2 in 1M Spi times by like 5 seconds, but takes 10-11 minutes more to do a frame of 2652 WinSMP folding, and is significantly slower in other benchmarks I have tried.
Latency is very different from bandwith, latency measures how fast data can move from one place to another whereas bandwith measures how much.
Latency would be how long it takes you to get from A to B on a specified road, while bandwith is concerned with how many cars you can fit on said road. You can't really compensate latency with bandwith because they are two entirely different things.
SuperPi only cares about getting 3 dudes to Oakland in an hour, it doesn't help if you can get 5 dudes to Oakland in two.
Also, the way the A64 memory controller is designed, the CPU pretty much has all the bandwith it needs.
SuperPi imho is a fun competitive benchmark, but imho is a poor indicator of overall performance. It seems to favor certain CPU architectures over others more than other benchmarks. For example my C2D laptop running at 2Ghz beats my X2 in 1M Spi times by like 5 seconds, but takes 10-11 minutes more to do a frame of 2652 WinSMP folding, and is significantly slower in other benchmarks I have tried.
- Joined
- Jan 23, 2005
- Location
- Keystone Lightville, MA
tweaking ddr2 is also very hard as well to get low latency. and most people who are benching spi with these 3600+'s, for lack of better words, arent benchers. they are average everyday ocer's who dont know how to optimize like you or others do.
ive seen some times from a sempron am2 that would make your 4400+ blush, its all about tweaking.
ive seen some times from a sempron am2 that would make your 4400+ blush, its all about tweaking.
OP
- Joined
- Jul 26, 2004
- Thread Starter
- #13
Can I see some links please Woudl liek to check the difference with there sub timmings vs some others .CCUABIDExORxDIE said:
my best time on my 4400+ is 28.4 I think that was @ 2.95ghz
- Joined
- Oct 2, 2001
- Location
- Corona, CA
console you have to remember, you have 1meg L2 cache vs 512KB on the Brisbane. Also the move to DDR2 cause a bit more latency to the cpu cache. Also super-PI loves tight timmings.||Console|| said:
- Joined
- Jan 23, 2005
- Location
- Keystone Lightville, MA
my friend got his sempron right around mid to low 28's at 2.95ghz. and thats with only 256k cache. its all about having ram that can run the right speeds. tuning tuning tuning. most of these brisbane systems we are seeing are using value ram, nothing is really done with really good fast ram (ddr2-1000 4-4-3)
- Joined
- Jan 23, 2005
- Location
- Keystone Lightville, MA
seems alot better then s939
- Joined
- Jan 23, 2005
- Location
- Keystone Lightville, MA
ddr2-1200 4-5-5 isnt pushing it
OP
- Joined
- Jul 26, 2004
- Thread Starter
- #19
CCUABIDE: Thx that is what I was looking for =)
- Joined
- May 11, 2004
- Location
- Montana tech Butte mt.
Most Am2 CPUs are only 512kb cache which is the biggest reason that you see a difference especially when compared to 1meg cache 939s.
As far as ram goes I happily trade latency for speed. With AM2 1t and 2t are about .1-.2s apart but the difference in being able to run my ram at over 500 MHZ 4,4,4 spanks anything I can do on a lower divider and 3,3,3. If my motherboard(maybe its my CPU) wasn't so stupid I think I could pull off close to 600mhz 4,4,4 which trust me no set of BH-5 or TCCD can touch in Superpi.
As far as speed bumps for each up in latency roughly 50mhz on ram speed is the difference between 3,3,3 and 4,4,4. So 1 divider and you covered the difference.
As far as ram goes I happily trade latency for speed. With AM2 1t and 2t are about .1-.2s apart but the difference in being able to run my ram at over 500 MHZ 4,4,4 spanks anything I can do on a lower divider and 3,3,3. If my motherboard(maybe its my CPU) wasn't so stupid I think I could pull off close to 600mhz 4,4,4 which trust me no set of BH-5 or TCCD can touch in Superpi.
As far as speed bumps for each up in latency roughly 50mhz on ram speed is the difference between 3,3,3 and 4,4,4. So 1 divider and you covered the difference.
Similar threads
