 10/07/2003 05:27 PM
|
edplayer Junior Member
Posts: 12
Joined: 10/07/2003
|
Whats the exact numbers on the Athlons bandwidth? 133FSB is advertised as 2100MB/sec 166FSB is 2700MB/sec and 200FSB is 3200MB/sec. But they are slightly off it seems cause it goes up 33MHz each time but jumps 600MB/sec and 500MB/sec. I want to figure out how much 220FSB would yield exactly.
|
|
|
|
10/07/2003 07:35 PM
|
xUNREALNEOx Member
Posts: 149
Joined: 10/06/2003
|
it's the MHz X fsb width (16 bit) so 220 x 16 = 3520 nForce 2's have about 90% memory effciency so 3520 X 0.9 = 3168 That's just about right for what my system puts out. remember, lower latency = more efficiency
-------------------------
14,434 9600Pro' ">http://wijzr.url.cjb.net/ 12,415 Ti200' ">http://batey.url.cjb.net/ 4,029 9600Pro' ">http://mgzue.url.cjb.net/.:Team cpucity:.' ">http://www.cpucity.co.uk :: MadOnion
|
|
|
|
10/07/2003 09:14 PM
|
edplayer Junior Member
Posts: 12
Joined: 10/07/2003
|
thanks xUNREALNEOx
Where did you learn about the NForce 2's efficiency? I recall reading some articles on memory efficiency at Vans Hardware but they were from a while back (comparing Athlon memory system to the P4s Rambus).
|
|
|
|
10/08/2003 03:14 AM
|
Brian Member
Posts: 98
Joined: 10/06/2003
|
actually not on Dual Channel systems, the memory bus width is 128bit on normal single channel DDR, it's 64-bit divide the bits by 8 to get bytes so 128/8=16 bytes and multiply it by the speed/frequency so, 16 bytes x 220MHz and you get 3520MB/sec
-------------------------
Folding For Team AnandTech Distributed Computing' ">http://forums.anandtech.com/categories.cfm?catid=39 My Humble Forums' ">http://forums.hellstormraiders.com
|
|
|
|
10/08/2003 03:53 AM
|
Greyhound Senior Member
Posts: 5449
Joined: 10/07/2003
|
Memory Bandwidth(in MB/s) = physical clockspeed(in MHz) x bus width(in bytes - 1 byte = 8 bits) x packets of data transferred per clock cycle(2 for DDR)
Example: PC3200 DDR-SDRAM: 200MHz physical clock x 8(64 bit = 8 byte bus) x 2(DDR) = 3200MB/s
if it's Dual-channel, the bus width effectively doubles since you have 2x64bit(NForce2) or 1x128bit(A64 FX/Opteron), so it's 6400MB/s
-------------------------
The opinions expressed above do not represent those of Advanced Micro Devices or any of their affiliates.
<center><font color=red>MODERATOR</center></font>
AMD Athlon 64 X2 6000+ 89W :: Asus M2N32SLI Deluxe :: 2x1024MB Muskin XP2-6400 DDR2-800 :: BeQuiet! Darkpower Pro 530W :: 2x WD Raptor 74GB/10k rpm, RAID 0 :: ATI Radeon X1950XTX
|
|
|
|
10/08/2003 03:58 AM
|
Dagalidis Senior Member
Posts: 243
Joined: 10/07/2003
|
Here is the ANSWER for you....
As you SEE at DUAL CHANNEL DDR SYNCH the Bandwidth Eficciency is 95% for me and some times can Rech up to 97%....
-------------------------
ATHLON64 3200+@2380(238X10) 1.644 Vcore Cooling By AMD Stock Heatsink ASUS K8V Deluxe (VIA K8T800 Chipset) 256x2 TWINMOS PC3700 5-3-2-2.5 at 2.8V Sound Blaster AUDIGY 2 Sound Blaster Megaworks Speakers 2.1 250D THX / 300 WRMS LEADTEK FX5900 128MB My ViVo
<span style='co
|
|
|
|
10/08/2003 04:04 AM
|
Brian Member
Posts: 98
Joined: 10/06/2003
|
yeah, that's what i do calculate mem bandwidth then use sandra to check
-------------------------
Folding For Team AnandTech Distributed Computing' ">http://forums.anandtech.com/categories.cfm?catid=39 My Humble Forums' ">http://forums.hellstormraiders.com
|
|
|
|
10/08/2003 08:07 PM
|
carltao Member
Posts: 35
Joined: 10/07/2003
|
QUOTE (Hellraiser @ Oct 7 2003, 11:53 PM) Memory Bandwidth(in MB/s) = physical clockspeed(in MHz) x bus width(in bytes - 1 byte = 8 bits) x packets of data transferred per clock cycle(2 for DDR)
Example: PC3200 DDR-SDRAM: 200MHz physical clock x 8(64 bit = 8 byte bus) x 2(DDR) = 3200MB/s
if it's Dual-channel, the bus width effectively doubles since you have 2x64bit(NForce2) or 1x128bit(A64 FX/Opteron), so it's 6400MB/s
Correct.
But actually in the AthlonXP(200MFSB)+DualChannelDDR400 configuration, CPU PSB(FSBx2, i.e. 400MHz=200MHzx2) limits the theoretical 6.4GB/s memory bandwidth to 3.2GB/s of CPU data pump(between north bridge). So the nForce2's memory efficiency is really very high.
Here are some interesting comparisons of different memory configurations:
Note: Red: Single Channel Cyan: Dual Channel
DDR200 parameters 2-2-2-3; DDR400 parameters 2-2-3-5
There is still a question that I don't have an answer: What does L2 cache contribute to actual MEM bandwidth?
-------------------------
AthlonXP1800+ 200x10@1.400V 200x11@1.575V 200x11.5@1.650V
AVC 112C86FBH
Huntkey(Maybe the best power brand of China) Rock-355
EPoX 8RDA3+ Rev 2.0
ATi Radeon9000(Not gamer ^_^)
TwinMOS CSP DDR400 256Mx2 2-2-3-11@2.77V
WD1
|
|
|
|
10/08/2003 08:19 PM
|
jonspd Member
Posts: 55
Joined: 10/07/2003
|
I always just used sandra but I think I will give this calculation a try.
-------------------------
XP2500-m------------------------A643000NC{No IHS}
NF7-S-----------------------------8KDA3J
KHX3500x2----------------------Mushkin3200x2
BFG 5900-------------------------EVGA6800
530 forton------------------------530 forton
SLK 800---------------------------SLKU948
|
|
|
|
10/09/2003 02:14 AM
|
Greyhound Senior Member
Posts: 5449
Joined: 10/07/2003
|
yeah, memory bandwidth can never exceed the FSB's bandwidth(if there IS a FSB, hehe). it can't be equal to the FSB's bandwidth either, because the northbridge does not only connect to the memory, but also AGP and southbridge(=>I/O).
The reason why the NForce2-chipsets' memory efficiency is so high is because it provides MUCH more bandwidth than the FSB could ever handle...
-------------------------
The opinions expressed above do not represent those of Advanced Micro Devices or any of their affiliates.
<center><font color=red>MODERATOR</center></font>
AMD Athlon 64 X2 6000+ 89W :: Asus M2N32SLI Deluxe :: 2x1024MB Muskin XP2-6400 DDR2-800 :: BeQuiet! Darkpower Pro 530W :: 2x WD Raptor 74GB/10k rpm, RAID 0 :: ATI Radeon X1950XTX
|
|
|
|
10/09/2003 07:37 AM
|
carltao Member
Posts: 35
Joined: 10/07/2003
|
QUOTE (Hellraiser @ Oct 8 2003, 10:14 PM) yeah, memory bandwidth can never exceed the FSB's bandwidth(if there IS a FSB, hehe). it can't be equal to the FSB's bandwidth either, because the northbridge does not only connect to the memory, but also AGP and southbridge(=>I/O).
The reason why the NForce2-chipsets' memory efficiency is so high is because it provides MUCH more bandwidth than the FSB could ever handle...
Single channel memory on nForce2 still can reach ~2.9GB/s bandwidth. See the above diagram.
So I think what makes nforce2's high memory efficiency is nVidia's DASP memory technology.
Dual channel does not help much except for the IGP's integrated GPU. Actually it is only for those who seek top performance.
-------------------------
AthlonXP1800+ 200x10@1.400V 200x11@1.575V 200x11.5@1.650V
AVC 112C86FBH
Huntkey(Maybe the best power brand of China) Rock-355
EPoX 8RDA3+ Rev 2.0
ATi Radeon9000(Not gamer ^_^)
TwinMOS CSP DDR400 256Mx2 2-2-3-11@2.77V
WD1
|
|
|
|
10/09/2003 11:10 AM
|
xUNREALNEOx Member
Posts: 149
Joined: 10/06/2003
|
QUOTE (carltao @ Oct 9 2003, 03:37 AM) QUOTE (Hellraiser @ Oct 8 2003, 10:14 PM) yeah, memory bandwidth can never exceed the FSB's bandwidth(if there IS a FSB, hehe). it can't be equal to the FSB's bandwidth either, because the northbridge does not only connect to the memory, but also AGP and southbridge(=>I/O). The reason why the NForce2-chipsets' memory efficiency is so high is because it provides MUCH more bandwidth than the FSB could ever handle... Single channel memory on nForce2 still can reach ~2.9GB/s bandwidth. See the above diagram. So I think what makes nforce2's high memory efficiency is nVidia's DASP memory technology. Dual channel does not help much except for the IGP's integrated GPU. Actually it is only for those who seek top performance. I can reach 3060 SC, 400fsb, 2-3-3-6. yeah, I forgot about DC, that is more efficient.
-------------------------
14,434 9600Pro' ">http://wijzr.url.cjb.net/ 12,415 Ti200' ">http://batey.url.cjb.net/ 4,029 9600Pro' ">http://mgzue.url.cjb.net/.:Team cpucity:.' ">http://www.cpucity.co.uk :: MadOnion
|
|
|
|
10/09/2003 08:09 PM
|
carltao Member
Posts: 35
Joined: 10/07/2003
|
QUOTE (xUNREALNEOx @ Oct 9 2003, 07:10 AM) QUOTE (carltao @ Oct 9 2003, 03:37 AM) QUOTE (Hellraiser @ Oct 8 2003, 10:14 PM) yeah, memory bandwidth can never exceed the FSB's bandwidth(if there IS a FSB, hehe). it can't be equal to the FSB's bandwidth either, because the northbridge does not only connect to the memory, but also AGP and southbridge(=>I/O).
The reason why the NForce2-chipsets' memory efficiency is so high is because it provides MUCH more bandwidth than the FSB could ever handle...
Single channel memory on nForce2 still can reach ~2.9GB/s bandwidth. See the above diagram.
So I think what makes nforce2's high memory efficiency is nVidia's DASP memory technology.
Dual channel does not help much except for the IGP's integrated GPU. Actually it is only for those who seek top performance.
I can reach 3060 SC, 400fsb, 2-3-3-6.
yeah, I forgot about DC, that is more efficient.
a screen copy is great enough...
-------------------------
AthlonXP1800+ 200x10@1.400V 200x11@1.575V 200x11.5@1.650V
AVC 112C86FBH
Huntkey(Maybe the best power brand of China) Rock-355
EPoX 8RDA3+ Rev 2.0
ATi Radeon9000(Not gamer ^_^)
TwinMOS CSP DDR400 256Mx2 2-2-3-11@2.77V
WD1
|
|
|
Search Topic
