800MHz FSB confuses me a bit..

pelleplu

Member
May 23, 2003
127
0
0
First off, let me see if I've got this straight:
The FSB is the speed between the CPU and the Chipset, and the Memory bus speed is between the chipset and the memory?
I've read here in the forums that it's best to keep the FSB and Memory speed the same, or at least close. But on a P4 with 800MHz FSB and DDR400 (200MHz Mem speed, right?), there is a huge difference (600MHz) between the FSB and mem speed. Can someone explain how this works?
 

pspada

Platinum Member
Dec 23, 2002
2,503
0
0
Nope, the actual speed of the memory is 200Mhz (PC3200), which the Athlon XP uses on both the up and down sides of the clockcycle (that's why it's called DDR (Double Data Rate), for a total of 400Mhz throughput. In the case of the 800Mhz Intel chip, it uses the same 200Mhz DDR memory, but basically accesses it 4 times during the clockcycle, what they call "quad pumped", for a total of 800Mhz.

And the FSB is the Front Side Bus, the speed at which the processor accesses the memory via (pun intended) the chipset.
 

redhatlinux

Senior member
Oct 6, 2001
493
0
0
Pelleplu, you got it right. The FSB is between the CPU and Chipset. The Northbridge has its own memory bus, its not the FSB. Data from memory uses the memory bus, to the memory controller, part of the Northbridge. The data are then passed on to the FSB to the cpu and cache. Note that Intel uses a quad-pumped FSB BUT the memory bus is dual-pumped (DDR). The issue of sync/async is really an issue of bandwidth. Athlon TBIRDS using 100mhz FSB (200 DDR) would show benefits from running the memory at 133 async because the FSB had the bandwidth to move the data. (Main memory ran SDR at the time). With the advent of DDR memory, effectively keeping pace with the FSB, running async has little if any benefit.
 

pspada

Platinum Member
Dec 23, 2002
2,503
0
0
I was under the impression that the access from the CPU to the memory and back was provided by a bridge (as in "Northbridge"), whereas the southbridge provides access from the CPU to the PCI and AGP buses and back. And the reason for 2 different bridges is so that the north is not limited by the slower speed of the south. I mean, is not the whole point of the bridges to provide access to devices external to the CPU itself, as opposed to simply communicating back and forth to the CPU? And then the FSB is the speed that the CPU has to communicate with the memory at, via the northbridge?

EDIT: You are correct, the FSB is only between the northbridge and CPU.
 

Spicedaddy

Platinum Member
Apr 18, 2002
2,305
77
91
But on a P4 with 800MHz FSB and DDR400 (200MHz Mem speed, right?), there is a huge difference (600MHz) between the FSB and mem speed. Can someone explain how this works?

P4-C is 200MHz Quad-Pumped = 800MHz FSB

The newer chipsets supporting them use dual-channel PC3200 DDR, so 200MHz DDR = 400MHz, then x2 since it's dual channel = 800MHz
 

pspada

Platinum Member
Dec 23, 2002
2,503
0
0
Originally posted by: Spicedaddy
But on a P4 with 800MHz FSB and DDR400 (200MHz Mem speed, right?), there is a huge difference (600MHz) between the FSB and mem speed. Can someone explain how this works?

P4-C is 200MHz Quad-Pumped = 800MHz FSB

The newer chipsets supporting them use dual-channel PC3200 DDR, so 200MHz DDR = 400MHz, then x2 since it's dual channel = 800MHz

So would this not also apply to the AMD NForce 2 motherboards, as they are dual channel? FSB x DDR X DC for 200 x 2 x 2 = 800Mhz?

 

Jeff7181

Lifer
Aug 21, 2002
18,368
11
81
One thing that people have neglected to clarify...

The front side bus of an Intel P4 is not actually 800 Mhz. It's 200... but because it's quad pumped, it's effectively 800 Mhz. PC3200 RAM also runs at 200 Mhz, but it's only double pumped, so effectively it's 400 Mhz. This would be a problem for Intel if they hadn't incorporated dual channel memory controllers like nVidia did with the nForce2 chipset. Now, even though PC3200 is 400 Mhz effectively, there's 2 channels, so there's 800 Mhz worth of bandwidth now, which is what P4's like, lots of memory bandwidth.
If you look back at the old 533 Mhz P4's, their FSB is 133 Mhz quad pumped. But they didn't utilize dual memory controllers... so even if you bought PC3200 RAM that's effectively 400 Mhz, your FSB is still fast than your memory, so it's constantly waiting for things that are in RAM. That is why increasing the memory bandwidth of a 533 Mhz bus P4 (by overclocking the RAM) resulted in such large increases in performance, the memory was always lagging behind, slowing everything down.
To put that in perspective for AMD users... that would be like using an XP2800 (333 Mhz front side bus) with PC2100 RAM, which runs at 266 Mhz effectively.
So... all those people who were so delighted that their P4 gained so much performance by overclocking sometimes didn't realize that the reason they gained so much more performance is because their P4 was handicapped to start with. Intel had hoped to use RDRAM, which would provide the memory bandwidth that the P4's needed, but it looks as though consumers rejected it because of it's cost.
</lecture> =)
 

Jeff7181

Lifer
Aug 21, 2002
18,368
11
81
Originally posted by: pspada
Originally posted by: Spicedaddy
But on a P4 with 800MHz FSB and DDR400 (200MHz Mem speed, right?), there is a huge difference (600MHz) between the FSB and mem speed. Can someone explain how this works?

P4-C is 200MHz Quad-Pumped = 800MHz FSB

The newer chipsets supporting them use dual-channel PC3200 DDR, so 200MHz DDR = 400MHz, then x2 since it's dual channel = 800MHz

So would this not also apply to the AMD NForce 2 motherboards, as they are dual channel? FSB x DDR X DC for 200 x 2 x 2 = 800Mhz?

The same numbers don't apply... but AMD processors work similar... for example...

Athlon XP2500 uses a 166 Mhz bus double pumped, so effectively it's 333 Mhz. PC2700 RAM is 166 Mhz, double data rate, so it's effectively 333 Mhz also. The RAM and FSB would be running in sync... that's good, very good.
But as you said, nForce2 motherboards are also dual channel boards... so... you have two channels that are capable of 333 Mhz... so effectively you have 666 Mhz worth of bandwidth. Hmmmmm... that brings up an interesting topic...

If with a dual channel board, you have twice as much memory bandwidth as an Athlon XP needs... could you cut the speed of the RAM in order to use lower timings and not loose a whole lot of performance? I may have to test that and see exactly what the results are =)
 

Spicedaddy

Platinum Member
Apr 18, 2002
2,305
77
91
Originally posted by: pspada
Originally posted by: Spicedaddy
But on a P4 with 800MHz FSB and DDR400 (200MHz Mem speed, right?), there is a huge difference (600MHz) between the FSB and mem speed. Can someone explain how this works?

P4-C is 200MHz Quad-Pumped = 800MHz FSB

The newer chipsets supporting them use dual-channel PC3200 DDR, so 200MHz DDR = 400MHz, then x2 since it's dual channel = 800MHz

So would this not also apply to the AMD NForce 2 motherboards, as they are dual channel? FSB x DDR X DC for 200 x 2 x 2 = 800Mhz?

Yep, except the Athlon can't use all that bandwidth. nVidia implemented dual-channel because it can be used by their integrated video.
 

Jeff7181

Lifer
Aug 21, 2002
18,368
11
81
Originally posted by: Spicedaddy
Originally posted by: pspada
Originally posted by: Spicedaddy
But on a P4 with 800MHz FSB and DDR400 (200MHz Mem speed, right?), there is a huge difference (600MHz) between the FSB and mem speed. Can someone explain how this works?

P4-C is 200MHz Quad-Pumped = 800MHz FSB

The newer chipsets supporting them use dual-channel PC3200 DDR, so 200MHz DDR = 400MHz, then x2 since it's dual channel = 800MHz

So would this not also apply to the AMD NForce 2 motherboards, as they are dual channel? FSB x DDR X DC for 200 x 2 x 2 = 800Mhz?

Yep, except the Athlon can't use all that bandwidth. nVidia implemented dual-channel because it can be used by their integrated video.

I prefer the phrase... "doesn't need" rather than "can't use"... =) It implies that it would benefit from it if it could use it... it simply doesn't need it.
 

zephyrprime

Diamond Member
Feb 18, 2001
7,512
2
81
The front side bus of an Intel P4 is not actually 800 Mhz. It's 200... but because it's quad pumped, it's effectively 800 Mhz.
I don't really like it when people talk about multi-pumped busses like this. Is this way of speaking of busses really correct? It seems to me that although the clock is only 200mhz, the data is truely 800mhz and the fact that's it's quadpumped or whatever is irelevant to everyone except the EEs that have to design the bus.
 

Jeff7181

Lifer
Aug 21, 2002
18,368
11
81
It confuses people who don't know about multi-pumping because then the math doesn't work out right...

"800 Mhz FSB times a multiplier of 12... WOW!!! I have a 9.6 Ghz processor!!!"