So my first question is, would a quad pumped FSB be "2x" better than a dual pumped FSB? I see that that AMD took the path of Dual x FSB and Intel took the path of Quad x FSB. However, when you compare the 1x FSB, most of the Athlons are 133MHz and the new ones are getting to be 166MHz, and now it is said that in May, the 200MHz FSB Athlons will be launched. The "400MHz" FSB P4's are just 100MHz x 4 and the 533's are just 133 x 4. So I'm starting to get confused with that.. like would a 200MHz x 2 be "better" than a 100MHz x 4?
Difference between quad / dual pumped FSB
- Thread starter MrEgo
- Start date
Here's a related question. Obviously it's possible to run a PC2700 (333 mhz) SDRAM in synchronous mode with a 333 mhz FSB Athlon at a 166/333 bus speed. But is it NOT possible to run PC2700 sync on a 533 mhz FSB P4 without overclocking the FSB? Seems like it would only be possible to run synchronous with PC2100 (133/266).
These FSB numbers are useful in determining the bandwidth between the CPU and the memory. Take an Athlon that has a FSB of 133MHz, but because there are two transfers per cycle, its effective FSB is 266MHz. To find the bandwidth, multiply the effective FSB by the bus width, which is 8 bytes and that will give you 2100 Megabytes of data per second. The optimal configuration for Athlons is to have memory that runs at the same effective speed as the FSB.
To answer your first question, the quad pumped FSB delivers twice as much bandwidth as a double pumped FSB of the same speed. Also, a 2 x 200FSB would have the same amount of bandwidth as a 4 x 100FSB. However, a CPU's performance does not scale with bandwidth. Once the CPU is has the bandwidth it needs, more bandwidth gives diminishing returns. Athlon processors need only the double pumped bus before they are saturated, while the Pentium 4's bandwidth demands require a quad pumped FSB. I hope to see this demonstrated when the quad pumped 200MHz FSB Pentium 2.4GHz is released by having a performance comparison between the 4 x 100MHz 2.4 and a 4 X 133MHz 2.4B and the 4 x 200 2.4C (?).
vicwang, you hit the nail on the head. However, single channel Pentium 4's need bandwidth more than they need to be synchronous, at least that is what I have observed from others. Peace.
To answer your first question, the quad pumped FSB delivers twice as much bandwidth as a double pumped FSB of the same speed. Also, a 2 x 200FSB would have the same amount of bandwidth as a 4 x 100FSB. However, a CPU's performance does not scale with bandwidth. Once the CPU is has the bandwidth it needs, more bandwidth gives diminishing returns. Athlon processors need only the double pumped bus before they are saturated, while the Pentium 4's bandwidth demands require a quad pumped FSB. I hope to see this demonstrated when the quad pumped 200MHz FSB Pentium 2.4GHz is released by having a performance comparison between the 4 x 100MHz 2.4 and a 4 X 133MHz 2.4B and the 4 x 200 2.4C (?).
vicwang, you hit the nail on the head. However, single channel Pentium 4's need bandwidth more than they need to be synchronous, at least that is what I have observed from others. Peace.
>So I'm starting to get confused with that.. like would a 200MHz x 2 be "better" than a 100MHz x 4
As far as the quantity of memory access delivered is concerned, it makes no difference if it is 2 X 200 or 4 x 100.
The problem is more what the CPU can do with the bandwidth. Looking at the history of the numbers reported in reviews as these numbers have changed, and the explanations given, the Athlons do not profit very much from an increased bandwidth, but the P4 does. Why exactly is that? You could reverse the implication and say the Athlon is well isolated from the effects of slow memory; therefore faster memory does not have a large effect. It sounds different depending on the way you put it.
The P4 runs at higher multiplier to duplicate the performance of the similar rated Athlon. Therefore memory access delays affect the P4 more than the Athlon. Beefing up the P4 memory system is more important for P4 performance than the Athlon's. When it comes to programs whose speed depends almost totally on how fast memory is accessed rather than instruction execution speed, the P4s superior memory system makes it a winner.
The nForce chipset operates 2 banks of memory simultaneously, doubling memory bandwidth. For the orginal nForce, the effect on the Athlon's performace was nearly zero. (Although the benefit is large for the on-board video if it has it.) But for nForce2, performance improvement is in some cases significant. The reason is supposed to be that the new nForce2 memory controllers have a smarter algorithm, which determines better what the Athlon is really going to need. In other words, raw bandwidth still doesn't really help the Athlon; it is anticipating correctly what the Athlon will need that helps. Still, with double the bandwidth, it should always benefit the Athlon somewhat; why doesn't it? After all, increasing the FSB in sync with memory does.
You're confused? I'm confused!
As far as the quantity of memory access delivered is concerned, it makes no difference if it is 2 X 200 or 4 x 100.
The problem is more what the CPU can do with the bandwidth. Looking at the history of the numbers reported in reviews as these numbers have changed, and the explanations given, the Athlons do not profit very much from an increased bandwidth, but the P4 does. Why exactly is that? You could reverse the implication and say the Athlon is well isolated from the effects of slow memory; therefore faster memory does not have a large effect. It sounds different depending on the way you put it.
The P4 runs at higher multiplier to duplicate the performance of the similar rated Athlon. Therefore memory access delays affect the P4 more than the Athlon. Beefing up the P4 memory system is more important for P4 performance than the Athlon's. When it comes to programs whose speed depends almost totally on how fast memory is accessed rather than instruction execution speed, the P4s superior memory system makes it a winner.
The nForce chipset operates 2 banks of memory simultaneously, doubling memory bandwidth. For the orginal nForce, the effect on the Athlon's performace was nearly zero. (Although the benefit is large for the on-board video if it has it.) But for nForce2, performance improvement is in some cases significant. The reason is supposed to be that the new nForce2 memory controllers have a smarter algorithm, which determines better what the Athlon is really going to need. In other words, raw bandwidth still doesn't really help the Athlon; it is anticipating correctly what the Athlon will need that helps. Still, with double the bandwidth, it should always benefit the Athlon somewhat; why doesn't it? After all, increasing the FSB in sync with memory does.
You're confused? I'm confused!
"vicwang, you hit the nail on the head. However, single channel Pentium 4's need bandwidth more than they need to be synchronous, at least that is what I have observed from others. Peace."
I saw an article recently where a 400 mhz DDR was actually SLOWER than 333 mhz DDR on a 533 mhz FSB P4 system. My first thought was "OK, well that's because the 400 mhz DDR is running asynch". But then I realized, isn't 333 mhz also asynch?
The article CLAIMED it was because 400 mhz DDR is "developing standard" or something like that, and that it is not yet an "official" standard. That just seemed like B.S. to me. Although if it's not the asynch issue, I don't know why DDR 400 would be slower than 333.
I saw an article recently where a 400 mhz DDR was actually SLOWER than 333 mhz DDR on a 533 mhz FSB P4 system. My first thought was "OK, well that's because the 400 mhz DDR is running asynch". But then I realized, isn't 333 mhz also asynch?
The article CLAIMED it was because 400 mhz DDR is "developing standard" or something like that, and that it is not yet an "official" standard. That just seemed like B.S. to me. Although if it's not the asynch issue, I don't know why DDR 400 would be slower than 333.
Cerb
Elite Member
- Aug 26, 2000
- 17,484
- 33
- 86
vicwang, it depends on the chipset and mobo. Sometimes there's a good difference, sometimes nill. I have a feeling it's BS. I think we all know that the mobo and chipset companies know JEDEC is a wee bit slow, and that there isn't much to figure out what DDR400 needs to be compared to 266 and 333.
>I think we all know that the mobo and chipset companies know JEDEC is a wee bit slow
The story goes more like this: The JEDEC committee, after interminable debate and delay, was about to issue the standard. Most of the companies that make memory thought it was a terrible standard. So the standard was put on hold and never issued. That was all a long time ago.
BIOS writers don't know what to do for DDR400, so they try to make the settings loose enough to work with anything. Therefore it is slow. OCing memory to 400 from a 333 setting does a lot better (and is more like what the memory manufactures actually wanted).
The story goes more like this: The JEDEC committee, after interminable debate and delay, was about to issue the standard. Most of the companies that make memory thought it was a terrible standard. So the standard was put on hold and never issued. That was all a long time ago.
BIOS writers don't know what to do for DDR400, so they try to make the settings loose enough to work with anything. Therefore it is slow. OCing memory to 400 from a 333 setting does a lot better (and is more like what the memory manufactures actually wanted).
The Athlon actually benefits from increased bandwidth more than I expected. If you check out the "Barton 3200+ hypothetical preview" at Lost Circuits, it shows that a simple move from sync'd DDR333 to DDR400 mem/FSB yields impressive performance increases at the same total clockspeed. Of course, this is assuming you can maintain cas 2 with tight timings as they did in the comparison. Bandwidth has always benefitted the XP (not as much as the P4), but I didn't think it made that much difference after 166MHz FSB. I still think AMD will have a tough time being competitive with the XP in its current form once Intel rolls out quad-pumped 200MHz P4's with HT2 on dual-channel DDR Canterwood boards.
Chiz
Chiz
"Originally posted by: THUGSROOK
youre forgetting about intel 4x 200fsb"
How well are these going to overclock ? The 800FSB 2.4 will have a multiplier of 12, right ? If it overclocks as well as current 2.4s, say 3.0g then the FSB will have to be 250x4 or 1000FSB, correct ?
Is this going to be feasable ? Those of you who have thought about this, what speed of DDR memory is going to be the target for a 250FSB ? Is it hoped that a memory to cpu ratio of 1:1 requiring DDR500 ram can be achieved ?
If it becomes feasible to have a dual ddr500 system, running a 2.4 at 3.0, with an FSB of 250, quad pumped to 1000FSB, is the theoritical bandwidth of the memory going to be too much or too little for the processor at 250FSB ?
If my figures are off, I'd appreciate being corrected. Also what other combinations of the new 800fsb processors and DDR motherboards are people hoping to overclock with ?
youre forgetting about intel 4x 200fsb"
How well are these going to overclock ? The 800FSB 2.4 will have a multiplier of 12, right ? If it overclocks as well as current 2.4s, say 3.0g then the FSB will have to be 250x4 or 1000FSB, correct ?
Is this going to be feasable ? Those of you who have thought about this, what speed of DDR memory is going to be the target for a 250FSB ? Is it hoped that a memory to cpu ratio of 1:1 requiring DDR500 ram can be achieved ?
If it becomes feasible to have a dual ddr500 system, running a 2.4 at 3.0, with an FSB of 250, quad pumped to 1000FSB, is the theoritical bandwidth of the memory going to be too much or too little for the processor at 250FSB ?
If my figures are off, I'd appreciate being corrected. Also what other combinations of the new 800fsb processors and DDR motherboards are people hoping to overclock with ?
That'd be 4ns wouldn't it? I don't know if there are any 4ns modules currently available for system memory, the closest I've seen is 4.6ns on Kingston HyperX PC3700. I don't think people will be shooting for 1:1 ratios with P4 platforms, as it seems the P4 doesn't suffer from async mem/fsb ratios like the Athlon does. In fact, an article (I believe it was by Ed over at overclockers.com) of Springdale/Canterwood boards have already indicated a lower mem/fsb ratio in order to accomodate the lack of high speed dimm modules for overclockers. There's also the rumor that Canterwood will allow multiplier unlocking on 200MHz P4's on either of the above boards, which kinda answers your other question (unconfirmed though).
Chiz
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 + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter