Processor Speed Versus System Bus speed

[ShinyRocks]

In an old issue of MaximumPC, they benchmarked two computers with these configurations

A) 3.58 Ghz (overclocked from 3.0 Ghz) Pentium
600+ Mhz Front Side Bus

B) 3.0 Ghz Pentium
800 Mhz Front Side Bus

Despite having nearly a 600 Mhz advantage, Computer A was slower in ALL the benchmarks. In fact, a higher Bus Speed is better than 600 Mhz in processor speed ratings. Now is that always the case or just the particular pentium 4 architecture? Thoughts?

 

[Ionizer86]

It really depends on how bottlenecked the bus is. If the processor needs more data to crunch, then it'll need a bigger bus to get data through. If it's already loaded on data, then widening the bus isn't doing much. It's like taking a 2 lane road and making it 8 lanes when 4 are enough.

From what I've seen, going from 400-->533 helped quite a bit with alleviating bottlenecks, while going from 533-->800 didn't help as much since the bottleneck wasn't as great at that point. I've also noticed that the 3.0/800 is sometimes faster and sometimes slow than a 3.06/533. The MaximumPC article may have not used a wide range of benchmark apps.

 

[ShinyRocks]

It's not a 3.06 Ghz. It was a 3.58 Ghz nearly a 3.6 Ghz processor speed. The processor was overclocked by nearly 600 mhz.

 

[DarkSarkas]

For the Pentium 4 architecture itself, the bus speed increase is a huge advantage due to available memory bandwidth. The Pentium 4 architecture is extremely efficient when it breaks down tasks, but on a project-by-project basis it is very inefficient when dealing with branching programs. A good example is the speed increases of the P4 EE over the P4 normal. The P4 EE wins in benchmarks because of increased efficiency of the memory substructure due to the L3 cache. Increasing the bus speed in a P4 gives it more memory bandwidth, which increases the speed with which it is able to queue up and execute code.

Remember the original p4? It had large amounts of bandwidth for a processor of it's time, but it was still hindered by the lack of access it had to memory. As the bus speed increased, the benchmarks that the p4 was subjected to gave more positive results as well.

P4 + Memory Bandwidth (and decreased memory latency) = increased performance

That's the basic concept behind the increases in speed that the P4 sees from increased bus speeds. This applies to any processor, but with the P4, because of its dependence on the availability of gobs of high-speed memory, it is more apparent.

 

[ShinyRocks]

So is a 2.4 GHz pentium with 800 mhz bus faster than a 2.8 Ghz pentium with a 533 mzh bus?

 

[nick1985]

Originally posted by: ShinyRocks
So is a 2.4 GHz pentium with 800 mhz bus faster than a 2.8 Ghz pentium with a 533 mzh bus?

they will probably be pretty close.

if you are amazed by this, check out this article and your jaw will drop.

 

[Duvie]

Originally posted by: ShinyRocks
In an old issue of MaximumPC, they benchmarked two computers with these configurations

A) 3.58 Ghz (overclocked from 3.0 Ghz) Pentium
533 Mhz Front Side Bus

B) 3.0 Ghz Pentium
800 Mhz Front Side Bus

Despite having nearly a 600 Mhz advantage, Computer A was slower in ALL the benchmarks. In fact, a 267 mhz Bus Speed is better than 600 Mhz in processor speed ratings. Now is that always the case or just the particular pentium 4 architecture? Thoughts?

First off I am confused....What has a system bus of 267mhz??? Also the B system was a stock 3.0ghz 800fsb cpu??? was the HyperThreading enabled or not???

I have tested this and I can tell you it is a pile of lies and crap unless it occurred in a few apps and a very limited testing. HT had to be enabled and the apps had to be extremely HT optimised.

Also for this to be true this had to happen...

1) HT is enabled
2) system A was not a dual channel DDR system...
3) HT optimsed programs


I have a thread around where I tested a p4b at 3.2ghz with single channel DDR and Dual channel ddr and a p4c at 3.2ghz...All same speed and the spreads were not that big until I started using specific apps that had HT optimization...IE TMPgenc I had a 22% increase alone versus same p4c chip with HT disabled. that would have been worth near 500-600mhz by itself but thisis not the norm. The p4b single channel DDR to p4c non HT enabled garner 5-15% max...

 

[ShinyRocks]

I'll try my best to answer.

The MaximumPC article was a couple months ago but HT was already out by then so both of them would be using HT.

My post was talking about bus speed not ddr channel memory. It seems to me that having a faster bus speed is better than raw processor speed. Plus it saves you a hell of a lot more money. However, please feel free to point out anything that is wrong. I'm not 100% sure...yet.

 

[MrPickins]

Unless they had an unlocked engineering sample, the P4B 3.06@3.58 would have to be running a 622Mhz bus speed, not 533.

I agree that Maximum PC must not have used a very good test method.

Edited because I always reply to fast.

 

[ShinyRocks]

Originally posted by: MrPickins
Unless they had an unlocked engineering sample, the P4B 3.06@3.58 would have to be running a 622Mhz bus speed.

You're right, the overclocked pc bus was above 533 mhz... around 600+. Don't have the article with me right now :-(

 

[NokiaDude]

BEHOLD!!!

 

[Acanthus]

There were also core improvements going from a B class to C class processor. Also was HT enabled on the C class cpu?

 

[RussianSensation]

It is fairly easy to find out the % of perfomance gained with pentium 4 since intel processors scale in speed almost always proportional with scaling of the MHZ (speed)

Therefore, to test this hypothesis you would take 2 systems:

1) 2.4ghz p4 C @ 3.0ghz with 250FSB
2) 3.0ghz p4 C @ 3.0ghz with 200FSB

the result will be 1) > 2) by X % points (probably not more than 5-7% increase in speed though)

now in general 3000 system is faster in most benchmarks then a 2400mhz system by (3000/2400 = (approximately) up to 25% speed increase)

so then naturally 3580 system /3000 system = (approximately) 19.3%
then you would take 19.3% and subtract potential gain of higher FSB (approximate this to be 5%) so the system with higher mhz will be faster by at least 12-14%

Now if maximum PC compared p4 b to C thats already not reasonable because the higher mhz speed of B is offset by the Hyperthreading and higher FSB and hyperthreading can gain huge %%% improvements so u dont know if the system 2 in your case won because the applications were HT optimized or because of faster front side bus alone

But following the logic I described above 2 systems based on P4C core with HT

Higher clock speed will almost always be better than faster front side bus with lower clock speed (considering you have larger than 100mhz improvement in speed at least)

cheers