Bookmark and Share

Articles: CPU

Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 ]

1066MHz Bus vs. 800MHz Bus

The first part of our test session will be devoted to discovering what positive effect results from moving the Pentium 4 Extreme Edition processors to 1066MHz bus. Since the faster bus can appear the only difference between future Pentium 4 XE processor and the Pentium 4 6XX family next year, knowing these numbers could be quite helpful for proper evaluation of future prospects of the Pentium 4 Extreme Edition processors. At the same time we are also interested in another aspect. Since the transition of Intel processors to the LGA775 form-factor and i925/i915 chipsets resulted into a slight performance drop compared with what we had on older platforms based on i875/i865 chipsets, it will also be en interesting question if the 1066MHz bus can actually help to solve this issue. Since in case of 1066MHz bus, the DDR2-533 SDRAM works synchronously with the processor bus in systems built on i925XE. Theoretically, it should speed up the system, so that in the end it turns out faster than an i875P based platforms with DDR400 memory.

For our testing needs we assembled three systems: one based on i925X chipset with 800MHz bus, one based on i925XE chipset with 1066MHz bus, and one on i875P chipset with 800MHz bus. In all three cases we used Pentium 4 XE processors working at 3.2GHz core clock (the samples we had at our disposal allows adjusting the clock frequency multiplier). In other words, in systems supporting 800MHz bus the processor frequency was set as 16 x 200MHz, while in the system supporting 1066MHz bus this setting looked as 12 x 266MHz.

First of all we took a look at the synthetic benchmark results, which illustrated the memory subsystem performance in all three cases:

i925X
FSB 800MHz
DDR2-533

i875P
FSB 800MHz
DDR400

i925XE
FSB 1066MHz
DDR2-533

SiSoft Sandra 2004, RAM Int Buffered Bandwidth, MB/s

4699

4956

5396

SiSoft Sandra 2004, RAM Float Buffered Bandwidth, MB/s

4699

4956

5393

ScienceMark 2.0, Memory Bandwidth, MB/s

4008.1

4351.34

4536.27

ScienceMark 2.0, Memory Latency, cycles

269

240

261

ScienceMark 2.0, Memory Latency, ns

84.06

75

81.56

Both synthetic benchmarks we used, SiSoft Sandra 2004 and ScienceMark 2.0 unanimously show that the transition to faster 1066MHz bus really does result into a noticeable practical increase in the data bandwidth of the bus between the CPU and the system memory. Moreover, in the system with 1066MHz bus and dual-channel DDR2-533 SDRAM the practical bandwidth is higher than the memory bandwidth in an i875P based system. In other words, when we speak about the new systems with DDR2-533 memory we can state that they do ensure higher practical bandwidth of the memory subsystem. However, when we measure the latencies of the memory subsystem, we find out that DDR2-533 cannot compete with DDR400 SDRAM at all here. As we see, even faster system bus and synchronous working mode do not help the system with DDR2 memory to outperform the one based on i875P chipset.

Here I have to make one thing clear first. Namely, you should understand that the practical latency of the memory subsystem, just like its bandwidth, depends not only on the type of the system memory, but also on its timings. DDR400 SDRAM has already reached the top of its potential here: the widely spread memory solutions with the 2-2-2-5 timings is the top, there will be no memory of this type with more aggressive timings out there. And in case of DDR2-533 SDRAM, the evolution has just started, so we can expect the memory of this type to go far beyond the currently common 4-4-4-11 timings settings in the future. Moreover, the engineering samples of the DDR2-533 memory with 3-3-3-8 timings do already exist. So far we will not run all the tests for the systems featuring this type of memory, because it hasn’t yet gone to the mass market. However, we will still offer you a few most essential results obtained on an i925XE based platform with 1066MHz bus and DDR2-533 SDRAM supporting such aggressive timings:

Well, the conclusions are evident. The use of DDR2-533 SDRAM memory with more aggressive timings can significantly increase the memory subsystem performance by raising its bandwidth on the one hand and reducing the latency on the other. It means that DDR2-533 SDRAM with 3-3-3-8 timings has every chance to speed the LGA775 systems up to the level of i875P based ones. However, this is still a future prospect, and in the meanwhile let’s check out the results we can obtain in real applications if we use the DDR2-533 SDRAM with 4-4-4-11 timings available in the stores today.

Pages: [ 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 ]

Discussion

Comments currently: 11
Discussion started: 11/02/04 02:24:45 AM
Latest comment: 06/08/07 09:02:30 PM

View comments

You must log in to add comments.

Forgot password? Registration

remember me