Random Phrase: Make yourself at home...


Read Me First!

This is the Spode's Abode archive. The old articles and forum have been kept here purely for historical purposes and are no longer updated.

Not all portions of this archive may work as expected.


Please visit the new site.



Athlon 64 Chipset Performance Analysis
Written by Peter Barnard, Spode (02/August/04)
Page 3 of 6

Untitled Document

The Test System

All of these tests were run with a Gigabyte K8NS Pro, equipped with an nForce 3 250 chipset and an AMD Athlon 3200+ (1MB) using 2 x 512MB of KingMax @ 2.5-4-4-8.

Although UT2004 was always run in software rendering mode, other benchmarks such as Far Cry, 3DMark and Aquamark where run using a Sapphire 9800XT 256MB.

HyperTransport Speed

The current generation of CPUs use 800MHz 16-Bit HyperTransport as standard, but how crucial is the HyperTransport speed to overall system performance? UT2004 in software rendering mode was chosen for the test, because it offers consistent results and is a real workout for the system, without being limited by GPU performance.

UT2004
HT Speed
1024 x 768
1280 x 1024
1200 x 1600
Difference
200 MHz
24.65
17.54
13.21
-1.3%
400 MHz
24.85
17.64
13.30
-0.6%
600 MHz
24.85
17.64
13.31
-0.6%
800 MHz
24.85
17.82
13.41
0.0% (Base)
1000 MHz
23.74
16.96
12.86
-4.5%

Aquamark
HT Speed
Default Settings
Difference
200 MHz
43,636
-5.7%
400 MHz
46,003
-0.5%
600 MHz
46,284
+0.1%
800 MHz
46,258
0.0% (Base)
1000 MHz
45,775
-1.0%

As the figures show, we are seeing only the smallest performance differences as the HyperTransport is slowed down. This means that the lower Hypertransport speed of the nForce 150 is not really any reason to reject it immediately. More puzzling is the fact that overclocking the HT speed has resulted in a lower score. We can only conclude that since neither the board nor CPU are designed for the higher speed so perhaps corruption is occurring but not badly enough to crash the system. It may well be a timing issue.

It has been suggested that the HyperTransport link offers far more bandwidth than the system could possibly use, unless the AGP, PCI, SATA and Ethernet interfaces are all transmitting data as fast as they can. The standard implementation of HyperTransport gives 6.4 GB/s, and the AGP slot can only manage 2.1 GB/s of data transfer, so it would seem that for gaming and everyday use, the HyperTransport bus has more bandwidth than it could ever need.

Thus, the 1GHz HyperTransport offered by the K8T800 Pro will only be of use once CPUs that support it are available and there is some doubt as to whether there will ever be any available for socket 754.


<< Back | 1 | | 2 | | 3 | | 4 | | 5 | | 6 | Next >>



Copyright Andrew Miller
Please read our disclaimer

Search the site:

Random

Outlook on the move - SyncPST