Gigabyte GA-K8NXP-SLI nForce 4 SLI - PAGE 6Terren Tong - Thursday, January 20th, 2005
Disk Benchmarks
We might as well blot out the names on the graphs, as there is nearly no performance difference between the VIA and NVIDIA disk controllers in a single drive configuration. CPU usage seems a touch higher on the Gigabyte but it is well within HD Tach's margin of error.
USB Performance
While the USB throughput of the nForce 4 is lower than the K8T800Pro, the difference in processor usage is pretty astounding - a whopping 80% difference. Transfer rates are about 11% off the mark of the VIA solution. Intel solutions with the 3.4EE and 3.6 also post around 8-10% CPU usage during this benchmark.
LAN Testing
The server system is a Soltek K8AN2E-GR (nForce 3 250Gb) paired with an Athlon 3400+.
LAN throughput on the Gigabyte is higher than that of the VIA counterparts but correspondingly the processor usage seems to be a lot higher. The drivers provided by Gigabyte did not include the new NVIDIA Firewall software so we decided to throw in the beta drivers and see what kind of difference it made and there were some surprising results. We decided to vary the firewall settings a bit to see what kind of results we got.
Throughput remained in the same general ballpark, +/- 10Mbps - however it is surprising to see the Firewall set to Medium having the highest transfer rate.
CPU Utilization however is the most interesting especially with the firewall set to off - we are ensured that both the NVIDIA firewall features and the Windows based one were both set to off, yet repeated tests still shwoed extremely high processor usage. The firewall settings at low medium and high are a lot more in line with what we expect from current chipsets. It is noteworthy that the nForce 4 MCP does manage to offload so much work from the processor especially when comparing results to the Windows SP2 firewall which topped out processor usage at nearly 90%.
We got in touch with NVIDIA about this apparent discrepancy and we were told that it is the ActiveArmor component that causes this - it handles TCP/IP offloading and with the firewall set to off, ActiveArmor is also disabled. With ActiveArmor on, processor usage again drops correspondingly. However we do not see this on the nForce 250Gb regardless of the firewall setting and we've been following up on this to see what the issue is and will update when we have further answers.