Socket 939 Roundup: Battle at the Top

Abit AV8: Overclocking and Stress Testing

FSB Overclocking Results

Front Side Bus Overclocking Testbed
Processor:	Athlon 64 FX53 Socket 939 2.4GHz
CPU Voltage:	1.5V (default)
Cooling:	Thermaltake Silent Boost K8
Power Supply:	Antec TruePower 430W
Maximum OC: (Standard Ratios)	203 x13 2642MHz (+10.1%)
Maximum FSB: (Lower Ratio)	280x9 (2520) at 1:1 Memory

Our FX53 easily achieved the next speed level of 2.6GHz on the Abit AV8. This will likely be the speed rating of the FX55, which will appear later this year. The best proof that the Abit has a working PCI/AGP lock can be seen in the fact that we were able to run DDR550 memory at 280x9 on the Abit. To achieve this overclock level, we needed to reduce the base HyperTransport frequency to 800. The AV8 could reach 300 CPU frequency at lower memory ratios and a 8 multiplier. We did find that the Abit did not really like an HT setting lower than 800, so we were limited to 300 as the maximum usable FSB.

Memory Stress Test Results:

The memory stress test measures the ability of the Abit AV8 to operate at its officially supported memory frequency (400MHz DDR), at the best performing memory timings that our Mushkin PC3500 Level 2 or OCZ PC3500 Platinum Ltd Modules will support. Memory stress testing was conducted by running RAM at 400MHz with 2 DIMM slots operating in Dual-Channel mode. The memory configuration of the Abit is a little unusual in that the first 2 DIMM slots represent a Dual-Channel. On most Dual-Channel motherboards, slots 1 and 3 are the first Dual Channel.

Stable DDR400 Timings - 2 DIMMs (2/4 DIMMs - 1 Dual-Channel Bank)
Clock Speed:	200MHz
Timing Mode:	N/A
CAS Latency:	2.0
Bank Interleave:	N/A
RAS to CAS Delay:	2T
RAS Precharge:	10T*
Precharge Delay:	2T
Command Rate:	1T

*Several memory tests have shown that memory performs fastest on the nVidia nForce and VIA K8T800 chipsets at a TRas (RAS Precharge) setting in the 9 to 13 range. We ran our own Memory Bandwidth tests with memtest86, with TRas settings from 5 to 15 at a wide range of different memory speeds. The best bandwidth was consistently at 9 to 11 at every speed, with TRas 10 always in the best range at every speed. The memory bandwidth improvement at TRas 10 was only 2% to 4% over TRas 5 and 6 depending on the speed, but the performance advantage was consistent across all tests. Since best performance was achieved at 2-2-2-10 timings, all Athlon 64 benchmarks were run at a TRas setting of 10.

The Abit KV8 PRO was completely stable with 2 DIMMs in Dual-Channel at the best performing settings of 2-2-2-10, at 2.6V default voltage. It should be noted that the BH5 memory modules that we used for testing are no longer available for purchase, but we have not yet established our standard memory for future testing. We will be using a new standard memory in future motherboard tests.

As we first found in our latest Socket 754 roundup, the Command Rate is very important for top performance on a VIA chipset motherboard. The best performance is at a Command Rate of 1T, and the Abit AV8 was completely stable at a 1T setting with 2 DIMMs. Standard memory bandwidth measured with SiSoft Sandra 2004 SP2 shows a 6000 MB/s bandwidth with 1T Command Rate compared to a 5000 MB/s bandwidth with a 2T setting. While we could not set "Bank Interleave" in the BIOS, Sandra 2004 reported that a 2-way Bank Interleave was being used by the memory controller. With an on-CPU memory controller with Athlon 64, this is not as important a setting as it is on with chipset-based memory controllers.

Filling all four available memory slots is more strenuous on the memory subsystem than testing 2 DIMMs on a motherboard.

Stable DDR400 Timings - 4 DIMMs (4/4 DIMMs - 2 Dual-Channel Banks)
Clock Speed:	200MHz
Timing Mode:	N/A
CAS Latency:	2.0
Bank Interleave:	N/A
RAS to CAS Delay:	2T
RAS Precharge:	10T*
Precharge Delay:	2T
Command Rate:	2T

The Abit AV8 was able to run with all 4 DIMM slots at the same aggressive 2-2-2-10 settings used for 2 DIMMs. However, Command Rate must be reduced to 2T when filling both Dual Channels.

During the course of testing for the 939 roundup, we experienced a failure of our FX53 processor. AMD was very gracious in quickly providing a replacement so testing could continue. The FX53 failure is probably rare, but the way that the CPU failed provided some insight into the Athlon 64 on-board memory controller. We first noticed new boards requiring much slower timings for memory; in most cases CAS 2.5 was required where CAS 2 had worked previously. This appeared to be motherboard differences, so we continued testing. Later, the FX53 failed. The replacement first went into a board where CAS 2.5 or 3 had been required, and now the same board performed fine at CAS 2 memory timings. Our point here is that the problem was the on-CPU memory controller failing and not the motherboard. The on-CPU memory controller is wonderful for reducing latency, but, unlike the past, memory problems may also be related to the CPU and not just the motherboard.