DDR SDRAM Comparison – Is it Worth Waiting For?

Pages: 1 2 3 4 5 6

DDR SDRAM By The Numbers

OK, you have decided your next system will use DDR SDRAM, but the question is what speed memory should you buy – PC1600 or PC2100, CAS 2 or CAS 2.5? What the heck does that all even mean, not just technically, but what kind of ‘Real World’ performance difference should you expect to see between the different types? And don’t forget about 200MHz FSB vs. 266MHz FSB for the CPU, or running the memory ASYNC at PC2100 (133MHz) with a 100MHz Host Clock (200MHz FSB) or SYNC (100 / 100MHz, 133 / 133MHz) … all very confusing.

So, to help you make that decision I ran a series of benchmark tests to see what type of applications would show a performance difference using DDR SDRAM of different speed ratings, and what effect different BIOS settings have. I chose a Socket A mainboard since it is AMD that is promoting DDR SDRAM at this time (I may also do another series of test with the VIA Apollo Pro 266 for Intel CPU’s). I chose an ALi MAGiK1 based mainboard for one major reason: I think ALi will hit the market first with the MAGiK1 chipset, as they seem to be the one in a position to ship chipsets in volume very soon. ALi has also priced the chipset very competitively, which is important to mainboard manufacturers. The AMD 760 is more of an OEM chipset that AMD never intends to ‘mass market’. In fact, it is pretty much an enablement chipset to get the market moving and stimulate mainboard and memory manufactures, not an effort by AMD to get into the chipset manufacturing business. VIA appears to be very late with the KT266 chipset, so I don’t expect to see it actually shipping on mainboards until well after those using the ALi MAGiK1 hit the market. I used the Soyo K7ADA because I had one (maybe one of the most important reasons) and have access to BIOS updates and tech support directly from Soyo Tek here in the USA.

If you’d like to look at the actual MAGiK1 technical specifications take a look at ALi’s web site, or this good article at Lostcircuits. But what interests me the most are the actual benchmark results. In other words, do I really care why the chipset does what it does, or do I just want to see what the results are and what they will mean to me while using the system?

One point to remember as you read this is that the Soyo K7ADA I used here is a pre-production mainboard with an evaluation BIOS. Results for the final production mainboard may vary, since there will likely be changes in the BIOS, maybe the mainboard design itself or even the chipset. I also ran a full set of benchmarks using an FIC AD11, which uses the AMD 761 Northbridge and VIA 686B Southbridge, just to make sure there were no issues with the Soyo K7ADA and its ALi MAGiK1 chipset. There were none that I could see.

The first DDR SDRAM modules I received to test with were Micron PC2100 CAS 2.5 engineering samples. So I wondered what difference I’d see if they were run at PC1600 using both CAS 2 and 2.5 settings vs. running them at PC2100 CAS 2 or 2.5. The other item I ran into was BIOS settings. Using the ALi MAGiK1 chipset I now had a ‘new’ (at least to us familiar with the VIA KT133 Socket A chipset) set of memory timing options in Advance Chipset Features section of the BIOS to adjust (along with auto set by SPD):

• Fail Safe
• Slow
• Normal
• Fast
• Ultra
• Ultra 2

Initially I was very disappointed in the ALi MAGiK 1 200MHz FSB test results (see DDR SDRAM – A Second Glance) when testing with the Soyo K7ADA. But I made one common error: I assumed (that’s the error) that setting the memory ASYNC to 133MHz when using a 100MHz HostClk would be faster than if setting SYNC to 100MHz without testing both settings… Boy was I wrong! Take a look at these results:

For some strange reason all the benchmarks except the 3D Mark 2000 are faster if the memory is set SYNC at 100MHz than if set ASYNC to 133MHz. Why? I don’t know yet, but do plan on finding out. Soyo uses a DIPswitch to select the Host and memory clock to 100/100MHz, 100/133MHz or 133/133MHz. It could be an issue with the mainboard design, BIOS even the chipset design. Any or all of these could change before the production version is released.

Due to this discrepancy, all further test results shown will be with SYNC memory speeds. If a 100MHz HostClk (200MHz FSB) is used the memory will be PC1600 (100MHz), if a 133MHz HostClk (266MHz FSB) then the memory will be PC2100 (133MHz)

Pages:   1 2 3 4 5 6  Next »