Pages: 1 2
There has been much hype and excitement over the past few weeks regarding the M Tech R581A and it’s ability to run at 100MHz, specifically with the AMD K6-266. Admittedly, we have contributed to some of this furor, however there have recently been some published benchmarks, subsequent failures to repeat and a few rationalizations about what the actual situation is. Based upon our tests and observations, we think that some misleading information is being spread and would like to present our own conclusions.
Early Successes
Our very first test with the R581A was with the B1 board revision. This one had a 12ns Tag SRAM chip, which prevented operation beyond 90MHz. With the receipt of the B2 revision, we still could not get the board to boot at 100MHz, but as reported in our previous updates it appeared mostly to be a processor issue.
The day we received the K6-266, we ran a test at 100MHz which was successful. We emailed several of the more prominent hardware review webmasters, M Technology and some vendors we knew were interested in these products. We then published our results.
Soon thereafter, some of the other vendors published some remarkable benchmarks and overclocking results with the R581A and other boards. It looked like it was only a matter of time before someone would be able to put together the combination of items that would allow stable 100MHz operation.
Puzzling Failures
Since we had only performed the initial tests on one set of products, we wanted to get to work duplicating our success with a larger sample. To do that, we pulled out 5 processors and 5 R581A motherboards. After spending an entire day fussing and testing, we published our second set of results, which were a bit less enthusiastic about achieving 100MHz bus speeds.
John Howland of Specialty Tech was also testing, and we traded success stories, as well as failures. At one point, we discovered that changing the L2 Cache access to Write-Thru (rather than Write-Back) would allow a fairly stable 100MHz operation, though the performance was reduced by about 1%-2%. Here at Real World, we were getting very mixed results, with some motherboards working much better at the higher clock speeds than others. When we mentioned this to John, he queried about the board revision because he had noticed his new boards were revision B3, rather than B2.
At this point, we discovered that several of our test boards were B2, while the rest were B3. We then tested more boards, and found that our best success came with the B2 revision boards, but we did not know what the difference was. Eventually, John was told by M Tech that the difference between the two revisions was the removal of a connection to ground on a chip that was overheating at high bus speeds. At this point, we focused on other issues, such as different video cards, etc.
It Must Be Contagious!
It now appears that everyone is having problems duplicating their original successes with these items. Anand has received a couple of boards and a 266 processor and has not been able to successfully duplicate any of the original results. We have found that even 90MHz operation is somewhat unstable, even though our first set of tests showed this speed to be very stable.
We started wondering what the heck was going on here, and decided to do some comprehensive testing, and some deep thinking. After another grueling test session with multiple processors and motherboards, we believe that we have some better understanding of the issues. Unfortunately, we have now seen some apparently thoughtless comments made about this that needs to be addressed.