Here are my first impressions of the NVMM Advance Path feature (NVMM = Nvidia Memory Manager) which is enabled in 2K090422, and was expected to bring about considerably faster memory access speeds.
According to the Lavalys EVEREST Cache & Memory Benchmark,
the memory read speed has increased very considerably, from 4847 to 5857 MB/s
the memory write speed has decreased slightly, from 4229 to 4158 MB/s
the memory copy speed has increased from 4208 to 4534 MB/s
the memory latency has dropped down very considerably, from 112.7 to 81.6 ns
(Intel E5200 2.5GHz, 4GB of dual channel DDR2-800 RAM, no overclocking, XP SP3).
Now, some potentially bad news. I still have not had the time for extensive testing, but it seems that the system has become more unstable. I had one of the two threads in prime95 crash within five minutes of the start of the program (but after restarting prime95, this was not repeated); there were several inexplicable random crashes of firefox 3.0.10 on news websites which have always worked without problems, internet explorer 8.0 also crashed, again randomly, on a couple of occasions with a "blue screen of death". All of these look like random memory access errors, which were not occurring under Bios 2K090202A . I intend to run full memtest86+ tests in the evening, both with NVMM Advance Path enabled and disabled, in an attempt to understand what is happening. For the record, I have a very well cooled system, and none of the temperatures ever exceed 50 deg C, even after running prime95 for hours.
According to one NVidia document, http://www.nvidia.co...DIA_EPP2_TB.pdf
Quote
The motherboard BIOS implements a dynamic Advance Path deterministic
algorithm, which is executed at BIOS initialization on a per-motherboard basis,
generating different Advance Path settings for each platform motherboard. You
gain additional overclocking headroom as a result.
algorithm, which is executed at BIOS initialization on a per-motherboard basis,
generating different Advance Path settings for each platform motherboard. You
gain additional overclocking headroom as a result.
This suggests that the NVMM Advance Path function is associated with a reasonably complicated piece of code in te Bios, and ensuring that this code works reliably in all situations could be a challenging task.
I would be interested to hear the impressions of other users who have switched to 2K090422, as well as comments from Technical Support. I will report my memtest86+ results once I have carried out the tests.
