Review: VIA EPIA M w/ new Nehemiah core

Viewing page 3 of 4 pages. Previous 1 2 3 4 Next


Physically, the CPU change from the M9000 is found only by removing the CPU heatsink: details are printed on the CPU casing ("133 x 7.5" and so on). Curiously, the review sample was identified by both Windows XP and Sisoft Sandra 2003 as a VIA C3 933. It's likely that in their haste to get this pre-production board to us, VIA neglected to update the multiplier in the board. Flashing the BIOS to the latest version,10B, did not change this incorrect data. Windows XP does identify it clearly as a Nehemiah.

Taking some advice from, WCPUID by H.Oda was used to try and change the CPU multipler. The utility did not recognize the board or CPU, even though the correct information was displayed. The change multiplier feature would not work on this EPIA M10000.

An email was fired off to VIA to determine how and whether the CPU could be made to run at the correct speed, but everyone there are likeLy too busy at CeBit in Germany, reputed to be the biggest computer show in the world. Any response from VIA pertaining to this issue will be published later as a postscript.

For the time being, it is probably safe to say that the benchmark results for a production-run 1 GHz M10000 Nehemiah would probably be a bit better than our 933 MHz sample. This problem is actually something of a benefit: we get to pit the Nehemiah core against the Ezra-T core at the same speed!


In one of their features / benefits pages, VIA provides a comparative graph showing the difference between a 1 GHz Ezra-T core C3 and a 1 GHz Nehemiah C3 on a VIA C3M266 main board, which uses the same Apollo CLE266 chipset with integrated VGA as the EPIA M boards. The differences appear significant, especially in 3D Mark 2001 SE.

The Nehemiah core does much better at 3D imaging than the previous Ezra-T.

This data needed to be verified, so a system was assembled by putting components onto an open frame platform, and then the motherboards tested with these components, one at a time. As the chipsets and video cards and all "peripheral devices" are identical, there was no need to make any driver adjustments when the motherboards were swapped.

At the last moment, an M6000 was also thrown into the fray for curiosity's sake. Let's check out the whole M family at once, shall we?

EPIA M10000, M9000 or M6000
PC2700 DDR RAM, 512 meg
Seagate Barracuda IV 40G hard drive
Nexus NX3000 power supply
Hitachi DVD 16X drive
Windows XP Pro SP1
Samsung 955DF monitor
PowerDVD XP 4.0

The EPIA test rig.


Sisoft Sandra 2003 was used to run some benchmarks. The improvement between the M10000 and M9000 in the SANDRA multimedia benchmark is 89%, even higher than the 74% obtained by VIA in 3D Mark 2001. The 22% gain in the CPU Arithmetic score is also interesting. This is despite the 2 boards running at the SAME clock speed.

Between the M6000 and M9000, one would expect to see only straight line gains based on the 50% increase in clock speed, but this is not the case. With CPU Arithmetic, the gain was 57%; with Multimedia Integer, it was 76%

Memory Bandwidth, which is one of the most often cited Sandra benchmarks, was pretty much the same for all three boards: 220/250. Wiping the floor in terms of comparisons against current chipsets for Intel and AMD. (An aside: Why did VIA move the board from SDRAM to DDRAM when the M series boards' bottleneck is clearly not memory bandwidth? The reason is a practical one: DDRAM is becoming the default standard for memory everywhere. Volume determines price and availability. Already SDRAM prices are a bit higher than DDRAM.)

% gain*
CPU Arithmetic / FPU
770 / 211
1212 / 330
1483 / 343
+ 22% / + 4%
MM Integer / Float iSSE
631 / 1195
1113 / 1836
2104 / 2126
+ 47% / + 14%

*NOTE: % gain is the improvement of M10000 scores over M9000

The Cache / Memory benchmark results are really interesting. Sandra contains typical benchmarks of a broad range of CPUs and chipsets in its database so that the system or CPU under test can be compared. Because all the current CPUs from AMD and Intel are running at speeds well in excess of 1 GHz, older comparison CPU/chipset combinations were chosen from Sandra's database.

933 MHz Ezra-T EPIA M9000 (blue line) scores 2.5 times worse than its closest competitor, a long discontinued P3-750 on the Intel 815 chipset. (Click on the image for a large view in a new window.) The EPIA M6000 scores about 40% lower.

Click for larger image

The 1GHz Nehemiah EPIA M1000 (blue line) makes a miraculous improvement over the M9000, moving up to the middle of the pack, a 300% better performance than the M9000. (Click on the image for a large view in a new window.)

Click for larger image

Previous 1 2 3 4 Next

CPUs|Motherboards - Article Index
Help support this site, buy from one of our affiliate retailers!