Memory Stress Testing

Optimum tRAS

In past reviews of AMD motherboards, memory bandwidth tests established that a tRAS of 6 was optimal for the nForce4 chipset, a tRAS of 10 was best for the nForce3 chipset and a tRAS setting of 11 or 12 was generally best for nForce2. Tests with Intel chipsets in the past have generally shown the lowest tRAS setting to provide the best bandwidth with Intel chipsets.

Since this was the first memory stress test of a production 955X board, tRAS timings were first tested with memtest86, a free diagnostic program with its own boot OS that will boot from either a floppy disk or optical disk. Bandwidth was measured from tRas 4 to tRAS 15 at the specified DDR2-667 to determine the best tRAS setting. Since Intel recommends 5-5-5-15 timings with DDR2-667, we wanted to establish recommended timings with lower latency memory from Corsair and OCZ.

 Memtest86 Bandwidth
Asus P5DW2 at DDR2-667
4 tRAS 3443
5 tRAS 3443
6 tRAS 3443
7 tRAS 3443
8 tRAS 3443
9 tRAS 3443
10 tRAS 3443
11 tRAS 3443
12 tRAS 3360
13 tRAS 3360
14 tRAS 3360
15 tRAS 2184

The best bandwidth was achieved with this combination of 955x and Pentium 560 in the 4 to 11 range. Since, historically, Intel chipsets have performed best at the lowest tRAS setting, we used a tRAS of 4 for memory tests. Any value from 4 to 11 should provide similar bandwidth.

Since memory was also tested at DDR2-800, we ran a similar set of tRAS timings at DDR2-800. At that speed, tRAS bandwidth was the same from tRAS 4 to tRAS13, so we tested with tRAS 4 for all DDR2-800 tests.

Memory Stress Tests

Memory stress tests look at the ability of the Asus P5WD2 Premium to operate at the officially supported memory frequency (667MHz DDR2), at the best performing memory timings that Corsair/OCZ PC2-5400 will support. Memory stress testing was conducted by running DDR2 at 667MHz (stock 3:5 ratio) with 2 DIMM slots operating in Dual-Channel mode.

Stable DDR667 Timings - 2 DIMMs
(2/4 DIMMs - 1 Dual-Channel Bank)
Clock Speed: 200MHz (800FSB)
Timing Mode: 3:5 (200:333 - Default)
CAS Latency: 4.0
Bank Interleave: Auto
RAS to CAS Delay: 2
RAS Precharge: 2
Cycle Time (tRAS): 4
Command Rate: NA

The Asus P5WD2 Premium was completely stable with 2 DDR2 modules in Dual-Channel at the settings of 4-2-2-4 at 1.9V.

Filling all four available memory slots is more strenuous on the memory subsystem than testing 2 DDR2 modules on a motherboard.

Stable DDR667 Timings - 4 DIMMs
(4/4 DIMMs - 2 Dual-Channel Banks)
Clock Speed: 200MHz (800FSB)
Timing Mode: 3:5 (200:333 - Default)
CAS Latency: 4.0
Bank Interleave: Auto
RAS to CAS Delay: 3
RAS Precharge: 2
Precharge Delay: 4
Command Rate: NA

4 DDR2 DIMMs were just as stable as 2 DIMMs on the new Asus 955x. We needed slightly slower 4-3-2-4 timings at DDR2-667 when using 4 DDR2 DIMMs at 1.9V. But at 2.0V, the 4 DDR2 dimms ran with complete stability at the same 4-2-2-4 timings that worked with 2 DIMMs. This is a remarkable improvement over the performance that we saw with 4 DDR2 DIMMs on the 925X. Considering the compromises that the AMD Athlon 64 has required with 4 DIMMs, Intel and Asus have done an outstanding job in equipping the 955X with the ability to run 4 DIMMs without much of a compromise.

Since both Corsair and OCZ had submitted new DDR2-667 DIMMs tweaked for best performance and low-latency in the latest DDR2 platforms, these were the only 2 pairs of DIMMs that we had for 4-DIMM testing on this new 955x board. We used a pair of Corsair and a pair of OCZ and these unmatched pairs performed fine at the same timings that worked with 2 DIMMs. Both Corsair and OCZ are based on Micron DDR2 chips, but the SPDs, bandwidth, and timings did differ in our tests of each pair, indicating that they are similar, but not identical in performance.

Asus P5WD2 Premium: Overclocking Corsair and Asus Reach DDR2-1066
Comments Locked

34 Comments

View All Comments

  • fitten - Wednesday, May 11, 2005 - link

    #5 Differences of 3% are usually in the noise of the type tests that most benchmark sites run. 3% is effectively 0% since that is beneath the precision of the tests.
  • Calin - Wednesday, May 11, 2005 - link

    Higher speed memory will only give you a real performance boost if you use one of the fastest processors. On a slower processor, all that extra memory bandwidth will not help at all (or very little)
  • Carfax - Wednesday, May 11, 2005 - link

    I wonder, is Intel ever going to introduce processors with a FSB greater than 1066 before they go to CSI? All this DDR2 bandwidth is going to waste..

  • Darth Farter - Wednesday, May 11, 2005 - link

    what is the msrp for these boards, cause that $255/$248 prices on newegg/ZZF are hard to ignore...
  • Pjotr - Wednesday, May 11, 2005 - link

    "i thought hypertransport was an amd thing, not an intel or nforce thing....."

    nVidia uses HyperTransport between NB and SB. They have since nForce1 on AMD and they also use it in X-Box, if you didn't know.
  • Wesley Fink - Wednesday, May 11, 2005 - link

    #5 - We didn't measure the difference in nF4 Intel and nF4 AMD at 2T. It was a subjective comment. So I have tried to better explain what I found in the paragraph you quote:

    "On the nF4 Intel platform, the performance impact of a 2T Command Rate appears to be rather small, as the nF4 Intel performance remains very competetive with the 955x as far as it goes. However, at just over DDR2-900, the nF4 Intel appears to hit a wall . . ."
  • Zebo - Wednesday, May 11, 2005 - link

    "it would have been better to include a fx-55 as competition "

    Not for INTC;)

    Man that's a nice chipset they got though so missed from nV:)
  • Wesley Fink - Wednesday, May 11, 2005 - link

    overclockingoodness -
    Both boards are rated at DDR2-667, but both easily ran DDR2-800 with the right memory, which is the next logical memory speed. We mainly wanted to see if DDR2-800 made any real performance difference, and the answer is no in Office and Multimedia, and yes in most gaming. For the future this also gives us a full set of benchmarks at DDR2-800 for comparison if we choose to use them.

    We also found the Asus 955X did a marginal DDR2-1066 in early testing so it seemed reasonable to at least test and report benchmarks at the very stable DDR2-800 in addition to DDR2-667. We won't be doing this with all future boards, but the tests did provide some answers to our questions.
  • mrwxyz - Wednesday, May 11, 2005 - link

    i thought hypertransport was an amd thing, not an intel or nforce thing.....
  • Lonyo - Wednesday, May 11, 2005 - link

    On page 5 you mention 2T having less of an impact than with AMD boards.
    Does that mean it has absolutely zero impact then?
    There's a thread on the forums showing 2T for AMD haveing REAL WORLD impacts of maybe 3% slowdown, nothing more except in synthetic tests, so I suppose on the Intel board it makes maybe 1% difference.

Log in

Don't have an account? Sign up now