Memtest86 & Z68 Chipset - false readings?

[Nelly]

Does anyone else have false readings when using memtest86 with a Z68 Chipset motherboard? I can verify the same problem when I use memtest86, just wondering if it is the same for other manufacturer motherboards.

Credit to this goes to Bernhard on the MSI Forums.

Motherboard: MSI Z68A-GD65

Chipset: Intel® Z68 Express Chipset (Sandybridge)

Processor: Intel i5 2500k

Please read this thread: http://forum-en.msi.com/index.php?topic=152693.0

I believe that memtest 86 needs some work to be compatible with the Z68 chipset as it is clear from the tests that there is no real world performance loss.

So where do we start this time ?
After having set up my new Z68A-GD65 (G3) board, which went without a hitch I must add, I went through a couple of beta BIOS versions to test and see what the restricted multi was all about.
The .N34 beta BIOS eventually had no more multi restriction, but was by no means perfect. Even with the minor issues that still exist, my board is solid and performs up to expectations, maybe even better than what I expected after now having run some comprehensive tests. But more about that later.
First the hardware:
MSI Z68-GD65 (G3)
i5-2500K
Arctic Cooling Freezer 13
Crucial 2X2GB CT2KIT25664BA1339 first mem kit tested
Corsair CML8GX3M2A1600C9W -1,35V Vengeance -kit currently on the MB
Nvidia 8600GT (currently for testing only)
Seagate 500GB SATA II
Sony Optiarc DVD RW
Corsair CX500 PSU
all mounted on a CM Lab test bench
OS Win 7-64bit
BIOS used for testing V23.3B4 (.N34)

Other hardware used: Fluke model 77 digital multimeter.

One thing that does need mentioning is that the .N34 BIOS applies the correct memory voltage for the Low voltage Vengeance when XMP is enabled. The previous BIOS did not do that and one has to manually adjust it.
Within the BIOS itself, the reported "Current DRAM voltage" in Auto and Manual does not reflect what is applied in "DRAM Voltage" and even that differs from the real voltage. I used a Fluke Digital Multimeter to eventually find the correct voltages and for anyone that wants to apply the correct voltages, here are two tables as tested.

Having used the multimeter to verify the VCore, I found that CPUZ reflects the most accurate reading out of all the applets used. A 1,36 Core voltage in CPUZ was reflected as 1,363V on the multi.

Now let's get to the real reasons for the tests.
In the recent past we had a user on a MSI sister forum that ran memtest 86 and found what he described as reduced base clocks whenever the CPU multiplier was increased by the "Adjust CPU Ratio" function in the BIOS. This led him to believe that although the board is running at an increased core frequency, that the performance would decrease as the BCLK and mem speed in memtest showed this clearly. No real benchmark tests had been done in the OS to either confirm or deny this suspicion, except a remark that it feels considerably slower than the same CPU on another manufacturer's P67.
At this point I must point out that it is not my intention nor the forum's intention to compare manufacturers motherboards for benchmarking. I shall include sample tests performed on that board which should clearly show any performance degradation, if the memtest data is anything to go by.
Having read the post I decided it would be a nice challenge to see if this memtest phenomena actually translates to a real performance reduction once the OS is loaded and benchmarks are performed. I am fortunate that I also happen to have the exact board that the user was referring to and mine has an i7-2600K on it.
First things first. Let's have a look at what the reported memtest issue is all about

This first screenshot shows the values that were monitored to be inconsistent with what is set in the BIOS. My first screenshot was taken with the BCLK at 100 (99.8) and the DDR3-1600 modules.

The second screenshot was taken with my 1333 memory installed and my mutiplier set at X45 and still a BCLK of 100 (99.8). You can clearly see -that the BCLK is at 73, the memory is at DDR3-975. I tested various multis from 33 through to 45 and as the multi increased, these reflected values decreased. Below are the results of my tests with the 1333 memory:
@ X33 everything is still normal
@X34 BCLK=96 -mem DDR3-1291
@X38 BCLK=86 -mem DDR3-1155
@X40 BCLK=82 -mem DDR3-1097
@X43 BCLK=76 -mem DDR3-1021
@X44 BCLK=74 -mem DDR3- 997
@X45 BCLK=73 -mem DDR3- 975

I checked this with the DDR3-1600 modules and the behaviour is the same. I only ran the DDR3-1600 up to X42 as it was enough to prove the similar results.
@X33 BCLK=99 -mem DDR3-1596
@X36 BCLK=91 -mem DDR3-1463
@X39 BCLK=84 -mem DDR3-1351
@X42 BCLK=78 -mem DDR3-1254

It must be remembered that at all times the memory was set at 1333 and 1600 respectively in the BIOS.
Thus the user's valid assumption that something is amiss with the Z68 board.
Since these values in memtest do tend to lead one to believe an associated performance penalty, it was time to put it to the test in the real world and run some applications that could measure overall memory performance and also the total system performance.
My choice was to use Maxxmem which produces comprehensive theoretical benchmark numbers and the 7ZIP which has a benchmark tool included, also providing a comprehensive system bench in terms of file compression and decompression.
The software used:
Win7-64bit OS
CPU-Z v1.58
Maxxmem v1.95
7Zip v9.2
Both the systems used are running with 2X4GB DDR3-1600 Vengeance 9.9.9.24 timings.
My Z68 has the i5-2500K and the P8P67-Pro has the i7-2600K.
Now before anyone cries foul that the i7-2600K has 8 threads and the 2500K only 4. My choice of 7Zip was done because it's benchmark has the ability to restrict the amount of threads for testing. So within the 7ZIP benchmark I used only 1 thread throughout all the tests to ensure a fair comparison.
I quote the following from the Help file.
The benchmark shows a rating in MIPS (million instructions per second). The rating value is calculated from the measured speed, and it is normalized with results of Intel Core 2 CPU with multi-threading option switched off. So if you have modern CPU from Intel or AMD, rating values in single-thread mode must be close to real CPU frequency.
The focus is on the Z68 tests and only X45 was used to do a performance comparison between the Z68 and P67 platform.

First the Maxxmem results:

This already indicates that the perceived performance penalty deduced from memtest 86 has no influence on real world performance. As the multi is increased the memory performance increases proportionally.


Now let's have a look at the 7ZIP benchmark tests. These tests were all done with one core, unless otherwise specified and all used a default dictionary size of 32mb.

This one came as a bit of a surprise. You can see that the Z68 actually outperforms the P67 with it's 2600K when both are running 4 threads.
Once all 8 threads are used, then everything is as per expectation.

A sample overview of the values as reflected within the benchmark applications.

And below a snapshot of my Z68 on the bench test. I know it looks like a mess, but when you are doing testing, nothing beats an open bench.

In closing, I believe that memtest 86 needs some work to be compatible with the Z68 chipset as it is clear from the tests that there is no real world performance loss.

Furthermore I invite other users of Z68 boards to do their own tests on MSI boards and post them for comparative purposes, especially those users that still have a BIOS causing them to have the mysterious throttling.