Announcement

**keith** · Nov-25-2015, 07:17 PM

Originally posted by g126 View Post

On the MacBookPro the UEFI version seems to work ok, with the exception of reporting a mem speed of over 15000 MB/s on PC3-12800 memory (which doesn't seem ok to me)...

The advertised speed of the RAM is actually the maximum theoretical bandwidth of the module itself; it does not actually come close to that in real operation which involves the CPU & memory controller. There are other factors that come into play such as the # channels, memory architecture, bus speed, interleaving, types of instructions being used, and many others. In your case, if it is a dual channel setup, the theoretically bandwidth is actually 2 x 12800 MB/s.

So the advertised speed should be taken with a grain of salt, as there are other factors that affect the actual memory bandwidth. If you do a google search on RAM maximum bandwidth, you'll get several pages on attempts to reach the maximum bandwidth.

Eg. http://codearcana.com/posts/2013/05/18/achieving-maximum-memory-bandwidth.html

Originally posted by g126 View Post

On the Mac Pro (8 cores/2 procs) I can't get multi processors to work due to a claimed EFI limitation... Is this really the case or does the software need to be updated to work with multi-procs on Mac Pro 2009 EFI?

MemTest86 relies on the multiprocessor services provided by the UEFI firmware. For older firmware (such as on the older Macs), the multiprocessor services are either unavailable, limited or buggy. Without a robust implementation of the MP services in UEFI, MemTest86 would not be able to reliably run in any of the multiprocessor modes.

Originally posted by g126 View Post

Another issue I'm having trouble diagnosing on the MAC PRO is the RAM Speed reported on PC3-8300 RAM. With the original 6GB (6x1GB) I got over 6000 MB/s (I was expecting around 8300 as the configuration allows for triple channel operation), however after upgrading to 24GB (6x4GB) the speed fell to 5200MB/s... The configuration hasn't changed (3 identical sticks so should be on triple channel mode), the memory specs (apart form size) hasn't changed... so why has the speed decreased and why I'm not seeing 8300 MB/s?

In general, higher capacity decreases the performance of the RAM. This is due to higher electrical load on the RAM module (eg. due to more ranks) and may cause the RAM to run at lower clock speeds. The number of ranks also plays a role in interleaving across ranks in a channel.

See http://frankdenneman.nl/2015/02/19/m...tem-bandwidth/

Originally posted by g126 View Post

One last question, when executing on plain old windows PCs memtest gives more information about the RAM, such as operating frequency, CAS timings, channel mode, etc... But on MACs this is not the case... Why? Does the EFI just not make that available for programs to read or is memtest not capable of extracting that info from the EFI?

The RAM details (eg. timings, clock speed, etc.) that are displayed are retrieved from the SPD of the RAM module. Accessing the SPD is highly dependent on the motherboard + CPU chipset so although MemTest86 is able to access the SPD module for most chipsets, there are a few chipsets (including Macs) that MemTest86 is unable to support for accessing the SPD modules. This is likely due to proprietary mechanisms used by the vendor to prevent such access. In such case, MemTest86 uses the SMBIOS as a fallback method of retrieving the RAM details. However, because the SMBIOS is filled by the BIOS, some of the RAM details may be incorrect or missing.

That being said, any of the reported RAM details (speed, timings, etc) in MemTest86 do not have any impact on the actual memory testing itself.

**g126** · Nov-26-2015, 01:01 PM

Keith,

Thank you very much for your detailed answer, it was very helpfull!

I has always thought that in the PCX-NNNN NNNN was the max bandwidth already taking into account the channels. Always good to learn new stuff! So cleared up point one nicely...

The second point is slightly harder to swallow, but okay, since MemTest86 is able to work nicely enough on the newer mac I can accept the on the older ones have UEFI issues... Perhaps issues isn't the right word, perhaps non-standard customizations that aren't public documented and hence there is no way of working around those...

Your third point would make sense if I'd moved from a 1GB 1333MHz DIMM to a 4GB 1066MHz DIMM, but in actual fact I substituted the 1GB 1066MHz DIMMs I had for 4GB 1066MHz DIMMs. So it was PC3-8500 for PC3-8500 RAM, neither the frequency, nor the number of channels changed. I suppose it when from 1R to 2R, but would this be responsible for a 15% drop in Bandwidth even though the frequency (and #channels) is the same? And thanks for the link, very useful, explains why higher density RAM has lower frequency than lower density

Cleared up the last detail about memory info nicely.

I ran SiSoft Sandra on Windows 7 Bootcamp on the Mac Pro to verify the memory bandwith and it came up to 22 GB/s (so over 20000 MB/s). Therefore, I think it's fair to say that, while memtest continues to be an excellent tool to test memory, in the early 2009 MAC PRO's case, it just doesn't perform well in its UEFI environment. That can be attested by the fact that it took 36 hours to run all the tests (4 passes) while in my MacBookPro, to test 16GB of memory it took around 8 hours.

Announcement

Memtest86 Pro on Macs...

Memtest86 Pro on Macs...

Comment

Comment