Bindee wrote:
PostPosted: Sat 05 Sep 2015, 04:20 Post subject:
I had the same with 16gb desktop of desktop ram , The first pass was fast but every pass after that got slower and slower until it was at a crawl.
But that's to be expected as you are electively DDOS'ing the ram and forcing continuous flipping of data storage and rewriting of the cells to try and force the capacitors to leak.
Hence why it's called hammering Razz....
Explanation of strange quote: I logged on to the forum, and wanted to quote the older post on Page 2 of the thread, but when I moved from page 1 to page 2 or from page 3 to page too, I was no longer "logged in", and so couldn't quote! I had to copy , then move to page three and paste.
Bindee wrote:Passmark recommend 4 passes of test #13 alone to be sure as it's not until it's at a crawl that some sticks start to leak , infact i'm not sure why they include it with the other test patterns instead of having it as a separate 4 run test on its own at the end of all the others as when you restart from #1 you are clearing the ram so you don't get the same hammering or slow down.
Temps wise it was not as extreme as your laptop but it was higher than normal , it started around the mid to high 30's and was in the mid to high 50's by the end of the test with the fan spinning up faster.
The heat doesn't do damage , the hammering or changing the refresh just drains the battery quicker because of the constant high power draw.
You reboot and clear the ram and the first pass will be quick again.
As I mentioned in my previous post, your conclusion that when running several passes of the whole memtest battery, RAM is cleared, and Row Hammer runs at full speed as at the start on the next pass.
In order to get a better perspective on this (since I'm not going to sit watching memtest run steadily for 14 hours with camera in hand to document the changing temps and speeds), I asked Kingston for the report of the 66 hour Row Hammer only series I was told was run on my last 16GB kit (the one that passed when I ran it for 14 hours), and it was kindly provided.
Sadly, it was a strictly text log (probably necessary because of the amount of disk space a regular report would have consumed - the log was some 300KB as it was). It also had no cpu temperature measurements.
But what it did record is the time it took for the Row Hammer test to complete each of 14 passes of the Row Hammer test alone (the fifteenth was aborted).
The interesting part is that it exactly replicates my own one-step speed change result using 8 passes of the entire memtest battery:
The speed halves after the first Row Hammer pass (within 1 minute), and then remains the same through all subsequent passes. I haven't received an explanation of this, and I don't have any idea what could cause it.
But the fact that it was consistent on two machines, one running the entire battery of tests for 8 passes, and the other running just the Row Hammer for 14 passes seems to me to rule out a
gradual, progressive slowing of the processing time.
I edited the log to show just the date/time stamps of the beginning of each pass, to make it more readable and smaller (still 56K, though).
I've tried to attach it as an odt and as a txt file, but the forum refuses both, so I'll try pasting it in.
I've kept the original log intact, if you'd care to have a copy by PM. My comments are in brackets [].
CPU speed: 2494.2MHz
L1 cache speed: 41135 MB/s
L2 cache speed: 12439 MB/s
L3 cache speed: 9408 MB/s
mem speed: 5559 MB/s
Memory latency: 65.684 ns
CPUSpeedTurboTheoreticalMax = 3092901
Available Memory: 0x3F3567000 (15GB)
Intel E3 Haswell-E chipset init
ECC polling disabled
This platform has 4 logical processors of which 4 are enabled.
TEST SESSION - 2015-08-28 14:36:06
2015-08-28 14:36:06 - CPU selection mode = 0
2015-08-28 14:36:06 - Starting pass #1
2015-08-28 14:36:06 - Running test #13 (Test 13 [Hammer test])
Enabling memory cache for test
2015-08-28 14:36:06 - Locking all memory ranges first...
2015-08-28 14:36:07 - Memory range locked: 0x100000000 - 0x41E500000 (1024KB of available memory left)
2015-08-28 17:04:58 - Finished Stage 2, Pattern=FFFFFFFF
2015-08-28 17:04:58 - Cleanup - Unlocking all memory ranges...
2015-08-28 17:04:58 - MtSupportRunAllTests - Test execution time: 8931.303
2015-08-28 17:04:58 - Starting pass #2
2015-08-28 17:04:58 - Running test #13 (Test 13 [Hammer test])
2015-08-28 21:50:42 - Starting pass #3
2015-08-29 02:36:26 - Starting pass #4
2015-08-29 07:22:10 - Starting pass #5
2015-08-29 12:07:55 - Starting pass #6
2015-08-29 16:53:39 - Starting pass #7
2015-08-29 21:39:24 - Starting pass #8
2015-08-30 02:25:08 - Starting pass #9
2015-08-30 07:10:52 - Starting pass #10
2015-08-30 11:56:35 - Starting pass #11
2015-08-30 16:42:19 - Starting pass #12
2015-08-30 21:28:01 - Starting pass #13
2015-08-31 02:13:45 - Starting pass #14
2015-08-31 06:59:30 - Starting pass #15
2015-08-31 08:51:51 - MtSupportRunAllTests - Test execution time: 6740.607
2015-08-31 08:51:51 - Test aborted
[Row Hammer test times for first (66 hour Row Hammer only) test session: Pass 1=149 min.,Pass 2=286 min., Pass 3=286 min., Pass 4=286 min.,Pass 5=285 min., Pass 6=286 min., Pass 7=285 min.,Pass 8=285 min., Pass 9=286 min., Pass 10=286 min., Pass 12=286 min., Pass 13=286 min, Pass 14=286 min., Pass 15 aborted after 112 min.]