0

I am using an Asus P6T Deluxe v2 motherboard with a first gen i7 processor. I was running 6x2 GB of Kingston HyperX PC1600 CAS 9 RAM in matched triple channel sets and had recently been experiencing my desktop getting much slower. Eventually, two sticks of the RAM (on two different channels, one on each bank) failed outright and stopped being recognized.

I decided to replace all 6 sticks with a new matched set of PC1600 CAS 9 RAM from Crucial (the only one model still available in matched sets) and was amazed by the level of improvement experienced. I did increase from 2GB DIMMs to 4GB DIMMs, (thus doubling my available RAM), but the performance improvement is seen even when total RAM utilization is below the 8GB point.

Improvements range from doubling to tripling of frame rates in games, large memory load times being about twice as fast and system boot and some process launches being 3 to 6 times faster.

Tests included frame rates in wow going from 30fps average to 97fps average, load times for Teamspeak going from 10+ seconds to under 2 seconds. System boot times dropping from around 40 seconds to around 20 seconds. Login to fully loaded desktop timings dropping from about a minute to about 30 seconds. Wow level load times dropping from about 20 seconds to about 10 seconds. 4K video render times in Davinci Resolve going twice as fast. SimC simulations running in about half the time they previously did for various different lengths of simulation (this one might be total ram related as I wasn't able to verify utilization over the entire runs), Warhammer 40k 3 load times from about a minute to about 10 seconds.

Unfortunately none of the timings are exact as I was not expecting the size or scope of improvement that occurred and didn't take exact measures prior to the change.

Even compared to prior to the failure of the two sticks I'm seeing 25 to 100% increases in performance for these tasks, even when not above a 12GB memory footprint.

What could have caused this? My understanding is that memory can't "get slower" with age, but then I'm not sure what could cause such a massive change in performance overall since I'm seeing huge performance differences even when not using more than 12gb or RAM on the system.

The old RAM was actually apparently not a triple kit, but rather 3 kits of KHX12800D3K2/4G. The new ram is CMZ24GX3M6A1600C9. There is a difference on the last part of the full CAS rating where the new RAM is faster by 3 cycles, so that may be a portion of it as well.

AJ Henderson
  • 515
  • 3
  • 11
  • "Even compared to prior to the failure of the two sticks I'm seeing 25 to 100% increases in performance for various tasks, even when not above a 12 GB memory footprint." - How are you measuring this improvement exactly? "What could have caused this?" - The only explanation would be the fact you doubled the amount of memory, and for whatever reason, your system was actually low on system memory. You don't provide any specifics about the usage. – Ramhound Jul 06 '17 at 16:00
  • "my best guess is that minor variations developed between the RAM that eventually degraded the triple channel performance because of loss of ability to maintain sync across the sticks." - You had 6 identical memory sticks, there was absolutely, no performance degradation due to you use 2 GB memory sticks. The only other explanation is that your memory was not originally working at their base frequency, and your new memory is, without being able to confirm that I cannot submit this theory as an answer. – Ramhound Jul 06 '17 at 16:00
  • I tested on various different loads from running WoW, to running video encodings, to running simulations. I'm measuring performance as framerates for wow and time to complete for discreet tasks (such as encodes, program launches, boot time, etc). I also checked system memory utilization at these times and the total utilization was below 8GB. – AJ Henderson Jul 06 '17 at 16:04
  • Just because sticks started out identical doesn't mean they stay identical. Kit pairs are shipped as such to minimize variation between the sticks. Over time, it would seem that deviation could occur that could compromise the level of precision needed for the memory controller to keep the triple channel functioning correctly. I did verify that when I first built the computer (8 years ago) the RAM was functioning correctly at it's intended frequency. I had assumed my recent performance slow downs were due to software issues until replacing the RAM produced an unexpectedly large boost. – AJ Henderson Jul 06 '17 at 16:07
  • Still no documentation or details on specific tests and their results....this can't be answered. –  Jul 06 '17 at 16:09
  • There is **absolutely** no difference, between the single memory stick kits and the multi-memory stick kits. Memory does NOT degrade over time, however, firmware settings can be reset to their defaults accidentally over a period of 8 years. "Over time, it would seem that deviation could occur that could compromise the level of precision needed for the memory controller to keep the triple channel functioning correctly." - This didn't happen, you replace the memory with functionally identical memory, and the same memory controller is being used. – Ramhound Jul 06 '17 at 16:10
  • Agree with Ramhound. I think the Kingstons were incompatible with the motherboard and thus were working below nominal. More detail on parts (Kingston HyperX PC1600 CAS 9 RAM is not enough) is required, else this question is unanswerable – AcePL Jul 06 '17 at 16:24
  • It sounds more likely that your processor was underclocking for some reason and the new memory made the BIOS do a recheck and reset to defaults and restore processor performance. Either that or the act of replacing the memory moved the heatsink and reseated it to give better contact with the processor to stop it throttling. Or your graphics card wasn't seated properly and may have been in PCIe x1 and got moved and reseated well enough to work at x16. Could have been a lot of things really. – Mokubai Jul 06 '17 at 16:25
  • The fact that it improved gaming performance, which is almost entirely graphics and CPU bound (memory is only really relevant at texture loading stage - before you see a picture basically) makes me believe it is unlikely to be a memory related problem. – Mokubai Jul 06 '17 at 16:29
  • @AcePL - they originally worked at a similar kind of level and I had verified it with performance tests back when it first went in, it's possible since other things in the machine have changed since then, but it doesn't seem like the main culprit. This restored performance to feeling more like the computer originally did prior to slow downs that started about a year ago. – AJ Henderson Jul 06 '17 at 16:32
  • @Mokubai - the bios recheck sounds likely. I had previously had a major thermal throttling issue due to a clogged cpu cooler (windows 10 killed my temp monitor, so I didn't get notified for a while until BIOS started complaining on a reboot). Clearing that had given me about a doubling in performance from removing the thermal throttling, but that was only a few weeks ago and only 4 or 5 reboots ago, so that might be something. It is, however, worth mentioning the FPS increase was only seen in WoW. Other games only showed load time improvements but not framerate. – AJ Henderson Jul 06 '17 at 16:34
  • Sounds like you had thermal throttling around, which would effect loading performance, the frequency of your memory unless it was significantly impacted would performance wise would be immeasurable. – Ramhound Jul 06 '17 at 16:49
  • Well WoW is highly CPU limited, especially in the major cities where it has to deal with a lot of player position management, so if no other games got improved performance then we roll back to the CPU. From an electronics standpoint a faulty link *can* degrade performance, but unless the RAM is ECC or Registered I would expect outright crashes rather than a simple degradation in performance. RAM simply doesn't have the protections that other protocol based busses have. – Mokubai Jul 06 '17 at 16:51
  • Comments are not for extended discussion; this conversation has been [moved to chat](http://chat.stackexchange.com/rooms/61706/discussion-on-question-by-aj-henderson-why-would-new-ram-run-substantially-faste). – Mokubai Jul 06 '17 at 16:56

1 Answers1

1

After discussion in chat, I'm fairly confident that this issue started as a CPU thermal throttling issue. At some point after the CPU throttling began (due to a clogged CPU cooler), two sticks of RAM failed which pushed the system out of being able to use triple channel memory bandwidth.

Since the throttled CPU couldn't keep up with the memory bandwidth anyway, this additional performance loss wasn't noted at the time, however, when I restored CPU cooling performance and removed the thermal throttle, the lack of full memory channel would have kept performance limited to a lesser extent.

Replacing the RAM then had the impact of restoring full memory bandwidth which would have produced most of the observed behaviors. The 5x improvement in TS load time is a bit strange still, but is an outlier to what is otherwise improvements on the x2 to x3 scale that would be consistent with moving from a failed single channel memory state back to functional triple channel performance. Unfortunately they don't seem to make the Kingston RAM anymore for me to be able to validate the theory, but it seems to explain everything pretty well.

AJ Henderson
  • 515
  • 3
  • 11