May be in that particular orientation, the chips on the 4th stick that facing the CPU heatsink really hit the throttling temperature.
My setup only has 3 DIMM, so, the cooling won't be affected that much by the heat sink.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Remove rear case pci holder to dissipate better heat?
- Thread starter Cuzzupino
- Start date
- Sort by reaction score
May be in that particular orientation, the chips on the 4th stick that facing the CPU heatsink really hit the throttling temperature.
My setup only has 3 DIMM, so, the cooling won't be affected that much by the heat sink.
The bench Mac Pro has 4 sticks and the office Mac Pro has 3 sticks. With the slowdown issue hitting both systems.
4GB/s throughput seems to be the ceiling of DDR 3 in a well cooled Mac. I’ll have to look up what the max speed is rated at.
Here is a shot with the slot covers installed. I just rearranged slots 1 & 2, showing the air flow friendly slot covers.
4GB/s throughput seems to be the ceiling of DDR 3 in a well cooled Mac. I’ll have to look up what the max speed is rated at.
Here is a shot with the slot covers installed. I just rearranged slots 1 & 2, showing the air flow friendly slot covers.
Last edited:
Hi Handheld,
It looks like you have a similar machine like I do. If you tell me which heat test you used for your benchmark to test your ram modules, I can do that test as well to compare values. My ram modules have a very thin heat spreader on it.
However I think you are lucky, because the optional available heat spreader available for the PC world are really great an a lot better. They are almost as tall as the heatsink. There is also an active fan available as well. You might want to consider that if it turns out that heat actually throttles down the performance.
My ram, as well as @h9826790's does not have heat spreaders. Thanks for the standard and extreme cooling ideas. I'm very pleased with cooling performance of the intake and exhaust fans with the side on well enough to forego heat spreaders.
I've been working with Quickbench's numeric extended 20mb-100mb file transfer test on 2 stripe and 3 stripe raid ram disks.
Micron > Samsung
I read technical documents. Power consumption difference is significant.
Micron > Hynix > Samsung
I read technical documents. Power consumption difference is significant.
Micron > Hynix > Samsung
Interesting to hear the Samsungs are power hungry.
Just to be clear, this is the part# for my Samsung rdimms.
16GB 2Rx4 PC3L 10600R ECC REG 1.35V Samsung M393B2G70QH0-YH9Q8 HP 647653-181
Sharing legacy x58 PC Ramdisk benchmarks from the net. It gives a good idea of what should be possible on our x58 based Mac.
http://fiehnlab.ucdavis.edu/downloads/staff/kind/Collector/Benchmark/RamDisk/ramdisk-benchmarks.pdf
A free copy of Amorpheous Disk Mark may also expose slowdowns encountered in QuickBench.
Last edited:
Found it:
https://www.infoq.com/articles/power-consumption-servers
It mustn't be a problem with your DIMM's either though. cMP has way better cooling than usual servers. These Samsung sticks are used in those servers without a problem.
Edit: I suppose you're going after a very small amount of difference. Eg. the first Geekbench result has always been the best for me on a Mac Pro. This is because Apple loves silent computers. This is how Mac Pro is designed. Fans won't ramp up before CPUs gets hotter than about 95 celcius. I think there's no need to worry about, since 99% of cMPs are still working like Swiss watches.
https://www.infoq.com/articles/power-consumption-servers
It mustn't be a problem with your DIMM's either though. cMP has way better cooling than usual servers. These Samsung sticks are used in those servers without a problem.
Edit: I suppose you're going after a very small amount of difference. Eg. the first Geekbench result has always been the best for me on a Mac Pro. This is because Apple loves silent computers. This is how Mac Pro is designed. Fans won't ramp up before CPUs gets hotter than about 95 celcius. I think there's no need to worry about, since 99% of cMPs are still working like Swiss watches.
Last edited:
No worry.. Just having fun SQUEEZING every last bit of performance out of the BEST MAC I've ever owned. 9+ years and counting. No buyers remorse for buying it or skipping the trash can. Although I'd like to buy the shiny lid from the 6,1 to cover a tissue box on my desk.
EDIT: More research on the net reveals the evils of DDR3 thermal throttling. While most references to it lack source material, I've found a college level research paper on the subject. Confirming that RAM thermal management can result in as much as a 15.8% improvement in performance, 4.1% on average.
http://users.eecs.northwestern.edu/~hardav/papers/2011-HPCA-DRAMThermal-Liu.pdf
Last edited:
That's some serious difference. Anything above 5% is worth wasting time. I'll read the document.
Another great document from the same search. This time from intel, discussing ddr3 thermal throttling mechanisms on a variant of the cMP chipset. Rather than the effects of thermal throttling in a research paper, Intel discusses the technology at the base of the research observations.
https://www.manualslib.com/manual/397357/Intel-Bb5520ur.html?page=82
Intel BB5520UR Technical Product Specification: Memory Open And Closed Loop Thermal Throttling
Product specification. Platform Management 4.3.1
Memory Open and Closed Loop Thermal Throttling
Open-Loop Thermal Throttling (OLTT)
Throttling is a solution to cool the DIMMs by reducing memory traffic allowed on the memory
bus, which reduces power consumption and thermal output. With OLTT, the system throttles in
response to memory bandwidth demands instead of actual memory temperature. Since there is
no direct temperature feedback from the DDR3 DIMMs, the throttling behavior is preset rather
than conservatively based on the worst cooling conditions (for example, high inlet temperature
and low fan speeds). Additionally, the fans that provide cooling to the memory region are also
set to conservative settings (for example, higher minimal fan speed). OLTT produces a slightly
louder system than CLTT because minimal fan speeds must be set high enough to support any
DDR3 DIMMs in the worst memory cooling conditions.
Closed-Loop Thermal Throttling (CLTT)
CLTT works by throttling the DDR3 DIMMs response directly to memory temperature via
thermal sensors integrated on the Serial Presence Detect (SPD) of the DDR3 DIMMs. This is
the preferred throttling method because this approach lowers limitations on both memory ower
and thermal threshold, therefore minimizing throttling impact on memory performance. This
reduces the utilization of high fan speeds because CLTT does not have to accommodate for the
worst memory cooling conditions; with a higher thermal threshold, CLTT enables memory
performance to achieve optimal levels.
I think it's important to consider the environmental state when these were designed. The earlier designs - the G5 in the plastic case - were small wind tunnels. At least they sounded like them. Apple was finally able to react to user complaints, and came out with the big aluminum case, and they refined the cooling capabilities over the next few years. Simultaneously, Intel was discovering that that their CPUs were small space heaters as they marched toward higher frequencies and greater densities. I was working with one of their OEMs at the time, and Intel made a big deal about reducing power consumption/increasing efficiency. Many of the components in the 4/5,1 were designed to go in 1U and 2U systems. They were able to accomplish this, with server-grade reliability, by putting fans everywhere inside the case. It was hard to hold a conversation next to just a single 1U system.
So Apple was using server grade components, and trying to cool them without too much noise.
I just read the Intel document about the Intel 5520 chipset
https://www.intel.com/content/dam/w...atasheets/5520-5500-chipset-ioh-datasheet.pdf
It seems the thermal throttling start at 80C on this chip.
I am wondering is this is real root cause of the performance penalty. The Apple default fan profile obviously will allow the NB go above 80C. And when you increase the fan speed, the NB can easily cool down for a few degrees.
When you stress the RAM disk, the NB will heat up as well. If it will start to throttle itself at 80C, it can also explain why your RAM drive is suffered from thermal issue, but mine doesn't. Because my single processor cMP using the X58 chipset, which doesn't has thermal throttling ability.
May I know if you monitored the NB's temperature as well (when you perform this RAM drive performance test).
[doublepost=1537398918][/doublepost]And I am thinking. If that's real RAM thermal throttling. GeekBench should able to see the difference as well.
If only RAM drive performance is affect, but not RAM performance is generally affected. Then may be the performance difference is actually coming from the NB thermal throttling.