Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

SpacemanSpiffed

macrumors regular
Mar 27, 2013
192
282
Pacific NW
I do like the 2018 model...but I do rather wish they had doubled the size of the case, and put in a larger cooler so that it can accommodate 95W TDP chips (without throttling), and a spare m2 slot. That would have been superb.

It can be done, though I'd recommend moving to the 45w TDP mobile chips, as everyone testing it seems to show the current Mini thermal throttling a lot under load. This would give it head room to offer the I9 or Xenon E 8-core CPUs in the current package.

Seriously, we're living in a time of exploding core-counts thanks to AMD (32 core Threadripper with 128MB L3 and a single NUMA Domain??? Be still my beating heart). For the Mini to stay near the top of the heap, it should offer an 8-core/16-thread model that's not thermally hampered.

Anyway, back to Spectrum's quote... and case in point for your desire for dual M.2 slots...

I just got a Zotac EN72070V.

It has the i7-9750H - a 45w TDP chip, nearly as powerful as the 65W i7-8700B in the top mini: Both are 6/12 cores/thread, but the 9750H is 2.6/4.5Ghz to the Mini's 3.2/4.6Ghz. But it doesn't seem to thermal throttle as badly.

Where it beats the mini is component access. No tools needed. 2x SO-DIMM slots, takes 64GB (confirmed personally, not properly documented), and has 2x M.2 slots - one it obstinately for an Optane module, but I've gotten the BIOS to recognize dual HP EX 950 M.2 Drives in it, but it doesn't retain that info correctly across reboots. Based on a number of things elsewhere, it should be able to do so with just a BIOS update. I've put in a request through Zotac support and sent them a video of it showing it recognizing the dual M.2 drives. Hopefully that comes in the next BIOS update. It also has a 2.5" drive slot.

It's about twice the volume of the Mini 2.65l vs 1.36L ( 0.57 inches wider, 0.29 inches deeper, 1.05 inches taller. External Power Supply though.

Well that sucks.. twice the volume!?!? External Power supply?!? What the bleep is it doing in there??

OH YEAH... It packs an 8GB nVidia RTX 2070 in there.

Click spoiler for a visual size comparison:
j638RTp.jpg


2012 mini sitting on top of the Zotac's predecessor (same dimensions).

Relative to the mini:
Pros: SD Card reader, Microphone Jack, 6x USB 3.0/3.1, 4x Simultaneous Display @ 4K, Dual Ethernet, Cheaper
Cons: No Thunderbolt. No 10G Ethernet (only 2.5G), No OSX, External Power supply (could be a pro)

Cost:
Zotac w/ 64GB, 2TB M.2 SSD, OS - $2,100
Mini w/ 64GB, 2TB SSD 10Gb - $3,200

tl;dr - Given that an RTX 2070 is a $500 video card, and the already > $1K price difference, there's room for Apple to improve the Mini.

[automerge]1571727190[/automerge]
A 50% increase in size would make the Mac Mini less mini, and less transportable.

Although the Mac Mini is a desktop, it is small enough to be easily transportable in a backpack. That is one of the features that attracted me, and maybe others, to the Mini.

When I got mine, I did not want a portable computer to tote everywhere, every day. I did want something that could be easily transported occasionally locally on a bicycle, or when relocating by public transport. The Mac Mini can be disconnected in a minute or so, taken some place and set up as quickly, with locally available peripherals. As a teacher I have taken my Mini from home to class from time to time, connecting it to an LCD projector and audio system in class.

FWIW: I've been taking a Zotac Magnus, like the one EN72070 I just got done posting about, back and forth between Seattle, Phoenix, and Houston via plane in my carry-on backpack (no way in hell will I let it be checked). Along with my 13" MacBook Pro in the same backpack, and bunch of other stuff. Much bigger and you got a problem, but ~3 liters and under seems be the sweet zone for a super portable "desktop" .
 
Last edited:
  • Like
Reactions: Neodym and Spectrum

Neodym

macrumors 68020
Jul 5, 2002
2,429
1,065
Re: Installing macOS / OS X

Apple has re-released many Installers due to the old Installers' expiring security certificates (Link). Make sure to download the latest versions from Apple, as old ones in a personal repository probably won't work anymore End of October (will be reported "damaged").
 
  • Like
Reactions: Synchro3

albusseverus

macrumors 6502a
Nov 28, 2007
744
154
There's no sign Apple has rid itself of the 'pay today's prices for years old technology' mindset. Apple's apparent renewed interest in Mac doesn't extend to updating the Mac line at least yearly. MacBook Pro has had updates, but iMac Pro, for example, languishes. iMac Pro was going to be a flagship product when Mac Pro appeared to be cancelled. Not that the flagship Mac Pro of old didn't suffer from outdated tech at Apple-style prices.

Where does that leave Mac mini? Where it always was, it seems. Updates every 5 years whether we want them or not.

Mini is a specialised product, for users that don't need GPU/can't afford a Mac with GPU. Those that don't need, like developers, probably have budget for such a thing. Those that can't afford a Mac with a GPU, are welcomed to the new Apple pricing strategy… just keep putting the price up until buyers stop buying (iPhone XS style). Maybe we'll get the iPhone 11 of Minis??

Naaaa, I'm dreaming! Apple might do that for a money spinner like iPhone. Mac Mini will see the usual neglect. Apple's stuck it to us for price and we're just going to have to lump it.
 

Spectrum

macrumors 68000
Mar 23, 2005
1,799
1,112
Never quite sure
It can be done, though I'd recommend moving to the 45w TDP mobile chips, as everyone testing it seems to show the current Mini thermal throttling a lot under load. This would give it head room to offer the I9 or Xenon E 8-core CPUs in the current package.
I think this misunderstands the issue. Current cooling in the Mac mini is good for around 60W stable. It makes no difference which CPU is inside.
The CPU TDP merely specifies the thermal load at rated *base* frequency all core load.
A 45W mobile 8 core i9 CPU will be just as thermally restricted in the Mac mini due to the ~60W thermals.
 
  • Like
Reactions: SpacemanSpiffed

SpacemanSpiffed

macrumors regular
Mar 27, 2013
192
282
Pacific NW
I think this misunderstands the issue. Current cooling in the Mac mini is good for around 60W stable. It makes no difference which CPU is inside.
The CPU TDP merely specifies the thermal load at rated *base* frequency all core load.
A 45W mobile 8 core i9 CPU will be just as thermally restricted in the Mac mini due to the ~60W thermals.

You're right that TDP hasn't been measured that way for a number of generations now, and a complete whiff on my part when I was posting. I have no problem being corrected with facts. :)

Though I still stand by my assertion that an 8-core part, say the i9-9900, should be considered, especially given the audience that still uses Minis. As I said, AMD has changed the PC landscape, and many people I know who do things like video rendering, software dev, etc, are all saying "give me the cores".

Apple could either switch to the HK parts and drop the peaks speed back a little bit to acknowledged that the current cooling in the Mini is just a little short of handling the chips at their max, or redesign the cooling system and allow for peak turbo and temps that equal larger PCs with the same CPUs.
 

Krevnik

macrumors 601
Sep 8, 2003
4,100
1,308
Though I still stand by my assertion that an 8-core part, say the i9-9900, should be considered, especially given the audience that still uses Minis. As I said, AMD has changed the PC landscape, and many people I know who do things like video rendering, software dev, etc, are all saying "give me the cores".

Apple could either switch to the HK parts and drop the peaks speed back a little bit to acknowledged that the current cooling in the Mini is just a little short of handling the chips at their max, or redesign the cooling system and allow for peak turbo and temps that equal larger PCs with the same CPUs.

To get 8 cores on Intel, their best bet is the 9700. The 9900HK would be slower, overall. But the desktop 9900 would be throttled even with boost disabled. And that monster can easily wind up drawing in the 180W range while it is boosting. So really, the most performant parts that will work here that are currently available are the 9700 from Intel or the 3700X from AMD.

My own measurements of the Mini suggest the thermal limit is around 80W, based on what I can get out of the Mini power wise while temps and clock speed remain stable (but high). I forget what the PSU is rated for, but it’s not that high, IIRC.

But let’s say we beefed up the PSU and cooling to handle a higher load. Do we spend that on more CPU, or would it make sense to spend it on GPU instead? *shrug*
 

Spectrum

macrumors 68000
Mar 23, 2005
1,799
1,112
Never quite sure
To get 8 cores on Intel, their best bet is the 9700. The 9900HK would be slower, overall. But the desktop 9900 would be throttled even with boost disabled. And that monster can easily wind up drawing in the 180W range while it is boosting. So really, the most performant parts that will work here that are currently available are the 9700 from Intel or the 3700X from AMD.

My own measurements of the Mini suggest the thermal limit is around 80W, based on what I can get out of the Mini power wise while temps and clock speed remain stable (but high). I forget what the PSU is rated for, but it’s not that high, IIRC.

But let’s say we beefed up the PSU and cooling to handle a higher load. Do we spend that on more CPU, or would it make sense to spend it on GPU instead? *shrug*
interesting...I am basing ~60W on the stablised stats reported by intel power gadget. Initially (lasting ~5-10 seconds or so) I see power draw up to ~80W, but it can't be maintained due to the temperature rapidly hitting 100C. Clocks then drop down to 3.8-4.3, depending on workload, and a steady ~60-65W, 99C.

I can't really see how an 8 core CPU would perform any differently in the mini without a better cooler (and\or PSU) unless the CPU is dramatically more power efficient. Load up 8 cores, and I don't see anything other than the clock speeds dropping lower than the 6 core CPU in order to maintain the steady 65W max Across 8 cores.

Happy to be proved differently though!
 
  • Like
Reactions: SpacemanSpiffed

Krevnik

macrumors 601
Sep 8, 2003
4,100
1,308
interesting...I am basing ~60W on the stablised stats reported by intel power gadget. Initially (lasting ~5-10 seconds or so) I see power draw up to ~80W, but it can't be maintained due to the temperature rapidly hitting 100C. Clocks then drop down to 3.8-4.3, depending on workload, and a steady ~60-65W, 99C.

I can't really see how an 8 core CPU would perform any differently in the mini without a better cooler (and\or PSU) unless the CPU is dramatically more power efficient. Load up 8 cores, and I don't see anything other than the clock speeds dropping lower than the 6 core CPU in order to maintain the steady 65W max Across 8 cores.

Happy to be proved differently though!

Binning is fun so it’s hard for our examples to really prove/disprove what’s going on. Add in differences in efficiency moving heat to the heat sink and gets harder. My estimate for ~80W is somewhat idealized, assuming the heat is efficiently being moved to the heat sink, and that ambient temperature isn’t a factor. But if you are getting 3.8+ from your Mini under multi-core workloads you are already doing well. Doubly so if you are in the TDP envelope while doing it.

I’m using the 8500, but I tend to sit at 3.8Ghz all core and 95C doing it. But I don’t need this sort of workload frequently, which is why I didn’t opt for the 8700. But I do sit above TDP doing this.

The 9700 is roughly similar to the 8700 in actual consumption. It will have a similar issue as the 8700 that it won’t be able to fully boost all the time, but it should still at least be similar to the HK at the end of the day, if not beating it outright. But the 9700 (3.0Ghz) base clock is lower than the 8700 (3.2Ghz), which is partly how it stays in the TDP envelope.

I could see a good argument for going with the HK to leverage the headroom it has to also include a mobile dGPU though. It can boost a bit higher (better binning), but I haven’t played with the 2019 MBP to have an idea what the actual power usage looks like.
 

Krevnik

macrumors 601
Sep 8, 2003
4,100
1,308
interesting...I am basing ~60W on the stablised stats reported by intel power gadget. Initially (lasting ~5-10 seconds or so) I see power draw up to ~80W, but it can't be maintained due to the temperature rapidly hitting 100C. Clocks then drop down to 3.8-4.3, depending on workload, and a steady ~60-65W, 99C.

I've been playing around and comparing a new 16" MBP to the Mini to see if it can be a good desktop replacement. I got reminded of this thread because I was looking at thermals/clocks to get a better idea why the gap in performance was smaller than I expected (i5 Mini vs i9 MBP). But this is why I made the comment I did. Temps holding steady at 95C under boost, right at about 80W for the package. ~3.7Ghz all clock during compiling, which isn't too bad. But this is the i5 without HT.

But your chip can hit over 4Ghz which mine never really has. I pretty much never see the 4.1Ghz boost in my work. 3.9 is about as good as it gets.

So I'm wondering what's going on here, because if anything, my CPU should be more wasteful with power, and producing more heat at 80W than yours at 60-65W. I'm confused.
Screen Shot 2019-11-20 at 8.17.38 PM.png
 

Spectrum

macrumors 68000
Mar 23, 2005
1,799
1,112
Never quite sure
I've been playing around and comparing a new 16" MBP to the Mini to see if it can be a good desktop replacement. I got reminded of this thread because I was looking at thermals/clocks to get a better idea why the gap in performance was smaller than I expected (i5 Mini vs i9 MBP). But this is why I made the comment I did. Temps holding steady at 95C under boost, right at about 80W for the package. ~3.7Ghz all clock during compiling, which isn't too bad. But this is the i5 without HT.

But your chip can hit over 4Ghz which mine never really has. I pretty much never see the 4.1Ghz boost in my work. 3.9 is about as good as it gets.

So I'm wondering what's going on here, because if anything, my CPU should be more wasteful with power, and producing more heat at 80W than yours at 60-65W. I'm confused.
View attachment 878340
Interesting...are they the results for the i5 mini? Compared to what I've seen with the i7, I'm rather amazed that it maintains a stable 80W power draw...
What workload are you running for this test?
Or is this the new 16 inch MBPro? (If so, which model?) It can't be since the horizontal TDP line is at 65W.
 
Last edited:

Spectrum

macrumors 68000
Mar 23, 2005
1,799
1,112
Never quite sure
I just re-ran 12 instances of: "yes > /dev/null &" and it is clear that my i7 mini peaks at 85 Watts, but rapidly stabilises to ~60W, pulling 3.5 Ghz for a 12-thread workload. It simply doesn't have the cooling capacity (it seems) to permit 80Watts for more than a few seconds. I don't understand how your machine manages this. Better thermal paste lottery?

If mine were able to stay up at 80W, the stable frequency would be ~4.0 Ghz for this 12-thread load.

1574333150879.png


Abd throwing a 6-core workload at it (and using Instruments to disable multi-threading), provides almost the same result - maxed out at 60Watts, but slightly higher at 3.7-3.8 Ghz:
1574333733059.png


What's really weird about this, is that my i7 is effectively providing the same performance as the i5 (6-cores at 3.7 Ghz), but at a lower Wattage (greater efficiency). But not actually any improved performance due to being thermally capped at 60 Watts rather than at 80 Watts! (like yours). I think it is very strange to see such a large stable Wattage difference, since both have the same cooling system.

Unless there is a bug in the Intel Power Gadget? I notice that you are using an older version. I have 3.6.1, and there is a bug in how it reports the CPU graph: it shows half what it should (see charts).
 
Last edited:

Spectrum

macrumors 68000
Mar 23, 2005
1,799
1,112
Never quite sure
Finally...I can get slightly better performance by turning the mini upside down and removign the bottom cover :)

Now around 70 Watts and 6-threads at 4.0 Ghz.

1574335595965.png
 

D.T.

macrumors G4
Sep 15, 2011
11,050
12,460
Vilano Beach, FL
Finally...I can get slightly better performance by turning the mini upside down and removign the bottom cover :)

Now around 70 Watts and 6-threads at 4.0 Ghz.

View attachment 878383


When I get a moment for a reboot, I'll add some data to this, I did a fresh reinstall of the IPG app (requiring a reboot), and I've got a big BB fan, but I'm going to also remove my bottom cover.
 

Krevnik

macrumors 601
Sep 8, 2003
4,100
1,308
Interesting...are they the results for the i5 mini? Compared to what I've seen with the i7, I'm rather amazed that it maintains a stable 80W power draw...
What workload are you running for this test?
Or is this the new 16 inch MBPro? (If so, which model?) It can't be since the horizontal TDP line is at 65W.

Yes, it’s the i5. The i7 should really just be a better binned version of the same chip. So your results with HT disabled are weird.

In my case, my tests were running clean builds of Vapor using Swift in release mode. It reflects my actual work load. But it’s very to what I see running something like Handbrake for 20+ minutes, so it reaches equilibrium pretty fast. That it stabilizes at 95C means it’s not thermally limited either, so the cooling is able to handle 80W, assuming the readings are right.

I didn’t realize there was newer version available. I was wondering why the MBP had a different display. I’ll try it again, just to standardize the results.

For airflow, my Mini sits on top of a blu-ray drive. But that shouldn’t make much of a difference. But that does seem like the simplest explanation maybe. Wonder how airflow can get choked with these things.



The MBP thermally throttles more, is louder, and isn’t that much faster than the i5 Mac Mini because it drops down closer to base clocks (boosting at 2.6Ghz or so) after the initial burst. It’s still double digits, but smaller than the Geekbench scores of the 15” (which is slower than the 16”) suggest. And I suspect it’s because 3.7Ghz all-core, all-day for 6 cores is nothing to sneeze at, even against 2 extra (slower) cores.

Honestly, I can see it being a desktop replacement for many, but the noise is still too much for me. The fans also stay spun up for a good minute after my build is finished, which isn’t great. The introduction of turbo boost has really made thermal management more complicated for these SFF computers and laptops. Especially when you have 45W chips spiking up into the 80+W range. I’m still tempted to have one to replace a 13” MBP, but I might wait to see if the changes trickle down to the 13” too (new keyboard wins), or just go with the base model 16” instead of something meant to replace the Mini.

The Mini + eGPU is simply surprising at how good it is in real world, even now. The i7 in the Mini I bet would still equal or pull ahead of the 9980HK in the 16” MBP for real benchmarks. But the HK might do better if given the Mini’s cooling. But then the 9700 might be even better still. Hard to say. These chips are all very similar in terms of real world performance.
 

Spectrum

macrumors 68000
Mar 23, 2005
1,799
1,112
Never quite sure
Yes, it’s the i5. The i7 should really just be a better binned version of the same chip. So your results with HT disabled are weird.
Well, in many ways the i7 demonstrating that it is a better performing chip: I am getting the same CPU speed as your i5, but at lower wattage...

The issue is that your Mini is happy to boost up to a higher Wattage stably, whereas mine is being forced to keep wattage at ~60W, which is giving a stable CPU performance result that is very similar to your i5 at 80W.

It basically seems like the i7 is better/more efficient than the i5 (and can in theory boost higher too), but that my mini has a crappy cooling subsystem (relatively). The fact that I gained ~10 Watts of headroom by removing the bottom covers is also telling. You might want to try it! :)

May I ask what the ambient temperature is where your mini is?
 

Krevnik

macrumors 601
Sep 8, 2003
4,100
1,308
Well, in many ways the i7 demonstrating that it is a better performing chip: I am getting the same CPU speed as your i5, but at lower wattage...

The issue is that your Mini is happy to boost up to a higher Wattage stably, whereas mine is being forced to keep wattage at ~60W, which is giving a stable CPU performance result that is very similar to your i5 at 80W.

Yeah, it's more that the amount of heat measured from the package is slamming up against the limits at almost 20W less, which is what I find weird. The heatsink and fan assembly are clearly designed to handle it, so maybe it's an issue of getting the heat into the heatsink.

I've seen reports that de-lidding the CPU helps in the self-build community due to variation in how the lid is applied to the die. And there's always variation in thermal paste between the "lid" and the heatsink. My guess is one of the two is probably your likely culprit.

It basically seems like the i7 is better/more efficient than the i5 (and can in theory boost higher too), but that my mini has a crappy cooling subsystem (relatively). The fact that I gained ~10 Watts of headroom by removing the bottom covers is also telling. You might want to try it! :)

May I ask what the ambient temperature is where your mini is?

My ambient temps tend to be about 20C during the day.

I did get a bit of a boost from removing the bottom. Instead of waffling between 3.6-3.7Ghz running x265 in Handbrake. I now waffle between 3.7-3.8Ghz. And power consumption now jumps between ~78-90W, which is about 10W higher than before. So temps are at play, but the general perf gain is "meh". It pretty much has no impact on build times, and less than a frame a second in Handbrake.

And yes, the airflow design of the Mini isn't the best. The opening for intake is fairly small. But the impact in real-world doesn't seem to be huge. 2% extra clock speed at the cost of 12% more power is what I'm missing out on, for example.

If you really wanted to get this all the time, you could grab something like this: https://forums.macrumors.com/threads/mac-mini-custom-bottom-plate-decreased-temps.1953828

The only issue I see is that the original design assumes the twist design of the pre-2014 models. Not 100% sure it'll fit the 2018 very well as-is.
 
  • Like
Reactions: Spectrum

Spectrum

macrumors 68000
Mar 23, 2005
1,799
1,112
Never quite sure
Yeah, it's more that the amount of heat measured from the package is slamming up against the limits at almost 20W less, which is what I find weird. The heatsink and fan assembly are clearly designed to handle it, so maybe it's an issue of getting the heat into the heatsink.

I've seen reports that de-lidding the CPU helps in the self-build community due to variation in how the lid is applied to the die. And there's always variation in thermal paste between the "lid" and the heatsink. My guess is one of the two is probably your likely culprit.



My ambient temps tend to be about 20C during the day.

I did get a bit of a boost from removing the bottom. Instead of waffling between 3.6-3.7Ghz running x265 in Handbrake. I now waffle between 3.7-3.8Ghz. And power consumption now jumps between ~78-90W, which is about 10W higher than before. So temps are at play, but the general perf gain is "meh". It pretty much has no impact on build times, and less than a frame a second in Handbrake.

And yes, the airflow design of the Mini isn't the best. The opening for intake is fairly small. But the impact in real-world doesn't seem to be huge. 2% extra clock speed at the cost of 12% more power is what I'm missing out on, for example.

If you really wanted to get this all the time, you could grab something like this: https://forums.macrumors.com/threads/mac-mini-custom-bottom-plate-decreased-temps.1953828

The only issue I see is that the original design assumes the twist design of the pre-2014 models. Not 100% sure it'll fit the 2018 very well as-is.
I am astounded that your mini can handle 90Watts fairly stably. That is incredible compared to mine.
I am tempted to look at applying new thermal paste, but quite worried about permanently damaging something...

Edit: I suppose another option is that you or I have a dodgy temp sensor which is allowing the CPU to use more or less power than it should be...not sure how that can be tested.
 
Last edited:

Krevnik

macrumors 601
Sep 8, 2003
4,100
1,308
I am astounded that your mini can handle 90Watts fairly stably. That is incredible compared to mine.
I am tempted to look at applying new thermal paste, but quite worried about permanently damaging something...

Edit: I suppose another option is that you or I have a dodgy temp sensor which is allowing the CPU to use more or less power than it should be...not sure how that can be tested.

Yeah, when adding RAM to mine, I screwed up and broke the LED connection. And I'm just not good enough with my soldering skill to fix anything that delicate. 1990s surface mount is about as good as it gets for me. Finding the part needed was easy enough, I just can't fix it without taking it to someone.

Oh there's no way the sensors report the same temp. The dE on these things are a couple degrees, which is why the design limit is 105C before shutdown, while the throttle limit is 100C, to account for that.

That said, my testing is a bit different. All my measurements are using stuff like Handbrake and Xcode because I use them in my workflow. It's possible that my load changes the profile of heat vs power consumption.
 

Spectrum

macrumors 68000
Mar 23, 2005
1,799
1,112
Never quite sure
That said, my testing is a bit different. All my measurements are using stuff like Handbrake and Xcode because I use them in my workflow. It's possible that my load changes the profile of heat vs power consumption.
Would you mind trying out the "yes > /dev/null & " test?
Set off as many instances as you wish to test, then stop with: killall yes
 

Krevnik

macrumors 601
Sep 8, 2003
4,100
1,308
Would you mind trying out the "yes > /dev/null & " test?

On a lark, I tried it and I get 65-75W consumption and ~3.8Ghz. Temp bounces around 95-100C.

I usually don't do this to load a CPU, because it is completely artificial, and ironically in this case, doesn't even draw as much power as a real workload.
 

Spectrum

macrumors 68000
Mar 23, 2005
1,799
1,112
Never quite sure
On a lark, I tried it and I get 65-75W consumption and ~3.8Ghz. Temp bounces around 95-100C.

I usually don't do this to load a CPU, because it is completely artificial, and ironically in this case, doesn't even draw as much power as a real workload.
Intriguing...that suggests that my i7 will perform even worse than your i5 given the same real workload...even if I limit it to 6 cores. Something isn't right...
 

Spectrum

macrumors 68000
Mar 23, 2005
1,799
1,112
Never quite sure
On a lark, I tried it and I get 65-75W consumption and ~3.8Ghz. Temp bounces around 95-100C.

I usually don't do this to load a CPU, because it is completely artificial, and ironically in this case, doesn't even draw as much power as a real workload.
Still wondering about this, I checked CPU temps under load. It appears that CPU Core 2 is consistently reporting it to be ~10˚C hotter than all the other cores. I wonder if this is what is triggering the CPU throttling on my machine?
1574771781971.png
 

Krevnik

macrumors 601
Sep 8, 2003
4,100
1,308
Possible, and it aligns with the behavior you are seeing. Not all cores are created equal, and some are better than others on the same die at boosting. AMD feeds this information into its boost algorithm so it favors some cores over others.

It could also be that Intel‘s algorithm is favoring this core? Not sure. But loading down all cores is supposed to be the worst case scenario for boost as well.

Turbo boost is a weird thing, TBH. Base clocks are picked based on what all chips in a bin can hit. But with the push for more performance, building in an overclock algorithm is one way to improve performance. But not all cores can overclock to the same level, or do it at the same power consumption.

So for the sake of higher performance, we are getting more variability between what different copies of a chip can pull off under boost. It’s very possible we’re comparing two copies on different points of the bell curve here.
 
  • Like
Reactions: Spectrum

Spectrum

macrumors 68000
Mar 23, 2005
1,799
1,112
Never quite sure
Yeah. Something like that. I suspect the multicore workload is requesting all cores to boost as much as possible, but Core number 2 is less efficient at doing so, needing more power for its highest speed. And therefore it gets hotter before the others. This then triggers a clock speed reduction across all the other cores in order to keep core number 2 below an acceptable temperature.

Next time you are doing a render, I'd be interested in seeing a screenshot of your individual core temps.
I predict they might be more homogenously close to 100C. Thus, overall more power draw and CPU speed.

Possible, and it aligns with the behavior you are seeing. Not all cores are created equal, and some are better than others on the same die at boosting. AMD feeds this information into its boost algorithm so it favors some cores over others.

It could also be that Intel‘s algorithm is favoring this core? Not sure. But loading down all cores is supposed to be the worst case scenario for boost as well.

Turbo boost is a weird thing, TBH. Base clocks are picked based on what all chips in a bin can hit. But with the push for more performance, building in an overclock algorithm is one way to improve performance. But not all cores can overclock to the same level, or do it at the same power consumption.

So for the sake of higher performance, we are getting more variability between what different copies of a chip can pull off under boost. It’s very possible we’re comparing two copies on different points of the bell curve here.
 

Krevnik

macrumors 601
Sep 8, 2003
4,100
1,308
In general, I see wider variation than you: 86-96C when encoding, with most cores in the 90s, and the 'yes' test is 87-97C, with 3 cores in the 90s, and 3 in the 80s.
 
  • Like
Reactions: Spectrum
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.