2013 i5 3.4 vs i7 3.5 temperature for given load

[propower]

I know I'm crazy!

I bought the $1999 i5 model to test running pro audio (90% of what this machine will do). On this machine I typically run about a 25% total CPU load. It has handled everything I have thrown at it. I even used this CPU stress test http://osxdaily.com/2012/10/02/stress-test-mac-cpu/ to put an artificial 50% extra load on the machine. No problem plus temps stayed under 70degC - no extra fan noise.

So I ordered a BTO 3.5 i7. If I load two instances of the above test on each machine (=50% CPU for the i5 and =25% CPU for the i7) the i7 is running WAY hotter. The i7 CPU even got to 90degC and kicked the fan up to 1800rpm (no like). Although I could load much more on to the i7 the fans would be a total bummer (for me - many would't care).

My first wonder is if the i7 BTO didn't get attached to the heat sink correctly. I mean the i7 and i5 die are the same size and for the same load I was assuming they would have very similar thermal performance. Maybe I'm wrong though

First I will have to talk to Applecare - but unless there is something clearly wrong - and crazy as it is - I may just keep the i5 or BTO an i5 with an SSD.

FWIW: My main use is low track count at the lowest latency I can do (currently 96kHz/64buffer). Loading up huge sessions is not really my game and in ProTools about 30% to 40% CPU usage on the i5 is as high as I can imagine going.

 

[Zimmerman]

I will hopefully get my new iMac with an i7 tomorrow, I will try the same loadtest and report here.

 

[propower]

I took some time and ran the following using the terminal command listed in the first post and Temperature Gauge Pro (free demo available http://www.tunabellysoftware.com/tgpro/)

A single instance of: yes > /dev/null &
i7 CPU temp peak ~66degC (12.5% CPU load) Fan @ 1200 (minimum for imac)
i5 pk temp ~ 62degC (25% CPU load) Fan @ 1200

two instances of: yes > /dev/null &
i7 CPU temp peak ~90degC (25% CPU load) Fan @ 1200
i5 pk temp ~ 73degC (50% CPU load) Fan @ 1200

Three instances of: yes > /dev/null &
i7 CPU temp peak ~95+degC (38% CPU load) Fan @ 1900rpm
i5 pk temp ~ 81degC (76% CPU load) Fan @ 1200

Four instances of: yes > /dev/null &
i7 CPU temp peak ~99+degC (50% CPU load) Fan @ 2640rpm (max for imac)
i5 pk temp ~ 89degC (100% CPU load) Fan @ 1200
Note: I didn't let this last one go for long (5 minutes or so). The i5 would probably have gotten over 90 after a while (10 to 20 minutes)...

I did not expect this result - same load, same size CPU die and same thermal management system. I expected the results to be the same! THe i5 does great but the i7 heats up very quickly. It would be great if someone (or more) could run these same tests to see if maybe I got a bad heatsink to cpu attach on the i7 or if this is consistent for others. Takes ~20 minutes to do the four tests.

THANKS!

 

[Nirurin]

I took some time and ran the following using the terminal command listed in the first post and Temperature Gauge Pro (free demo available http://www.tunabellysoftware.com/tgpro/)

A single instance of: yes > /dev/null &
i7 CPU temp peak ~66degC (12.5% CPU load) Fan @ 1200 (minimum for imac)
i5 pk temp ~ 62degC (25% CPU load) Fan @ 1200

two instances of: yes > /dev/null &
i7 CPU temp peak ~90degC (25% CPU load) Fan @ 1200
i5 pk temp ~ 73degC (50% CPU load) Fan @ 1200

Three instances of: yes > /dev/null &
i7 CPU temp peak ~95+degC (38% CPU load) Fan @ 1900rpm
i5 pk temp ~ 81degC (76% CPU load) Fan @ 1200

Four instances of: yes > /dev/null &
i7 CPU temp peak ~99+degC (50% CPU load) Fan @ 2640rpm (max for imac)
i5 pk temp ~ 89degC (100% CPU load) Fan @ 1200
Note: I didn't let this last one go for long (5 minutes or so). The i5 would probably have gotten over 90 after a while (10 to 20 minutes)...

I did not expect this result - same load, same size CPU die and same thermal management system. I expected the results to be the same! THe i5 does great but the i7 heats up very quickly. It would be great if someone (or more) could run these same tests to see if maybe I got a bad heatsink to cpu attach on the i7 or if this is consistent for others. Takes ~20 minutes to do the four tests.

THANKS!

Im sorry to say this, but I really really hope that those temperatures are an error, or that you have a bad heatsink connection!

I say this because I have just ordered an i7 over an i5, but 90degrees + at anything over a quarter load seems like it woukd make the chip unusable for any decent work. Running a long video edit overnight would burn out the chip, or reduce its life a hell of a lot.

If thats true for all i7s then it would be a lot better to get an i5 instead.

Edit:
I just noticed that the fan speeds arent even the same! The i5 is keeping cool on slow fan speed even at 100%, while the i7 is burning up at 50% even with maxed out fans? This has to be an error, noone would buy something that noisy and hot surely

 

[Frostie]

I'm also awaiting an i7, so hope these measurements are wrong/faulty in some fashion.

Looking at the following test...

http://www.techpowerup.com/reviews/Intel/Core_i5_4670K_and_i7_4770K_Comparison/9.html

...there should be little difference at idle and under load the i5 should run hotter than the i7.

I'll certainly be interested if anyone can run similar tests on their i7's to see what results you get.

 

[bp1000]

Dont they i7's run at higher frequencies and have more transistors.

We have had this discussion over on the MBA forums and the i7's normally run around 2c hotter at idle, fairly insignificant. Under normal operation they tend to rund between 5% - 10% hotter and give a 10% - 15% performance boost to the non hyper-threaded i5. Only in applications that support hyper-threading, i.e. benchmarks, handbrake, FCPm after effects etc...

In most cases people report them running hotter without even monitoring temperatures and usually report fans come on earlier with the i7.

Despite the case of the mac computers being a beautiful svelte heat sink the restricting internal space is likely to create some heat build up, like with any PC but the performance from a night long video encode will only be retarded if the cpu temp reaches 100c and thermal throttling steps in.

I do think it is a bit odd the difference in temperature you are seeing is so great between the two but people have reported under heavy load the i7 does struggle to bring down the heat. This doesnt necessarily mean the performance is going to suffer, it just means the chips are running so hot the cooling design can't bring it down until the task is completed.

 

[Krevnik]

Four instances of: yes > /dev/null &
i7 CPU temp peak ~99+degC (50% CPU load) Fan @ 2640rpm (max for imac)
i5 pk temp ~ 89degC (100% CPU load) Fan @ 1200
Note: I didn't let this last one go for long (5 minutes or so). The i5 would probably have gotten over 90 after a while (10 to 20 minutes)...

A couple things to note about this particular test:
- Hyper-threading doesn't actually give you extra cores, but rather, makes it possible to squeeze more efficiency out of the existing ones. So "50% CPU load" is incorrect. You are actually loading down both roughly equally. Unfortunately, this is reported incorrectly by the OS which really only sees the 4 extra threads as "cores".
- There are minor variations in sensors from machine to machine, so there is a margin of error here to account for.

That said, I'll note that I'm seeing high temperatures in my case, but they peak at around 93C, and the fans peak out at under 2k RPM. It took me running this myself to understand someone's comment about "noisy" as I've been loading my 27" i7 down the last couple days with Handbrake encodes, and haven't gotten much noise. But turning the fans up to max, it does make quite a bit of noise.

I wouldn't call your results normal here. The temperatures being reported are high, and the fans are at max because of the reported temperatures. Now, is it because the temperatures are that high and there is something wrong with the hardware, or is some of the variation of the reported temps by the probes (these are off by couple degrees in one way or another) causing your machine to ramp things up when the temps may actually be a little lower, or maybe both?

 

[Nirurin]

Despite the case of the mac computers being a beautiful svelte heat sink the restricting internal space is likely to create some heat build up, like with any PC but the performance from a night long video encode will only be retarded if the cpu temp reaches 100c and thermal throttling steps in.

Ahh yeh, I wasn't referring to the speed being effected from an overnight run... I was more worried that the chips life expectancy will be reduced significently if it keeps running hot like that.
Spending an extra £150 on a chip, that will mean the iMac will only last 3 years instead of 5... is not much of an investment!!

Edit: To put it another way, my friends desktop i7 cpu is running at a maximum of something like 60c under heavy load. HAswells begin to throttle at 100c, but the general recommendation for overclocking etc is to keep the max temp below 80c at most, under 70c is even better.

These processors seem to run very close to overload temp even when doing basic tasks, according to this test. I cant see how they can survive long like it.

 

[Zimmerman]

My results

Four instances of: yes > /dev/null &
i7 CPU temp peak ~92 degC (50% CPU load) Fan @ 1380rpm

Note: I did let this one go for about 7-8 minutes or so, but I could not see any change in temps or fanspeed after 4-5 minutes.

 

[propower]

Four instances of: yes > /dev/null &
i7 CPU temp peak ~92 degC (50% CPU load) Fan @ 1380rpm

Note: I did let this one go for about 7-8 minutes or so, but I could not see any change in temps or fanspeed after 4-5 minutes.

Super helpful info !! Thanks!!
Any way you could run the other three data points (one instance, two and three) - just to have a more to compare against.

FWIW: After doing more work today I think my BTO is not assembled correctly and will be contacting AC about a replacement unit.

 

[Nirurin]

Super helpful info !! Thanks!!
Any way you could run the other three data points (one instance, two and three) - just to have a more to compare against.

FWIW: After doing more work today I think my BTO is not assembled correctly and will be contacting AC about a replacement unit.

Yeh I would be interested in updates on this too, as I would like to see what others are getting. My unit is meant to arrive on the 15th, so I will try and run these tests myself and I'll put up my results

 

[propower]

OK - a bunch of thoughts - try not to make an epic post ---

Lots of good input from people here!!

1) As far as %CPU load, the numbers I reported are the way Activity monitor and istat look at things - its right when viewed a particular way. Yes, the i7 and i5 are both true quad cores (I tried to find out if they were the exact same size die and I thought it was yes but maybe someone knows a better link). The i7 is able to use any wait states or dead time in each core to support a second thread/core - hyperthreading.

2) The dev/null test may be flawed in comparing instances to CPU load in these two processors. To the best of my understanding the command tells one core to execute a command and then throw the data away (as fast as it can??). Thus if you tell a hyperthreaded core to do this what is the result?? Really not sure.

3) I did read the TechPower article. I fully expected the thermal perf of the i5 and i7 at the same load to be the same!! The i7 would just have more headroom should it ever be needed.

4) Although Dev/Null is interesting I use this machine for Pro Audio. So I created a test session in my system (ProTools 11 with HD Native Thunderbolt hardware). The session is 96kHz, 64 buffer, 20 tracks with audio, 10 instances of Avid Channel strip per track. Here are the loading and temp results I got.

a) Playback: i7 /// i5 (load readings are istat not protools usage meter)
i7 = 34% CPU load ////// i5 = 42%CPU load
i7 Temp ~ 65degC fan 1200 //////// i5 Temp ~ 53degC fan 1200

b) Record all tracks i7 //// i5
i7 = 41% cpu ~68degC fan 1200 ///// i5 = 53%cpu ~62degC

SO: my "real world" loading yields different (and easier to live with) results than the synthetic benchmark. Makes me worry that the Dev/Null test is not quite doing the same thing on the i5 and i7!

5) I also encoded at 12 minute video using imovie.
i7 ///// i5 movie export (12.06 min 1280X720 H.264 AAC)
export 5:35 //// 5:34 SAME !!!!!!
~31% cpu usage ///// ~61% cpu usage
Used 4 cores at 60% each pk temp = 76dec //// 4 cores 60% ea. pk Temp =61degC

 

[Nirurin]

OK - a bunch of thoughts - try not to make an epic post ---

Lots of good input from people here!!

1) As far as %CPU load, the numbers I reported are the way Activity monitor and istat look at things - its right when viewed a particular way. Yes, the i7 and i5 are both true quad cores (I tried to find out if they were the exact same size die and I thought it was yes but maybe someone knows a better link). The i7 is able to use any wait states or dead time in each core to support a second thread/core - hyperthreading.

2) The dev/null test may be flawed in comparing instances to CPU load in these two processors. To the best of my understanding the command tells one core to execute a command and then throw the data away (as fast as it can??). Thus if you tell a hyperthreaded core to do this what is the result?? Really not sure.

3) I did read the TechPower article. I fully expected the thermal perf of the i5 and i7 at the same load to be the same!! The i7 would just have more headroom should it ever be needed.

4) Although Dev/Null is interesting I use this machine for Pro Audio. So I created a test session in my system (ProTools 11 with HD Native Thunderbolt hardware). The session is 96kHz, 64 buffer, 20 tracks with audio, 10 instances of Avid Channel strip per track. Here are the loading and temp results I got.

a) Playback: i7 /// i5 (load readings are istat not protools usage meter)
i7 = 34% CPU load ////// i5 = 42%CPU load
i7 Temp ~ 65degC fan 1200 //////// i5 Temp ~ 53degC fan 1200

b) Record all tracks i7 //// i5
i7 = 41% cpu ~68degC fan 1200 ///// i5 = 53%cpu ~62degC

SO: my "real world" loading yields different (and easier to live with) results than the synthetic benchmark. Makes me worry that the Dev/Null test is not quite doing the same thing on the i5 and i7!

5) I also encoded at 12 minute video using imovie.
i7 ///// i5 movie export (12.06 min 1280X720 H.264 AAC)
export 5:35 //// 5:34 SAME !!!!!!
~31% cpu usage ///// ~61% cpu usage
Used 4 cores at 60% each pk temp = 76dec //// 4 cores 60% ea. pk Temp =61degC

According to these results, the i7 is still running between 10% and 20% hotter than the i5... this shouldn't be happening I'm sure.
Under an equivalent "maximum load", the i7 should only be about 5c hotter than the i5. You're getting closer to 15c-20c, under only 60% load, so maxed out it would probably be more of a difference.

I'd get it taken back to the apple store for an exchange.

 

[propower]

According to these results, the i7 is still running between 10% and 20% hotter than the i5... this shouldn't be happening I'm sure.
Under an equivalent "maximum load", the i7 should only be about 5c hotter than the i5. You're getting closer to 15c-20c, under only 60% load, so maxed out it would probably be more of a difference.

I'd get it taken back to the apple store for an exchange.

This is my belief as well -
I think there is a thermal attach issue with my BTO - now to try an convince them!
Though for my uses - I am giving pause to maybe going i5 after all! Near sa I can tell for all "realistic" loads I care about - the i5 is actually performing better not only in temps but in load balancing per core (all detailed pro tools stuff ). Like my first sentence - yes - I know I'm crazy....

 

[Nirurin]

This is my belief as well -
I think there is a thermal attach issue with my BTO - now to try an convince them!
Though for my uses - I am giving pause to maybe going i5 after all! Near sa I can tell for all "realistic" loads I care about - the i5 is actually performing better not only in temps but in load balancing per core (all detailed pro tools stuff ). Like my first sentence - yes - I know I'm crazy....

Not crazy at all. Up until today I had assumed that the i7 and i5 would have the same temperatures. I thought the only downside to getting the i7 was the extra cost.
After seeing this... I'm wondering if I made a mistake getting the i7 myself. But I will run my own tests when it turns up, and if my temperatures are ok then I doubt I will bother going through the chore of a return.

However for the price I'm paying, I expect this machine to be perfect. Any high temps, scratches, pixels, any flaws at all and I will be going for a replacement.

Let us know what you end up doing, and if you get a replacement machine let us know the new temps!

 

[woddy]

late 2013 i7 (8gb+780nvidia+256ssd) here:

8 instances of yes after 10 minutes give me:
fan around 1600-1800 fluctuating. Not much of a noise doh.

 

[Mac32]

I researched the issue before I purchased my iMac 8 months ago. The best place is PC overclocking forums. PC overclockers usually disable hyperthreading to get higher overclocks. So hyperthreading will raise the CPU temperature, obviously the higher clock speed and turbo boost speed on the i7 iMac CPU also has an effect on the temperature.

 

[DianeK]

I am looking at configuring a new iMac BTO and was going to get the i7 but am now having second thoughts based on this discussion. My computer is used mostly for Photoshop, InDesign and Illustrator. Only minor video editing. Can anyone here tell me if the Adobe products I use benefit from hyperthreading? That would help me decide between the i5 and i7.
Thank you

 

[propower]

late 2013 i7 (8gb+780nvidia+256ssd) here:

8 instances of yes after 10 minutes give me:
fan around 1600-1800 fluctuating. Not much of a noise doh.

Very interesting!

Could you possibly run only 4 and see what fans and temps you get? Be really great if you could run 2 as well but any additional data is great. Thanks!!

 

[bp1000]

I am looking at configuring a new iMac BTO and was going to get the i7 but am now having second thoughts based on this discussion. My computer is used mostly for Photoshop, InDesign and Illustrator. Only minor video editing. Can anyone here tell me if the Adobe products I use benefit from hyperthreading? That would help me decide between the i5 and i7.
Thank you

At these clock speeds photoshop doesn't really saturate the CPU cores to take advantage of hyper threading. It does hyper thread but is also offloads a lot of processing to the gpu so this is just as important with photoshop work.

This maybe of interest as a very high level indicator

Please remember this shows times for a chip 2 gens old and an older version of PS.

The i5 has a slower clock speed too.

The reports from people benchmarking these things, ie setting up benchmark scripts to apply rules and filters and timing it seem to suggest an i7 advantage is fairly indistinguishable with only around 5% improvement.

However on batch operations you may just see some advantages. I think certainly you would see render advantages in video providing the program supports ht, iMovie doesn't but something serious like FCP does.

 

[propower]

Update:

Me saying there is a thermal issue does not equal them saying it is so But no way I am keeping the BTO i7 with the way it has tested for me!

New i5/256SSD/16G ram ordered tonight i7 will be returned when the new one arrives.

Note: Getting the 16G ram option on the 2013 imac gets two 8G sticks instead of 2 4G sticks (if it was four 4G sticks that would suck!). $200 is not that far off from 16G on the retail market (OWC about that delivered). Still doesn't account that for the same dough yo could upgrade to 24G on your own... but -- now I'm only two sticks away from maxing it out someday

I sure hope the new i5 is as good as my test i5 machine was!

 

[am2am]

This topic is REALLY interesting. I am still fighting i5 vs i7 in my new future mac.
Honestly after reviewing this thread I am closer to i5 than ever. Not only less expensive but also quieter/colder with similar performance.

I want 780 graphics and would like to avoid GPU/CPU throttling.
The worst case would be to invest in i7 and have lower performance due to temp limits and throttling.

 

[bp1000]

This topic is REALLY interesting. I am still fighting i5 vs i7 in my new future mac.
Honestly after reviewing this thread I am closer to i5 than ever. Not only less expensive but also quieter/colder with similar performance.

I want 780 graphics and would like to avoid GPU/CPU throttling.
The worst case would be to invest in i7 and have lower performance due to temp limits and throttling.

If you read my post that highlighted thermal throttling please remember that the iMacs rear is one huge heat sink designed to radiate heat to the outside and for the cooler air to reduce temperatures therefore it is a lot less likely to encounter thermal throttle than something smaller like a macbook air.

I would try not to complicate your decision by including thermal throttling into the mix.

Simply think how you divide your computer time. If you are a pro-am videographer or 3d modeller you get the i7 no thought needed.

If you occasionally use final cut pro or after effects its not really needed but a nice to have. Export times may increase with the i7 (general operation would be dictated by your free RAM and storage to an extend). If you cut and export video all day you really dont want to be waiting the extra few minutes for a large export. Most at this level would probably be looking at a mac pro anyway.

For anything less, photoshop, iPhoto, iMovie, browsing/email/etc it is really a waste of money. Most of those apps dont support hyper threading. Photoshop does but rarely would you max out a CPU at 3.4GHZ to benefit from HT in typical PS tasks... unless you load a shoot with batch processing to apply complex filters to each photo. Even then some people report the bottlenecks elsewhere.

They say some games are going to be designed to support HT. Again i really cannot see a significant advantage from this because even more so in gaming the GPU is likely to be the weak link.

 

[joema2]

...
For anything less, photoshop, iPhoto, iMovie, browsing/email/etc it is really a waste of money. Most of those apps dont support hyper threading. Photoshop does but rarely would you max out a CPU at 3.4GHZ to benefit from HT in typical PS tasks... unless you load a shoot with batch processing to apply complex filters to each photo. Even then some people report the bottlenecks elsewhere...

This is generally correct, however hyperthreading support is not an app decision. Apps are either multithreaded or not, and if multithreaded a hyperthreaded CPU will automatically be used. The only requirement is multiple threads must be in a runnable state, not pending on an event.

By contrast at the operating system level there is hyperthreading support. The OS thread dispatcher, if aware if a hyperthreaded CPU, will preferentially schedule threads on every other core. Scheduling four threads on four cores is more efficient than on two hyperthreaded cores of a four core CPU.

Increasingly apps are effectively multithreaded. If so they benefit from more cores, whether real or virtual (in the case of hyperthreading).

Also the issue isn't a single app, but all the runnable threads of all the apps on a machine.

If you use the Mac OS X Activity Monitor and look under the "threads" column, you'll see most processes have many threads. Recently I looked at an iMac on which Lightroom 5.2 had 24 threads, iTunes 18 threads, and Firefox 38 threads. All the runnable threads on the entire machine constitute a pool which the OS tries to assign to available cores (real or virtual).

It is obvious a typical system has many hundred or thousands of threads. At any instant of time, most of those are not in a runnable state but pending on an event (I/O, system call, intraprocess signal, etc). But with enough threads you eventually have four or eight on average in a ready-to-run state, so more cores (physical or hyperthreaded) will benefit.

The bottom line is there's no such thing as an app that takes advantage of hyperthreading. Rather there are multithreaded apps, which will automatically be scheduled on all available cores by the OS. If all the ready-to-run threads across all the apps on a given machine are equal or greater than the number of cores (real or virtual), then having those cores (real or virtual) will generally be beneficial.

 

[DianeK]

joema2: thank you for a very good explanation of hyperthreading.