iMac Pro benchmarks

[Flint Ironstag]

cMP
The Imac forum is at https://forums.macrumors.com/forums/imac.91/ . Discuss Imacs there, and the cMP and nMP and mMP here.

I think there's plenty to merit discussion right here. Glad you relented and joined us!

Mac Pro users are cross - shopping and considering iMac Pro simply because:

1. the Pro moniker
2. those for whom 6,1 was a success, maybe they can live with more of the same in an iMac form factor
3. iMac Pro is faster, and available NOW.
4. Apple's own eGPU dev kit supplier, Sonnet, has listed the 6,1 as incompatible with their enclosures!

 

[MarkJames68]

I think there's plenty to merit discussion right here. Glad you relented and joined us!

Mac Pro users are cross - shopping and considering iMac Pro simply because:

1. the Pro moniker
2. those for whom 6,1 was a success, maybe they can live with more of the same in an iMac form factor
3. iMac Pro is faster, and available NOW.
4. Apple's own eGPU dev kit supplier, Sonnet, has listed the 6,1 as incompatible with their enclosures!

But for #4, Akitio hasn’t said this about the Node

On a separate but related note my 6,1 with eGPU did a spontaneous reboot today with a GPU error logged. Crossing my fingers.

 

[fastlanephil]

Apple just released Logic Pro X 10.3.3 which now recognizes 36 processing threads, up from 24. I assume that’s intended for the 18-Core iMac Pro. It sounds like it might be a pretty quiet machine regardless of which processor it uses.

 

[rawweb]

The real question...does the thing have a boot chime?

 

[orph]

id gess we will chat about it as news but the imac forum is it's home, then more specialized questions will be on appropriate software forums.
relay most the questions here are what GPU to use or i cant boot whats wrong

there are vary few high level software discussions and i gess most people look to youtube for "how to" help now.

id like to see how it runs but it's not what im looking to buy, >.< unless apple pulls something special out ill go hack next.

so is the bigger Vram size for 8K video ? id like to see how resolve runs on it

 

[OS6-OSX]

I thought about getting a max out iMac Pro. But I added an extra 1K and got this instead! I just have to find a way to get it off pedestal!

 

[FredT2]

Does that mean they are basically the same when running programs (including benchmark programs) that are single-threaded (until the system cannot handle the heat)?​

Yes - but there's some confusion in your statement. Assuming that the system has reasonable cooling (perhaps not a good assumption for an Imac) a single thread will never "not be able to handle the heat". All single-threaded apps should sit at the turbo speed start to finish unless the thermal system is way, way weak.

Under what circumstances does the CPU drop down to the base clock speed?

 

[h9826790]

Under what circumstances does the CPU drop down to the base clock speed?

AFAIK, when thermal limiting. In other words, when the CPU temperature is above an Intel pre-defined threshold. Therefore, it may be a big issue on iMac Pro. Even Apple says iMac Pro's cooling improved by 80%, however, the iMac Pro's TDP also increase by 67%. That means, actual improvement may be as low as 7%. By considering the original iMac's cooling is really bad (both CPU and GPU can easily reach 100C when under stress). 7% on top of a very bad cooling may not even enough jump from "very bad" to "bad". And I won't expect the CPU on an iMac Pro can stay at the turbo clock for a long period of time. I suspect Apple only care to make it good enough to boost under those famous benchmarking software (e.g. Geekbench). As long as the data looks good on those tech magazine / website, it's good enough for them to tell the public how fast the iMac Pro is.

 

[whwang]

I saw in multiple places that when more cores are used, the speed will be lower, closer to the base speed rather than the turbo boosted speed. For example, here:
https://macperformanceguide.com/blog/2017/20171214_0823-iMacPro2017-released.html
"when more cores are used, the iMac Pro CPUs drop towards base clock speed"

Is this what will definitely happen (i.e., lower clock rate) when more cores are used? Or this only happens when the cooling cannot keep up with the heat? I am a bit confused.

 

[h9826790]

I saw in multiple places that when more cores are used, the speed will be lower, closer to the base speed rather than the turbo boosted speed. For example, here:
https://macperformanceguide.com/blog/2017/20171214_0823-iMacPro2017-released.html
"when more cores are used, the iMac Pro CPUs drop towards base clock speed"

Is this what will definitely happen (i.e., lower clock rate) when more cores are used? Or this only happens when the cooling cannot keep up with the heat? I am a bit confused.

Only happen when cooling is bad.

The old Xeon (e.g. X5690) only has Turbo Boost 1.0. Which cannot boost all cores to the max speed at the same time.

But on the newer Xeons, they have Turbo Boost 2.0. Which allow all cores boost to the max turbo speed at the same time.

 

[FredT2]

Only happen when cooling is bad.

The old Xeon (e.g. X5690) only has Turbo Boost 1.0. Which cannot boot all cores to the max speed at the same time.

But on the newer Xeons, they have Turbo Boost 2.0. Which allow all cores boost to the max turbo speed at the same time.

Ah, OK. I've been confused as to what base clock speed means vs. boost speed. My 2014 i7 iMac quickly drops to between 3.5 and 3.8 GHz (below base clock speed of 4.0) running Handbrake, whereas my i7 Mac Mini happily runs Handbrake for hours at close to it's boost speed of 3.4 GHz.

 

[beaker7]

Only happen when cooling is bad.

The old Xeon (e.g. X5690) only has Turbo Boost 1.0. Which cannot boot all cores to the max speed at the same time.

But on the newer Xeons, they have Turbo Boost 2.0. Which allow all cores boost to the max turbo speed at the same time.

That’s sort of true. They don’t boost all cores to max when all cores are in use. Depending on the cpu you’ll get max boost for 1-3 cores, then it tapers off after that.

 

[h9826790]

Ah, OK. I've been confused as to what base clock speed means vs. boost speed. My 2014 i7 iMac quickly drops to between 3.5 and 3.8 GHz (below base clock speed of 4.0) running Handbrake, whereas my i7 Mac Mini happily runs Handbrake for hours at close to it's boost speed of 3.4 GHz.

Turbo boost 2 "allow" that happen, doesn't mean it will happen. Also, not all CPU are configured to stay at max when all cores in use. e.g. E7-8891V4 (10C 20T). Base clock 2.8GHz, Max turbo at 3.5GHz (only happen when 1-2 cores in use), and when all cores in use, max turbo is 3.3GHz. It still allow 18% boost when all cores in use, but not really stay at "max".
[doublepost=1513450978][/doublepost]

That’s sort of true. They don’t boost all cores to max when all cores are in use. Depending on the cpu you’ll get max boost for 1-3 cores, then it tapers off after that.

Thanks for pointing that out. I also realised that I should make it more clear. So, a supplementary post is added.

 

[Fl0r!an]

There are even detailed tables describing the behavior (didn't find one for iMac Processors, but I guess it won't be much different):

 

[Macintosh IIcx]

Regarding turbo speed and core numbers in use, this graphic from Anandtech explains it very well:

https://images.anandtech.com/doci/11839/turbos2.png

Note how the 10 core CPU should stay at 4.0 for all 10 cores working, if temperature doesn’t become an issue. This might be different with the versions used in the iMac Pro, but the principle remains the same.

Ack, Fl0r!an beat me to it!
[doublepost=1513451981][/doublepost]One other note to add: Those AVX instructions are extremely power hungry and will eat into those turbo speed clock fast. Not many apps use AVX as far as I understand, so not a big issue, and if they do, it is for a reason and will speed things up fairly massively.

 

[flat five]

i wonder where an upper-end regular iMac sits on this benchmark..
it probably bests them all.

 

[orph]

dont forget if apple has tuned the fans for low speed you can just juck the fans faster and see it boost more.

 

[pl1984]

I saw in multiple places that when more cores are used, the speed will be lower, closer to the base speed rather than the turbo boosted speed. For example, here:
https://macperformanceguide.com/blog/2017/20171214_0823-iMacPro2017-released.html
"when more cores are used, the iMac Pro CPUs drop towards base clock speed"

Is this what will definitely happen (i.e., lower clock rate) when more cores are used? Or this only happens when the cooling cannot keep up with the heat? I am a bit confused.

Every processor has a maximum power rating which it is certified for. In the case of the 8-core iMac Pro the processor is rated at 140W (assuming the Xeon W-2145 processor). In the days of single core processors the power rating would be met by clocking the processor up to a frequency which would result in operation at the 140W rating. As cores are added, if clock frequency remained the same, the power consumption increased which would the exceed the power rating. In order to accommodate this the clock speed was reduced in order to maintain the same power rating. For multithreaded applications this was a reasonable trade off. But for single threaded applications it meant a decrease in performance. Once needed to choose either more cores and less clock speed or more clock speed and fewer cores.

Enter Turbo Boost. The idea being the processor would shut down cores when they are not needed and increase the clock speed. As more cores were needed the processor would clock down. All the while remaining within the stated power rating.

At a very high level the processor, given sufficient cooling to dissipate heat at the rated power rating, will operate at maximum Turbo Boost speed when only one core is needed. As more cores are needed the processor will clock down to some level between the maximum Turbo Boost speed and base clock speed. When all cores are needed the processor will clock down to the base clock speed. Under normal operation the processor should never drop below the base clock speed no matter how many cores are in use.

The above paragraph is an simplified explanation to illustrate the concept. There are other factors, such as cooling, which can affect how the processor clocks. If sufficient cooling is provided then the processor should be able to bring more cores online at a higher clock frequency. If cooling is insufficient there's the possibility the processor could clock lower than the base clock frequency in order to protect itself.

HTH

 

[Fernand]

I'm a Logic user primarily. I use a 12 core upgraded cMP. Logic's design requires every DSP function on the currently open-for-recording track, including the virtual instrument(s), the effects, AND any reverbs and other send'ed effects that are common to it, to run on a single thread. This was originally done to keep perfect sync, and it's still like that. What happens then is that the single core Geekbench score predicts how easily the whole system saturates. Processing all the "playback" tracks keeps the "input" core from running at full Turbo. All the Xeon-based macpros to date have been shamefully crushed by i7 systems, so severely that people add $1k slave PCs after spending a fortune on a nMP.

Thermal constraints have especially limited the very high-core count Xeon machines' single core performance, and for Logic users a 12 core 3.47 cMP is only about 10% slower than a $10k 12 core trashcan nMP. So, single core Geekbench is the benchmark I want to see for the $14k iToaster-Pro. It might work better for video editors, until its GPU becomes old hat, but for Logic it's entirely possible that an i7 based iMac or Hack'tosh will trounce it.

 

[pl1984]

I'm a Logic user primarily. I use a 12 core upgraded cMP. Logic's design requires every DSP function on the currently open-for-recording track, including the virtual instrument(s), the effects, AND any reverbs and other send'ed effects that are common to it, to run on a single thread. This was originally done to keep perfect sync, and it's still like that. What happens then is that the single core Geekbench score predicts how easily the whole system saturates. Processing all the "playback" tracks keeps the "input" core from running at full Turbo. All the Xeon-based macpros to date have been shamefully crushed by i7 systems, so severely that people add $1k slave PCs after spending a fortune on a nMP.

Thermal constraints have especially limited the very high-core count Xeon machines' single core performance, and for Logic users a 12 core 3.47 cMP is only about 10% slower than a $10k 12 core trashcan nMP. So, single core Geekbench is the benchmark I want to see for the $14k iToaster-Pro. It might work better for video editors, until its GPU becomes old hat, but for Logic it's entirely possible that an i7 based iMac or Hack'tosh will trounce it.

Based on what you've said about Logic it sounds as if the four or six core nMP would be more appropriate than the twelve core model.

 

[MarkJames68]

Based on what you've said about Logic it sounds as if the four or six core nMP would be more appropriate than the twelve core model.

For a nMP the E5-2667 V2, not a factory option, works best. 8 cores at 3.3 that turbos up to 4.0, higher than any other CPU.

 

[Fernand]

For me, with high track counts and connected devices, I like my 3.33 Ghz 12 core cMP w/ 64GB. Or my 4.2 Ghz i7-4790K.

 

[pl1984]

For a nMP the E5-2667 V2, not a factory option, works best. 8 cores at 3.3 that turbos up to 4.0, higher than any other CPU.

As this CPU has the same power rating as other CPUs Apple uses I'm surprised Apple never bumped the specs of the nMP with it.

 

[orph]

the nMP problem was the GPU not the CPU, there's been a few articles about it.
the nmp heat sink was designed for two GPU's that output heat X and GPU's moved to a higher GPU heat output and apple got stuck.

 

[Fernand]

First person to show us GeekBench results (even the free 32 bit version 3 is fine, in fact preferred as it's the most common denominator) for one or more iToaster Pros (especially the high end ones) might be eligible to receive a valuable coupon that might entitle them to a rebate of $5 or less on their purchase.