Marco.org • About ▾

I’m : a programmer, writer, podcaster, geek, and coffee enthusiast.

Turbo Boost and the New Mac Pro’s CPUs

With the new Mac Pro’s release imminent, it’s important for prospective buyers to understand the odd-looking CPU options:

ModelCoresSpeedL3 CacheTDPLikely costTurbo Boost
E5-1620 v243.7 GHz10 MB130 W+$0(0/0/0/2)
E5-1650 v263.5 GHz12 MB130 W+$500(1/1/2/2/2/4)
E5-1680 v283.0 GHz25 MB130 W+$1500(4/4/4/4/5/7/8/9)
E5-2697 v2122.7 GHz30 MB130 W+$2500(3/3/3/3/3/3/3/4/5/6/7/8)

It looks like you’re paying a lot for slower clock speeds as the cores increase, but that’s not the entire story. Those weird Turbo Boost numbers, which are easy to pull from here and here, are worth understanding before choosing a modern Intel processor.

They indicate the number of extra 100 MHz increments by which the CPU may ramp up its speed with a given number of cores in an active, high-power state. The sequence begins with all cores active, then counts down to just one core active. For instance, the 6-core’s increments are “(1/1/2/2/2/4)”, which means:

This is probably a more helpful way to compare:

Active Cores
CPU123456789101112
4-core 3.73.93.73.73.7
6-core 3.53.93.73.73.73.63.6
6-core 3.74.03.93.83.83.83.8
8-core 3.03.93.83.73.53.43.43.43.4
8-core 3.34.03.93.83.73.63.63.63.6
12-core 2.73.53.43.33.23.13.03.03.03.03.03.03.0
Maximum Turbo GHz Per Core

(The two red entries — the 6-core E5-1660 v2 and 8-core E5-2667 v2 — are not available in the new Mac Pro, but I wish they were. Faster, the same TDP, and the same or larger cache, for a few hundred bucks more.)

This is why the AAPLJ90,1 Geekbench results make sense: the single-threaded performance on all but the 12-core is effectively identical, and the 6- and 8-core’s multithreaded results scale almost perfectly linearly with their respective core count despite an advertised 500 MHz base clock difference.

You can also see why I don’t recommend the 12-core model to anyone except those whose software will definitely make very good use of all of its cores, at least most of the time — because for any other conditions, it’ll be slower than the others.

Turbo Boost is also why the iMac and 15” MacBook Pro are matching the new Mac Pro already in single-threaded benchmarks, why the 15” MacBook Pro is so much faster than the 13”, and why the Air can keep up despite a much lower base clock speed:

Active Cores
CPU1234TDPCache
13” MacBook Air 1.73.32.9 15 W4 MB
13” RMBP 2.83.33.1 28 W4 MB
15” RMBP 2.63.83.73.63.647 W6 MB
27” iMac 3.53.93.93.83.784 W8 MB
Mac Pro 4-core 3.73.93.73.73.7130 W10 MB
Maximum Turbo GHz Per Core

So why buy a Mac Pro for CPU performance at all?

The increased L3 cache helps certain workloads, but the biggest difference is the TDP, which specifies the maximum sustained heat output of each CPU.

Turbo Boost can only sustain its higher speeds as long as it’s being adequately cooled and is under its TDP limit. This is why the clocks decrease as the core count increases: since all of the Mac Pro’s CPUs have the same TDP, Intel can’t just offer a 12-core that can sustain 3.9 GHz.

While the MacBook Air can match the 13” MacBook Pro’s clock briefly, it won’t hold it for as long because it can’t afford the heat. Those giant 130 W TDPs in the Mac Pro can accommodate much more than the laptops and iMac under a sustained heavy workload — especially if the CPU is being stressed but the GPUs aren’t, due to the shared-giant-heatsink design. (And if the GPUs are being stressed, the Mac Pro should be justifying itself quite well already.)

But there’s little reason to get the higher-end Mac Pro CPUs unless you know you’ll use all of the cores. And if you won’t be sustaining heavy parallel loads and you won’t take advantage of heavy GPU power, there’s a lot less reason to get the Mac Pro at all.

Ads via The Deck