MP 7,1 Effective single thread turbo speed(for owners of 28core/16core 7.1)

[sirio76]

I guys, I just want to know if somebody can run Intel Power Gadget on their 28core and 16core 7.1 models to see if they reach the 4.4/4.6Ghz on single threaded tasks.
Before spending on a 28core CPU I want to be sure that the number of cores does not affect the capacity to reach full turbo, otherwise I'll get a 16core.
Thanks in advance

 

[sirio76]

Anyone? Singe thread speed is one of the most important aspect for a workstation, looks weird nobody check what turbo can be achieved on the 7.1

 

[Parzival]

Interested in this as well

 

[ZombiePhysicist]

How would I test this? I've seen my machine go over 4GHz a few times but I really haven't been pushing this, and it's involved multiple cores. Usually about 3 or so peak load wise, but not sure how I see what speed the individual cores are running at?

 

[Ph.D.]

I'm interested in this as well.

In my linux box, I routinely see sustained 5GHz speeds, and that's a 9900k on air cooling. Yet in this super-engineered 'pro, what I've seen so far is that max boost speeds are rarely if ever maintained. I know this is intended to be a conservative 24/7 machine, and so it may not be reasonable to expect high sustained turbo performance, but it's feeling a little too conservative.

 

[ZombiePhysicist]

my machine, when I'm doing nothing and it's mostly idling, is down around 2.6GHz and the overall system is eating up about 150w (measured by my UPS). Which I like, meaning when it's not doing anything, there is no reason the frequency should be spiked.

If someone mentions a benchmark or something that would do the single thread long enough, I'm happy to test it and tell you what it spikes to and for how long and what it throttles down to.

 

[bxs]

In order to observe a single core reach its max turbo frequency it will likely be necessary for all cores, except one, to be disabled and then running a CPU intensive code to encourage the remaining operating core to reach its max turbo boost frequency. The Stockfish.app (Chess)is a good one for doing this.

This can be done using the Xcode's Instruments.app.

I just did this on my late 2016 15" rMBP13,3 that has a max turbo boost frequency of 3.5 GHz. I set core count to 1 and ran a CPU intensive code while Intel PowerGadget was running.... see attached image showing the 3.5 GHz was attained.

From my experience, the only way to see a core reach its max turbo boost frequency is to perform the above. It may be possible with having just two cores active, but any more than this it's unlikely to see any core reach the max turbo boost frequency.

 

[sirio76]

Yes, of course the max turbo can be achieved only artificially since the system will most likely use more than one thread at the time. That being said in real world usage there are still many tasks that should get you close to the full turbo, for example on my 8core 6.1 I can get about 3.85Ghz(theoretical max turbo 3.9) running intensive Cinema4D editing stuff.
I will be happy to see a 28core getting close to 4.4Ghz(even though theoretically can run the best core up to 4.6), if that's the case than I'll justify the price, otherwise I'll go for the 16core CPU, hoping that fewer core can help to reach higher performance on single thread(that's why should be useful to see also the 16core results).

 

[thexash]

I have a 28cores, I havent checked the frequency in single core runs, but from Geekbench - my runs and the run you can find on the internet the 28 cores is marginally faster / the same speed as the other variants in single core. Only the base 8 cores version is a bit slower than the others.

I have had ton of hardware pass through the past few years and all of them have about the same single core perf, within 10% (1000-1100GB5). A quick sample: Ryzen 7 1700, Macbook Pro 13" 2016 dual core, Macbook Pro 13" 2018 quad core, top spec 16" mbp, a threadripper 2950x and the Mac Pro 28 cores. Really, the single core performance are really almost equivalent - that is the difference is so small that I never would use that as a criteria to chose one over the other... The only CPU I've seen that is somewhat faster are the 9900K's that can be 15-20% faster in some single core scenario.

Imo if you're buying a machine today go for as many core as you can afford and you can almost completly forget single core since its going to be about the same for all of them. That being said the 12-28cores macpro also have the best single core except for those 9900K (which is on the top regular iMac - not pro - I think).

In general "fewer cores" dont help for the same CPU architecture since turbo boost aim to use the full power envelope and larger core count CPUs have a larger envelope than the lower core count variants.

In the end if you're not sure whether you need more cores because your workload doesnt use them then the 24-28cores price jump is hard to justify... Its "only" about 30% faster than the 16cores version even in multicore workloads.

(I use geekbench as a proxy because it indicates reasonably the performance I'll get for the work I do... ymmv)

 

[defjam]

If someone mentions a benchmark or something that would do the single thread long enough, I'm happy to test it and tell you what it spikes to and for how long and what it throttles down to.

Maybe Cinebench R20 would run long enough to provide this information?

 

[ZombiePhysicist]

OK so good news is Cinebench has a mode to test just one CPU. And I ran it.

So the intel power gadget shows 4 items for frequency.

The Max, Avg, min, and Request.

So it ran around max 4.3-4.4Ghz and the avg was 3.5Ghz (between 3.4 and 3.6 for the most part) and min was like 2.0Ghz. The request was always 4.4. iStat shows the same as the average measure, so it was showing around 3.5ghz.

Pkg temp was from 54-70 (seemed to work to 70 and fans must have kicked up but I couldn't hear them) and then went down back to around 50-60.

I ran it like 3 times and got bored. Hope that helps.

 

[sirio76]

Thanks a lot for the info!

 

[AdamSeen]

I had 3 cores running at 100% usage each, which dropped down to around 4GHz when the 4th core went to 100%. That’s on a 16 core.

I tested this by launching a new ‘yes’ bin in terminal until it dropped below turbo. The test was very basic, but what it’s saying is with 3 cores running at max, you should be able to keep full turbo speeds on the 16 core.

 

[Coyote2006]

I have a 28cores, I havent checked the frequency in single core runs, but from Geekbench - my runs and the run you can find on the internet the 28 cores is marginally faster / the same speed as the other variants in single core. Only the base 8 cores version is a bit slower than the others.

I have had ton of hardware pass through the past few years and all of them have about the same single core perf, within 10% (1000-1100GB5). A quick sample: Ryzen 7 1700, Macbook Pro 13" 2016 dual core, Macbook Pro 13" 2018 quad core, top spec 16" mbp, a threadripper 2950x and the Mac Pro 28 cores. Really, the single core performance are really almost equivalent - that is the difference is so small that I never would use that as a criteria to chose one over the other... The only CPU I've seen that is somewhat faster are the 9900K's that can be 15-20% faster in some single core scenario.

Imo if you're buying a machine today go for as many core as you can afford and you can almost completly forget single core since its going to be about the same for all of them. That being said the 12-28cores macpro also have the best single core except for those 9900K (which is on the top regular iMac - not pro - I think).

In general "fewer cores" dont help for the same CPU architecture since turbo boost aim to use the full power envelope and larger core count CPUs have a larger envelope than the lower core count variants.

In the end if you're not sure whether you need more cores because your workload doesnt use them then the 24-28cores price jump is hard to justify... Its "only" about 30% faster than the 16cores version even in multicore workloads.

(I use geekbench as a proxy because it indicates reasonably the performance I'll get for the work I do... ymmv)

so you would recommend the 12 core MP 2019 over the 8 core?

 

[OkiRun]

I use the 7.1 16-core for video production with no speed issues running software. If future software updates make it 'necessary' $$/time wise to upgrade, I will but not until I see real-word evidence that the increase in core count 'matters' in regards to idle editors; and if that is happening to a significant degree, I will add another computer - not upgrade a CPU.

 

[thexash]

so you would recommend the 12 core MP 2019 over the 8 core?

Definitely, it also uses faster memory (2933 vs 2666 for the 8 cores - thar's a cpu limitation not the ram itself). As the base spec I would suggest a 12cores, 1Tb ssd, 48gb ram and the 580x. If you need more GPU the best is to wait for the W5700XT or put your own, the Vega II is expensive for what it is - unless you need 32GB of video memory. If you need more CPU the 16cores is decent value, the jump in price to 24/28cores is not worth it unless you know you need it.

 

[Coyote2006]

Definitely, it also uses faster memory (2933 vs 2666 for the 8 cores - thar's a cpu limitation not the ram itself). As the base spec I would suggest a 12cores, 1Tb ssd, 48gb ram and the 580x. If you need more GPU the best is to wait for the W5700XT or put your own, the Vega II is expensive for what it is - unless you need 32GB of video memory. If you need more CPU the 16cores is decent value, the jump in price to 24/28cores is not worth it unless you know you need it.

Thanks for your input. I'll probably got with 12/1TB/32GB/5700 (RAM upgrade by myself). That exceeds my needs but should be future proove as I want to have long time replacement for my good old 5,1.