MP All Models How many Mac Pro cores would you get?

[AidenShaw]

10% is about the intended speed difference.

You just said it benchmarked about a 10% difference. That's about the correct number. We're in agreement about what the performance boost is. You just think Hyperthreading is something extra that it is not supposed to be.

Actually, this tangent started over claims about SMT4 and SMT8 vs SMT2. We agree that 10% improvement for SMT2 is about the norm, but it's application dependent.

I question whether making a "Xeon" or "Epyc" processor that's SMT4 will have even a marginal improvement - *unless the SMT4 CPU has more execution units*.

To oversimplify that point - imagine an SMT2 CPU with one floating-point arithmetic execution unit and one integer arithmetic unit, and an application workflow with multiple threads of mixed integer/float content.

• With SMT disabled, the CPU will round robin between threads - sometime waiting on memory, sometimes waiting on the floating-point unit, sometimes waiting on the integer unit.
• With SMT enabled, when one thread stalls on memory/floating/integer - it can run the second thread if it's not waiting on whatever the first thread has busy.
• Not even close to twice as fast, but a significant boost for little extra cost.

Adding more threads without improving the physical core won't be a big win for most workloads. I suspect that the Power9 has far more execution units than x64 processors. (That happens when cost is no object.)

And I'm arguing against the people who want SMT to be more than it is. I usually disable HT, but for some workloads I turn it on.

 

[mmomega]

Have you looked at the price tag for the 28-core upgrade?

Nope, it's not in my budget to even look at it.

But if I'm spending someone else money.... 28 CORES!

I am sarcastic a bit but no, I haven't even spec'd one out because I'm not in the market. I don't typically look until I am ready to purchase, not much of a window shopper anymore.
I do know how much my 18 core was and that was a tough pill to swallow.

 

[telequest]

The reliability of the MP7,1 is unknown. The MP6,1 had some serious issues.

Apple will make you haul a broken system to a store and leave it for service. Dell and HP will come to your home/office the next day for repairs, and offer 5 year service warranties.

OK thanks for that perspective. So I now have 2 unanswered questions:
1) What CPU would be the "sweet spot" for video editing apps such as the Adobe CC suite?
2) How long after the 7,1 starts shipping will we get a general sense of its reliability? How long did it take for 6,1 "serious issues" to become known? I'm definitely not placing an order until at least 2020 precisely because I'd want to hear users' real-world experience with the new machine before jumping in. Of course, some things don't fail for years, but obvious flaws ought to emerge after, how soon, maybe 6 months or so?

 

[goMac]

Actually, this tangent started over claims about SMT4 and SMT8 vs SMT2. We agree that 10% improvement for SMT2 is about the norm, but it's application dependent.

It's not really as application dependent as you seem to imply. I think it would actually be pretty hard to come up with a scenario where everything is even. You'd have to be in a case where there are absolutely no stalls, and all parts of the pipeline are either busy or irrelevant. That's a pretty rare situation. Basically you'd have to be in a process where you are entirely in cache (rare) and doing absolutely one type of math across the entire system including your drivers (very very very rare to impossible.)

I question whether making a "Xeon" or "Epyc" processor that's SMT4 will have even a marginal improvement - *unless the SMT4 CPU has more execution units*.

It depends if Xeon is being underfed. It doesn't have to have more execution units if it was underfed to begin with.

Not even close to twice as fast, but a significant boost for little extra cost.

Correct, but 10% is also not nothing. But the main point people argue against it is that it's a virtual core that doesn't double your speed, which is a total misunderstanding of what it's for to begin with.

Adding more threads without improving the physical core won't be a big win for most workloads.

No, but it's nearly always some sort of win.

And I'm arguing against the people who want SMT to be more than it is. I usually disable HT, but for some workloads I turn it on.

Again, SMT is 99% always some sort of win. It makes almost no sense to turn it off and on, unless you have benchmarks showing it slows you down, which I'd be very curious to see.

It won't be double, but that's a straw man. it was never meant to be double. Intel never said it was double. People fundamentally don't understand why Hyperthreading is, make up what it's supposed to be, and then "win" their own argument.

10% off a rendering time is not nothing. Especially for such little investment in the silicon.

 

[vaugha]

If I was rich, 28-core. Realistically speaking, I'd get the 16-core. Counting my days towards the online config options that should be hitting the web real soon.

 

[lowendlinux]

I'd welcome configuration suggestions for video editing using the Adobe CC suite (Premiere Pro, After Effects, Media Encoder, etc),. Will eventually replace my upgraded 5,1 as 4K becomes more common. Seems there is likely a "sweet spot" as well as a tradeoff between spending on the processor vs video card & RAM. Back in 2010, the 6-core 3.33GHz seemed to fit that sweet spot (good combo of high clock & decent number of cores vs the other options at the time) ... not the fastest clock or the greatest number of cores, but sort of a Goldilocks "just right." Any similar option for 2019?

If I lived in the creative suite you couldn't pay me to be on a Mac.

 

[skippermonkey]

I'm waiting to see prices, but for me 24-core feels right. It's twice the core count of my current machine, and I figure the price differential between 24 and 28 probably won't match the price differential. But who knows: if the 16-core is way cheaper and only marginally less capable, I could drop down, but I have the budget earmarked to go pretty high in cores.

(For interest, I use C4D and Keyshot for rendering, but looking forward to Octane in Metal, too, as an alternative option).

 

[Macintosh IIcx]

I'm waiting to see prices, but for me 24-core feels right. It's twice the core count of my current machine, and I figure the price differential between 24 and 28 probably won't match the price differential. But who knows: if the 16-core is way cheaper and only marginally less capable, I could drop down, but I have the budget earmarked to go pretty high in cores.

(For interest, I use C4D and Keyshot for rendering, but looking forward to Octane in Metal, too, as an alternative option).

You should definitely consider upgrading the CPU yourself then as you can buy the 24 and 28 cores versions that “only” supports 1 TB of RAM. The BTO options that Apple offers for 24 and 28 core, are the M-variants that cost significantly more for the added support of 1,5 TB of RAM.

We can’t be entirely sure that these non-M variant for the 24 and 28 cores are supported yet, but I would be surprised if they aren’t.

It is going to be an awesome machine for C4D.