Toggle sidebar Toggle sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

MP 7,1 Benefits of upgrading the CPU on the Mac Pro 7.1

AndreeOnline

AndreeOnline

macrumors 6502a
Original poster
I feel that the CPU is a weak spot and somewhat of a dead end in my Mac Pro.

I can more than double the core count, but single thread performance is forever locked in. It is what it is.

I'm wondering if anyone out there has made the switch from 12/16c to 28c and noticed a change?

Way back when, almost everything was single threaded. But the last few years I get the feeling that more and more apps, and perhaps macOS itself has gotten tuned to better leverage multi-threading? Is there in fact value to be found in an upgrade, apart from bragging rights?
 
Bigwaff said:
Geekbench results show 28c slower than 14c/12c at single core benchmarks.

Mac Benchmarks - Geekbench

browser.geekbench.com browser.geekbench.com
Click to expand...
Interesting. Didn't use to be that way. Many years ago, some of the top single thread results were from the 28s.

-----

ADDED: Then again, just look at the multi thread scores and you can see how broken Geekbench is. There is no way a 12c scores 9500 and a 28c scores 10800 in a well calibrated benchmark.
And since we've been able to test these CPUs under traditional loads, such as 3D rendering in CPU mode, for years we know they DO scale linearly.

And that presents a problem. If that one score that is easily identifiable as 'false', what else is way off?

I use Geekbench myself, as one datapoint but I don't rely much on it at all. Unfortunately, it's in-app testing that counts.

It may very well be that the Geekbench score IS representative of typical 'experienced' macOS performance between the CPUs. It can be that it represents a modern code base that is similar to how most apps are built these days. Who knows.... but it doesn't make me jump on Ebay to find a W-3275M.
 
Last edited:
I upgraded one of mine from 16 to 28 cores. The other is 28 from the start.

My workflows are different to others so can’t say anything more.
 
Last edited:
AndreeOnline said:
Interesting. Didn't use to be that way. Many years ago, some of the top single thread results were from the 28s.

-----

ADDED: Then again, just look at the multi thread scores and you can see how broken Geekbench is. There is no way a 12c scores 9500 and a 28c scores 10800 in a well calibrated benchmark.
And since we've been able to test these CPUs under traditional loads, such as 3D rendering in CPU mode, for years we know they DO scale linearly.

And that presents a problem. If that one score that is easily identifiable as 'false', what else is way off?

I use Geekbench myself, as one datapoint but I don't rely much on it at all. Unfortunately, it's in-app testing that counts.

It may very well be that the Geekbench score IS representative of typical 'experienced' macOS performance between the CPUs. It can be that it represents a modern code base that is similar to how most apps are built these days. Who knows.... but it doesn't make me jump on Ebay to find a W-3275M.
Click to expand...
Geekbench 5 multicore scores generally increased linearly with the number of cores. If an 8 core CPU scored 10,000 then a 16 core CPU would score 20,000 (assuming the CPU cores were identical across both chips). This garnered a bit of criticism since a lot of real world workflows are not sufficiently parallelized to take equal advantage of all cores.

Geekbench 6 attempted to address this by scoring multicore on a curve, and more than 8 cores produce increasingly diminished scoring improvements. It effectively penalizes high score counts, especially if the CPU design prioritizes core counts over single core performance, as Xeon chips usually do versus equivalent K-series desktop chips.

There's a strong argument that Geekbench 6 is better suited for consumer and prosumer/professional (non-workstation) CPUs but less so for workstation and server CPUs where workflows can and do scale up to take advantage of the added cores.
 
Last edited:
IMO the 16 core is the sweet spot as it offers 9% less single core performance (based on CPU clock speed) over the 8 core CPU but has double the cores. The 28 core CPU has 24% less single core performance than the 12 core CPU and 22% less single core performance over the 16 core CPU. Unless you have a workload you know that can scale to 24 or 28 cores I don't see either of the latter being something to consider even if free (assuming performance is your goal).

This is based solely off of clock speed and, IME, CPU based tasks increase fairly linearly with clock speed but not for core counts. Which is a nice segue into Geekbench (GB). IMO GB is only useful for determining how fast a computer runs GB. If you want to know how well (or not) a computer will work for you then you need to test it with your applications / workflow. If that's not possible search out others who have performed benchmarks with the applications / workflow you use or as close as possible.
 
Bigwaff said:
Let’s bring this conversation back to @AndreeOnline question in context of single thread performance …
Click to expand...
My question was not about single-thread performance. On the contrary, I said that single-thread performance is what it is and can't be changed.

What CAN change is that if mainstream computers are leaving the 2-4 core era from a decade ago and moving to 8 to 10 to 16 cores... AND software follows suit and optimizes for that, then the older 24/28 core Xeons might see some unexpected gains if they can leverage some of it. That is what the question is about.

Bigwaff said:
Agreed.. which is why this thread should sputter and die.
Click to expand...

Never hesitate to take the first door out of the thread if it's of no value or interest to you.

avro707 said:
I don’t think you’ll find too many folks using 28-core machines around here in the first place.
Click to expand...
No, there won't be many. Then again, you have two by yourself.

7.1s are cheaper now, and so are the QS CPUs on Ebay. I think that we'll see a little bit of the 5.1 upgrade hobby repeat itself. Not to the same extent, but still. We'll see.
 
  • Like
Reactions: Flint Ironstag, b17777, Regulus67 and 1 other person
m1maverick said:
IMO the 16 core is the sweet spot as it offers 9% less single core performance (based on CPU clock speed) over the 8 core CPU but has double the cores. The 28 core CPU has 24% less single core performance than the 12 core CPU and 22% less single core performance over the 16 core CPU. Unless you have a workload you know that can scale to 24 or 28 cores I don't see either of the latter being something to consider even if free (assuming performance is your goal).
Click to expand...
You are right about the 'on paper' clock speeds, but it doesn't work out like that. Most of the time, if the 28-core is "stuck" in single-thread mode, it will use its turbo mode and perform the same as the fewer cores counterparts.

In Geekbench 5 and earlier, it was the 28-core that led the single-core chart. I'm not saying it's BETTER, though, I'm just bunching them all together as "the same" in that department.

It might very well be that many apps today don't run in single-core mode, but they also won't really benefit from more than 4-6 cores. THAT scenario might penalize the 24/28 core models with lower clock speeds when outside of the Turbo Boost parameters.... but that only circles back to the original question: what is the situation today actually like?
 
Last edited:
  • Like
Reactions: b17777 and Regulus67
AndreeOnline said:
You are right about the 'on paper' clock speeds, but it doesn't work out like that. Most of the time, if the 28-core is "stuck" in single-thread mode, it will use its turbo mode and perform the same as the fewer cores counterparts.

In Geekbench 5 and earlier, it was the 28-core that led the single-core chart. I'm not saying it's BETTER, though, I'm just bunching them all together as "the same" in that department.

It might very well be that many apps today don't run in single-core mode, but they also won't really benefit from more than 4-6 cores. THAT scenario might penalize the 24/28 core models with lower clock speeds when outside of the Turbo Boost parameters.... but that only circles back to the original question: what is the situation today actually like?
Click to expand...
All core counts, except for the 8 core, have identical Turbo Boost speeds. Therefore for lower core utilization they should all perform similarly. However as the core utilization increases the 12 and 16 core CPUs will down clock to a higher clock speed than the 24 and 28 cores. Therefore, IMO, the 12 and 16 core are the sweet spot unless you know for certain your workload will utilize 24 or 28 cores (I say this as someone who has a 44 core / 88 thread system so I am not against high core counts).

As for the actual situation today that depends on ones use case. It's my opinion if you don't know whether your situation requires high core counts you can assume the 12 or 16 core CPU will probably be faster than a 24 or 28 core CPU.
 
The higher base clocks on the 16 core is probably your best bet for single core perf.

I would say the 28 core is more for multi-threaded performance.

I am feeling the lack of single core perf on the 28 core as days move forward and latest versions of macOS aren't helping either with drop in perf.
 
  • Like
Reactions: AndreeOnline
AndreeOnline said:
No, there won't be many. Then again, you have two by yourself.
Click to expand...

Well yes I do, but Bigwaff took it upon himself to dictate what we could say in your topic. So while I had a different reply originally, I removed it.

My workflows would go well on the latest Xeons with even more cores. And what I’m running will work fine in Windows 11 Pro for Workstations or any of the modern Linux distributions so I’m not tied to macOS.

Bigwaff said:
Agreed.. which is why this thread should sputter and die.
Click to expand...

What was the point of replying in that tone then? Do you even have one of these machines?

Perhaps you missed that I have two of them, both high spec machines.
 
Last edited:
Regulus67 said:
I got 30% boost (running Sonoma 14.5)

Single score stayed the same at 68.
Multi-core from 1219 (16-core) to 1584 (24-core)

edit: room temperature 23°C/74°F
Click to expand...
I mentioned Sonoma, as that was the installed OS
m1maverick said:
IMO the 16 core is the sweet spot as it offers 9% less single core performance (based on CPU clock speed) over the 8 core CPU but has double the cores.
Click to expand...
Do not forget the 8-core has slower memory speeds at 2666MHz, all the others run full speed at 2933MHz.

I got my 3rd Mac Pro three weeks ago. Came with an 8-core. But I didn't think to run tests on it, sorry. And I have already upgraded it to 16-cores.
 
Last edited:
  • Like
Reactions: b17777, AndreeOnline and avro707
AndreeOnline said:
You are right about the 'on paper' clock speeds, but it doesn't work out like that. Most of the time, if the 28-core is "stuck" in single-thread mode, it will use its turbo mode and perform the same as the fewer cores counterparts.
Click to expand...
should double check if these things work on a QS CPU under macOS. Could've sworn someone had trouble with that on this forum or somewhere else
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back