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

Primate Labs Launches Geekbench 6 Benchmarking Suite

Comparing M2 base model Mac mini with iMac 2020 5k i5 with radeon 5300 pro, it seems the new benchmarks favor the M2 more than the i5. Would be great to see if this is only in this instance or something that also applies to newer intel machines. Attachment showing single, multi and metal. The radeon pro 5300 still beats the M2, but less than it did in GB5 measured in %, and when it comes to CPU the M2 now has an even larger % win over the i5. Maybe because the M2 is more modern, and built for the newer tasks in GB6?

I also did similar calculations between M1 and M2 base models by looking at other peoples posts here, and there it seems to favor the M1, and closing the gap between M1 and M2 more in GB6. But then again I used the numbers from the GB5 official charts, and compared them to posted results here for a single computer, so that might also be the reason.
 

Attachments

  • IMG_0138.jpeg
    IMG_0138.jpeg
    492.8 KB · Views: 122
2019 Mac Pro Xeon (16-core) really showing it's age on the CPU front.

GB 6 Single = 1382 / Multi = 10469 (Max multicore usage was like 16 threads at 80% and the other 16 around 70%)
GB 6 W6800X Duo (Die1 / Die 2) Metal = 183202 / 179058 (Usage peaked at 63% for about 1 second, mostly stayed around 2%)
GB 6 W6800X Duo (Die1 / Die 2) OpenCL = 94647 / 91920

I am going to take the GPU results with a grain of salt, since real world GPU based 3D rendering the W6800X Duo crushes all current Apple Silicon; and the fact that usage barely peaked at all.

Based on the CPU usage percentage; I feel the multi-core test is flawed. Can someone with Apple Silicon run the test with activity viewer CPU and GPU charts up to see if they do the same thing.

I know whenever I render something in Cinebench or Lightwave 3D, every single CPU thread goes to 100% until the render is done.

(*edit, I have attached photos, 1st one is GPU % during compute metal test, 2nd one is max CPU usage during CPU test)

(*edit 2, added CPU usage during Lightwave Render, and the Lightwave render specifics and final scene, which was a 3840x2160 frame size; this seen has massively large texture maps also)
 

Attachments

  • Screenshot 2023-02-14 at 11.48.16 AM.png
    Screenshot 2023-02-14 at 11.48.16 AM.png
    138.1 KB · Views: 77
  • CPU Max Usage.png
    CPU Max Usage.png
    72.7 KB · Views: 84
  • Lightwave Render CPU.png
    Lightwave Render CPU.png
    74.5 KB · Views: 62
  • Lightwave Render.png
    Lightwave Render.png
    1.8 MB · Views: 70
Last edited:
avkills said:
2019 Mac Pro Xeon (16-core) really showing it's age on the CPU front.

GB 6 Single = 1382 / Multi = 10469 (Max multicore usage was like 16 threads at 80% and the other 16 around 70%)
GB 6 W6800X Duo (Die1 / Die 2) Metal = 183202 / 179058 (Usage peaked at 63% for about 1 second, mostly stayed around 2%)
GB 6 W6800X Duo (Die1 / Die 2) OpenCL = 94647 / 91920

I am going to take the GPU results with a grain of salt, since real world GPU based 3D rendering the W6800X Duo crushes all current Apple Silicon; and the fact that usage barely peaked at all.

Based on the CPU usage percentage; I feel the multi-core test is flawed. Can someone with Apple Silicon run the test with activity viewer CPU and GPU charts up to see if they do the same thing.

I know whenever I render something in Cinebench or Lightwave 3D, every single CPU thread goes to 100% until the render is done.

(*edit, I have attached photos, 1st one is GPU % during compute metal test, 2nd one is max CPU usage during CPU test)
Click to expand...

This is not a single test - you cannot base anything off GPU or CPU usage. There are multiple tasks being benchmarked, usage will go up and down (and are not at all accurate when a sample is only taken once per second). A specific task may only need 2% usage, if it's a fairly simple task for the P.U. to process. It's the time it takes to run each task that is measured... that is what is being rated, based off the time it took the base i7 to perform the same task.
 
obviouslogic said:
This is not a single test - you cannot base anything off GPU or CPU usage. There are multiple tasks being benchmarked, usage will go up and down (and are not at all accurate when a sample is only taken once per second). A specific task may only need 2% usage, if it's a fairly simple task for the P.U. to process. It's the time it takes to run each task that is measured... that is what is being rated, based off the time it took the base i7.
Click to expand...
Sure but that does not equate to real world usage then. I just offered a real world usage test that pegged all my CPU cores to 100% for like 7 minutes.

This is why I pretty much take any Geekbench results very lightly. Running quick little snippets of things that get done in the real world does not give a super accurate look at actual performance. Does not even take in account any throttling that might happen because the CPU or GPU is getting hammered for a long period of time.

And please do not take this post as a bashing of Apple Silicon, Apple has done an amazing job so far with it, the power/performance is very very good.

I would trust Cinebench more for CPU since I know it is going to hammer the CPU for a long time period.

This is more or less just an observation with what happens during the test. But I am still curious as to whether Apple Silicon behaves in the same way as Intel hardware, under Mac OS, using this benchmark suite.
 
avkills said:
Sure but that does not equate to real world usage then. I just offered a real world usage test that pegged all my CPU cores to 100% for like 7 minutes.

This is why I pretty much take any Geekbench results very lightly. Running quick little snippets of things that get done in the real world does not give a super accurate look at actual performance. Does not even take in account any throttling that might happen because the CPU or GPU is getting hammered for a long period of time.

And please do not take this post as a bashing of Apple Silicon, Apple has done an amazing job so far with it, the power/performance is very very good.

I would trust Cinebench more for CPU since I know it is going to hammer the CPU for a long time period.
Click to expand...

It does actually. Basing general purpose real-world benchmarks on highly optimized tasks is not indicative of real-world usage.

AMD and Nvidia GPUs have had raytracing hardware acceleration for a long time, ASi does not [yet]. This is how they pull those astronomical benchmark scores in some tests (along with power consumption). It would not be fair to base a general computing benchmark off such a bias. ASi has image, video and neural processors, etc. None of those were used in GPU testing even though the system would normally send those tasks to those specialized processors.

These tests aren't about "what's the fastest this system can run", it's about comparing similar tasks running across different systems. If you want more accurate scores for your particular workflow, find a benchmark that reflects that, and then compare the scores.
 
  • Like
Reactions: jdb8167 and EntropyQ3
obviouslogic said:
It does actually. Basing general purpose real-world benchmarks on highly optimized tasks is not indicative of real-world usage.

AMD and Nvidia GPUs have had raytracing hardware acceleration for a long time, ASi does not [yet]. This is how they pull those astronomical benchmark scores in some tests (along with power consumption). It would not be fair to base a general computing benchmark off such a bias. ASi has image, video and neural processors, etc. None of those were used in GPU testing even though the system would normally send those tasks to those specialized processors.

These tests aren't about "what's the fastest this system can run", it's about comparing similar tasks running across different systems. If you want more accurate scores for your particular workflow, find a benchmark that reflects that, and then compare the scores.
Click to expand...
Umm, Lightwave 3D is a CPU based rendering engine; so having GPU grunt is only helpful for viewport stuff or if you are offloading to a DCC GPU based rendering engine; but Lightwave is pretty much dead so it is kind of a moot point. It is just what I have used and owned for a long time for my basic 3D needs; and is something I know stresses the CPU.

My point is that Geekbench does not run long enough to stress the system at all (at least my system it seems.) Can we at least agree on that? 🙂
 
1676400512938.png

1676398545560.png


1676397532681.png


My systems: Let's see if the spec table will past properly:

Systems
NameFlight SimGameMPMSMBPiPPiPhP
DescriptionCustomCustomRack Mac ProMac Studio14" MacBook ProiPad Pro 12.9"iPhone 14 Pro
CPUAMD 5900X (12c)AMD 7900X (12c)Xeon W3245 (16c)M1 Ultra (16/4c)M1 Pro (6/2c)M1 (4/4c)A16 (2/4c)
GPUNvidia RTX 3080TiNvidia RTX 4900AMD Pro Vega IIM1 Ultra (48c)M1 Pro (16c)M1A16
RAM64GB RAM64GB RAM64GB RAM64GB RAM16GB RAM16GB RAM6GB
MotherboardGigabyte X570 Aurous MasterASUS ROG Crosshair X670EApple MPAppleAppleAppleApple
 
  • Like
Reactions: Leon Ze Professional and donth8
An issue with synthetic benchmarks is that it tests for performance that may not be identical to actual use cases of end users.

It being synthetic makes it neutral or crossplatform for ideally getting it apple to apple comparison but again it isn't day to day workflows that people actually do.

Something to think about when talking about Geekbench and other similar benchmarks.

Personally if I was a gamer I'd bench based on FPS or other triple A title.

Or be more focused on the performance of specific apps you actually use.

Thank you for attending my TED talk.
 
  • Haha
Reactions: avkills
Some interesting benchmarks. Very impressive that even an iPhone 14 Pro Max scores higher than the i7 in a Surface Pro 8 (and the Surface Pro 8 is certainly no slouch)

Surface Pro 8 - Intel Core i7-1185G7
Single Core: 1993
Multi-Core Score: 6510

iPhone 14 Pro Max
Single-Core Performance: 2498
Multi-Core Score: 6257

iPad Pro (11-inch, 4th generation)
Single-Core Performance: 2474
Multi-Core Score: 9357
 
GB6 Metal AMD Radeon RX 6900 XT eGPU score: 194703 (2017 iMac Pro.) Metal scores on even the latest hardware M2 ASi have a ways to go. Perhaps the M2 Ultra will match that score.
 
avkills said:
Umm, Lightwave 3D is a CPU based rendering engine; so having GPU grunt is only helpful for viewport stuff or if you are offloading to a DCC GPU based rendering engine; but Lightwave is pretty much dead so it is kind of a moot point. It is just what I have used and owned for a long time for my basic 3D needs; and is something I know stresses the CPU.

My point is that Geekbench does not run long enough to stress the system at all (at least my system it seems.) Can we at least agree on that? 🙂
Click to expand...
Geekbench's release notes indicate they're looking to provide more of an indicator of "real world" multicore performance rather than the raw potential of all of the cores, so the tasks the benchmark tests run are varied. CPU-based 3D rendering is something that scales incredibly well the more cores you have, whereas a sequence of filter tests in Photoshop does not saturate all of the cores. Part of me feels like there should be multiple multicore benchmark figures reflecting various workloads so that users can get a better sense for how a chip will perform for their specific needs.
 
  • Like
Reactions: jdb8167 and avkills
walkin said:
Interesting. Can someone please explain how the new benchmarks can still serve as a valid comparison with older hardware that has only run the previous version(s),
Click to expand...

They can’t. It’s a different test suite. You cannot meaningfully compare scores across versions.

walkin said:
Doesn’t changing the test defeat the very purpose of a benchmark?
Click to expand...

No. This is why it’s a major new release. 4, 5, and now 6 all changed up the suite.

walkin said:
Do we need to run the new benchmarks on old hardware before it can be compared with new hardware?
Click to expand...

Yes.
 
  • Like
Reactions: jdb8167 and walkin
So if an Intel i7-12700 is baseline (2500), then most of the M-series scores I'm seeing here are around the speed of an Intel i7? I thought M1/M2 killed Intel, performance-wise?
 
Interesting to see even A14X ( yes I prefer to call it by that name) is getting slightly higher single core score than A15. Normally it’s the other way around due to micro architecture
 
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