According to geekbench 7800x3d is 2400 single core, 15500 multicore. M4 pro is 3350 single core and 24750 multicore. Yet when I convert video using libsvtav1 with ffmpeg I’m getting noticeably faster performance on the desktop. And that’s with mbp, which doesn’t thermally throttle within 15 seconds.
Is it the 96mb cache? Avx-512? Are benchmarks bullshit when comparing different architectures?
Which version of svt-av1 are you using? Neon optimisations were added quite recently in svt-av1, and are still a bit incomplete, so it might make a difference.
The software identifies the hardware, performs the benchmarks, and sends the heading and results to the Geekbench DB. Somebody who had the machine at hand must have run it.
Internal leaks, these happen frequently. Employees have access to these devices for months prior to launch. Someone ran a benchmark and uploaded it on a machine they weren't supposed to.
I find that hard to believe at Apple. Not even possible that it’s accidental. They have their own internal tools, all packets leaving the network will be interrogated, leaking data is a sackable offence… if it’s in the geekbench database then it’s there because they put it there.
> Not even possible that it’s accidental. They have their own internal tools, all packets leaving the network will be interrogated, leaking data is a sackable offence
This is a story that folks like to tell, but it's not really how things work - at least since they stopped letting Steve Jobs lock all his engineers in the conference venue.
Leaks are a sackable offence at every tech firm, deep packet inspection firewalls are used all over the place, information leaks like this still happen pretty much every single product cycle.
No, not the same. Those are review samples who got sent the product early. Those "leaks" happen AFTER the products have been officially announced by Apple.
Leaked prior to announcement - Leaked Benchmarks Confirm M2 Chip is Up to 20% Faster Than M1 https://www.macrumors.com/2022/06/15/m2-geekbench-benchmark/
Date of article: June 15, 2022 10:32 am PDT. M2 announcement was June 6.[0]
I don't feel like doing it for your other sources. They follow the same pattern. You can check official announcement dates vs "leaked" dates yourself.
That was a mistake and I updated it - only the M2 Max benchmarks leaked prior to announcement, not the M2.
Those M2 Max Geekbench benchmarks appeared on November 30, 2022. Apple announced the Apple M2 Max chip on January 17, 2023.
Similarly, those M4 Geekbench benchmarks for Mac were leaked on October 6, 2024, prior to Apple announcing the product. Apple announced the M4 Macs on October 28, 2024.
I noticed a whiff of panic at the iphone 18 launch. Opened macrumors and there like 6 articles in a row about preordering the phone, how soon you can get the phone when you preorder, how long the queues are etc.
Guess they didn't get the expected # of preorders.
I'm currently in the middle of porting a game from 2004 to run on MacOS via wine. On a 5000$ Macbook Pro, I'm getting about 4 FPS. This has multiple causes, including the game being single threaded and running through Rosetta, but damn is it hard to get good single core performance out of Macbooks.
Out of curiosity, I also ported the game to arm Linux with fex + upstreamed the wine patches in about a third the time and there's no FPS issues
Did you forget to tell Claude to make it run fast?
(I am not joking. I did a port of an old utility, Dos Navigator, and I gave Claude+Codex a set of KPIs: cold start under 100ms, 30MB RAW image preview under 50ms, and a few others. I went to bed, and I woke up to a file manager written in Swift that is a joy to use, its so fast, esp. when previewing images. If I haven't requested it, it would likely be way slower).
Running something with "Wine" doesn't usually mean "porting", "porting" means making a native binary that doesn't run through an emulation layer.
Anyway, probably you are trying to run a 32-bit executable with Rosetta, that won't be fast, because Rosetta was never meant to run 32-bit things and has got a lot of pitfalls.
If anything it's the exact opposite. They're ridiculously powerful with more memory bandwidth than pretty much any equivalent PC. It is easy to make sloppy software because you can get away with so much.
Something is going horribly wrong. I don't know what it is (without your code I don't know), have you profiled where it's going wrong? There's no way a game for 2004 would be that slow.
If you want help, I've done some similar work, you could pop it up on a github repo or something. What is the game?
What porting you talking about exactly? Packaging existing binaries? Compiling source code making it more compatible with Wine?
In my experience as game developer and porting C++ code tells such difference usually mean there are something really wrong with syncronization primitives. So 99.99% of the time code does nothing just waiting for something to unlock.
I recommend you try giving your AI an open ended task like, "it's at 4 fps even though it's a 22 year old game and the raw hardware performance is many orders of magnitude higher. Maybe due to running under Wine and Rosetta. Please figure out the simplest way you can unlock usable performance (30-60 FPS) short of decompiling, reverse engineering, and reimplementing the full game yourself natively. Maybe there is something else you can figure out for the performance unlock. The hardware is definitely there, so there has to be a way. The game is running so make sure you don't break it. Keep the same game behavior."
Regardless of which AI you're using, it will be able to figure something out in a few hours. 2004 pc's were 500-800 mhz, 128 MB RAM, and if they had a GPU at all it was 16-32 MB, 0.2-1 GLOP.
Your AI should be able to figure out a way to get it running in realtime.
M5 Ultra CPU:
https://browser.geekbench.com/search?k=parkdale_cpu&q=mac17%...
M5 Ultra GPU:
https://browser.geekbench.com/search?k=grand_gpu&q=mac17%2C1...
Base M6 CPU:
https://browser.geekbench.com/search?k=parkdale_cpu&q=mac18%...
Base M6 GPU:
https://browser.geekbench.com/search?k=grand_gpu&q=mac18%2C5
The Base M6 CPU single core is averaging a bit over 4000 on Geekbench 7.
For comparison, the AMD Ryzen 9 9950X3D2 averages 3161 on the same test.
https://browser.geekbench.com/processors/amd-ryzen-9-9950x3d...
The Intel Core Ultra 7 270K Plus averages 2940 on the same test.
https://browser.geekbench.com/processors/intel-core-ultra-7-...
According to geekbench 7800x3d is 2400 single core, 15500 multicore. M4 pro is 3350 single core and 24750 multicore. Yet when I convert video using libsvtav1 with ffmpeg I’m getting noticeably faster performance on the desktop. And that’s with mbp, which doesn’t thermally throttle within 15 seconds.
Is it the 96mb cache? Avx-512? Are benchmarks bullshit when comparing different architectures?
Bingo, each CPU is too unique with its own strengths and weaknesses to make broad statements, marketing picks up what they like and ignore rest
https://browser.geekbench.com/v7/cpu/singlecore
here are the leaderboards for anyone else who was curious
The software identifies the hardware, performs the benchmarks, and sends the heading and results to the Geekbench DB. Somebody who had the machine at hand must have run it.
The M7 generation has updated the tensor instructions in the GPU cores and they want to ship that ASAP.
I would love to see macOS replacing Windows in virtual desktop environments.
This is a story that folks like to tell, but it's not really how things work - at least since they stopped letting Steve Jobs lock all his engineers in the conference venue.
Leaks are a sackable offence at every tech firm, deep packet inspection firewalls are used all over the place, information leaks like this still happen pretty much every single product cycle.
Not to mention that Apple has a huge number of teams, and no absolute standards across them. People would be shocked at the things Apple DOESN'T do.
Then again, with the current state of their QA, I think it's becoming clear what Apple doesn't do much of anymore.
So this is different and faster than such a result in the old scoring would have implied.
AMD beats Intel, but only by a couple hundred points.
Qualcomm beats them both.
It feels like a crime upgrading from this thing tbh.
In the current times performance/price seems to be increasingly important.
(Not to mention several other factors that once did not even exist, or were irrelevant.)
This is actually the M6 Pro and not the M6. It has 18 CPU cores to the M6's 12. The scores line up as well.
Reports have said all along that Apple is skipping M6 Pro/Max to focus on M7 generation.
I can't remember the last time an Apple chip got leaked like this.
Apple has not mentioned M6 Pro anywhere.
Leaked prior to launch - Leaked Benchmarks Confirm M2 Chip is Up to 20% Faster Than M1 https://www.macrumors.com/2022/06/15/m2-geekbench-benchmark/
Leaked prior to announcement - 'M2 Max' Geekbench Scores Leak Online, Revealing Rumored Specs and Performance https://www.macrumors.com/2022/11/30/m2-max-chip-geekbench-r...
Leaked prior to launch - M3 Max Chip Around as Fast as M2 Ultra in Early Benchmark Results https://www.macrumors.com/2023/11/01/m3-max-chip-benchmark-r...
Leaked prior to announcement - Mac M4 Chip Performance Unveiled in First Benchmarks https://www.macrumors.com/2024/10/07/m4-mac-chip-benchmarks/
Leaked prior to launch - M5 Chip Achieves Impressive Feat in 14-Inch MacBook Pro Speed Test https://www.macrumors.com/2025/10/17/m5-chip-macbook-pro-gee...
Leaked prior to launch - Apple's M5 Max Chip Achieves a New Record in First Benchmark Result
https://www.macrumors.com/2026/03/05/m5-max-geekbench-benchm...
Then add this one to the list.
I don't feel like doing it for your other sources. They follow the same pattern. You can check official announcement dates vs "leaked" dates yourself.
[0]https://www.apple.com/newsroom/2022/06/apple-unveils-m2-with...
Those M2 Max Geekbench benchmarks appeared on November 30, 2022. Apple announced the Apple M2 Max chip on January 17, 2023.
Similarly, those M4 Geekbench benchmarks for Mac were leaked on October 6, 2024, prior to Apple announcing the product. Apple announced the M4 Macs on October 28, 2024.
Guess they didn't get the expected # of preorders.
Out of curiosity, I also ported the game to arm Linux with fex + upstreamed the wine patches in about a third the time and there's no FPS issues
(I am not joking. I did a port of an old utility, Dos Navigator, and I gave Claude+Codex a set of KPIs: cold start under 100ms, 30MB RAW image preview under 50ms, and a few others. I went to bed, and I woke up to a file manager written in Swift that is a joy to use, its so fast, esp. when previewing images. If I haven't requested it, it would likely be way slower).
Anyway, probably you are trying to run a 32-bit executable with Rosetta, that won't be fast, because Rosetta was never meant to run 32-bit things and has got a lot of pitfalls.
If you want help, I've done some similar work, you could pop it up on a github repo or something. What is the game?
In my experience as game developer and porting C++ code tells such difference usually mean there are something really wrong with syncronization primitives. So 99.99% of the time code does nothing just waiting for something to unlock.
For me it's like they built a Ferrari with 4 thin bicycle tires. What a waste.
I recommend you try giving your AI an open ended task like, "it's at 4 fps even though it's a 22 year old game and the raw hardware performance is many orders of magnitude higher. Maybe due to running under Wine and Rosetta. Please figure out the simplest way you can unlock usable performance (30-60 FPS) short of decompiling, reverse engineering, and reimplementing the full game yourself natively. Maybe there is something else you can figure out for the performance unlock. The hardware is definitely there, so there has to be a way. The game is running so make sure you don't break it. Keep the same game behavior."
Regardless of which AI you're using, it will be able to figure something out in a few hours. 2004 pc's were 500-800 mhz, 128 MB RAM, and if they had a GPU at all it was 16-32 MB, 0.2-1 GLOP.
Your AI should be able to figure out a way to get it running in realtime.