M3 Ultra Chip is Only 10% Faster Than M4 Max in Benchmark Results

[MacRumors]

The first alleged benchmark result for Apple's new M3 Ultra chip has surfaced in the Geekbench 6 database tonight, allowing for more performance comparisons. The high-end chip is available in the new Mac Studio, introduced earlier this week.

Apple said the M3 Ultra chip is the "highest-performing chip it has ever created," and the unverified benchmark result seems to confirm that. In the single result, the 32-core M3 Ultra chip achieved a multi-core CPU score of 27,749, which makes it around 8% faster than the 16-core M4 Max chip that previously held the performance record. The result also reveals that the M3 Ultra chip is up to 30% faster than the 24-core M2 Ultra chip.

As expected, the M4 Max chip tops the M3 Ultra chip in terms of single-core CPU performance by nearly 20%, according to the result. This is due in part to the M4 Max chip being manufactured with TSMC's second-generation 3nm process, whereas the M3 Ultra is likely based on TSMC's first-generation 3nm process.

We now await additional M3 Ultra benchmark results to see if these scores are accurate, as they seem to be on the lower side compared to what was expected. For example, Apple advertised the M3 Ultra chip as being up to 1.5x faster than the M2 Ultra chip, so that 30% increase mentioned above should seemingly be closer to the 50% mark. Apple never said how the M3 Ultra chip's performance compares to the M4 Max chip, though.

As always, real-world performance may vary somewhat, but synthetic benchmark tools like Geekbench 6 provide a useful baseline for comparisons.

Watch this space, as we would not be surprised if additional Geekbench 6 results for the M3 Ultra chip end up having higher performance scores.

The benchmark was spotted by @jimmyjames_tech and shared by Vadim Yuryev.

Update: Three more M3 Ultra results have surfaced in the Geekbench 6 database, and the average multi-core CPU score has increased to 28,160. This means the M3 Ultra chip is around 10% faster than the M4 Max chip, up from the original 8% figure. Overall, it looks like the M3 Ultra chip is indeed not much faster than the M4 Max.

Article Link: M3 Ultra Chip is Only 10% Faster Than M4 Max in Benchmark Results

 

[nathansz]

I thought the point was more gpu cores?

 

[Neil J. Squillante]

Apple advertised the M3 Ultra chip as being up to 1.5x faster than the M2 Ultra chip, so that 30% increase mentioned above should seemingly be closer to the 50% mark. Apple never said how the M3 Ultra chip's performance compares to the M4 Max chip, though.

1.5x means 150% faster, not 50%.

 

[joeblow7777]

1.5x means 150% faster, not 50%.

By that math 1x faster (ie. the same speed) would be 100% faster. Am I missing something?

 

[EugW]

1.5x means 150% faster, not 50%.

Apple's language use here is not accurate.

1.5X = 50% faster or 150% as fast, but not 150% faster.
2.5X = 150% faster.

 

[Chancha]

By that math 1x faster (ie. the same speed) would be 100% faster. Am I missing something?

It is a linguistic ambiguity issue. If someone said "1.5x as fast" it is definitely a 150% multiplier, so "50% faster". But the article above chose to say the "x" multiplier and "faster" in the same sentence.

 

[Plutonius]

Didn't everyone expect the M4 to be faster in some cases ?

I'm assuming that the M3 Ultra was a fallback for Apple.

 

[svish]

Interesting. Waiting to see more tests/information. The chip should be more than enough for all the tasks but thought that it will be faster in benchmark numbers.

 

[rhodesy22]

Yeah, this is why I ended up selling the M1 Ultra and then the M2 Ultra, and I won't be buying any more Ultras. The next gen Max chip kept coming out soon after. So really this comes down to if you need more memory, higher memory bandwidth or more GPU cores. Software engineers should just stick with the M4 Max unless they have a specific use case for the extra AI prowess.

 

[Chancha]

I thought the point was more gpu cores?

There was hope that with the PSU TDP increasing from 370W to 480W, perhaps the Studio could overclock the M3 Ultra a bit to give it some advantage than the M3 Max inside a 16". But the 4.05GHz in the above GB6 benchmark record means there is no clocking difference at all.

And then the roughly 1.4x multiplier of M3 Ultra's multi-core score over the M3 Max means even scaling efficiency remains the same 70% as with the M2 Ultra gen.

So yes everything is within perfect prediction. Which is the disappointing part. But at least there is no regress, so.

 

[creediddy]

Some explaining is needed for $4k starting price.

 

[rp2011]

Yeah, it's important to see full benchmarks before pulling the trigger on this. So far an 8% improvement over the Max isnt' too impressive unless it's the memory speed you value.

 

[Return Zero]

I thought the point was more gpu cores?

Yeah, 2X GPU cores and also 4X max RAM which is perhaps an even bigger deal, especially since it’s unified (can all be VRAM).

 

[simonscheurer]

If you want to run large language models locally, then what you need is addressable RAM space for the GPU. Big drawback of the Nvidia cards is their small VRAM (while their tensor performance is still better by very, very far). For such a use case, the M3 Ultra may be a good option - way more GPU power and addressable RAM. With a 256 GB configuration you could run an 143B model with FP8 without an issue. But other than that I'd not see a lot of reason to go for an M3 Ultra.
I run models locally as well, quantizized models with 70B Params I can run on my 64GB Mac mini M4 Pro without a problem. I even ran bigger R1 models (thanks to MoE). Frankly for all usual use-cases as a developer that do not require huge training efforts or inference on massive local models, I never felt the need for more power.

 

[nutmac]

M3 Ultra seems to be mainly for getting more than 256 or 512 GB GB RAM and/or 16 TB SSD, or you want TB 5 ports on the front (instead of USB-C) and/or 5 or more external displays.

 

[VisceralRealist]

M4 seems to be a more significant jump in performance than the M3 was over the M2. Makes you wonder why they couldn't just attach two M4 Max chips and make an M4 Ultra.

 

[Quu]

I wonder if Apple going forward intends to do a tik-tock strategy with their chips. They mentioned the M4 Max lacks the fusion connector needed to make an M4 Ultra so perhaps it'll go like this:

M5 = Ultra Available
M6 = No Ultra
M7 = Ultra Available

That would allow them to upgrade the Mac Studio and potentially the Mac Pro every 2 years while the other products (especially laptops) that sell in higher volumes receive yearly updates with the latest architecture.

 

[GMShadow]

There was hope that with the PSU TDP increasing from 370W to 480W, perhaps the Studio could overclock the M3 Ultra a bit to give it some advantage than the M3 Max inside a 16". But the 4.05GHz in the above GB6 benchmark record means there is no clocking difference at all.

And then the roughly 1.4x multiplier of M3 Ultra's multi-core score over the M3 Max means even scaling efficiency remains the same 70% as with the M2 Ultra gen.

So yes everything is within perfect prediction. Which is the disappointing part. But at least there is no regress, so.

Did both models increase, or just Ultra equipped Studios?

 

[Hogswarts]

Anecdotal results shouldn't be trusted.

 

[amaze1499]

1.5x means 150% faster, not 50%.

No. Check your math.

 

[iSayuSay]

1.5x means 150% faster, not 50%.

The reference point is always 100%. Then you add or subtract based on performance difference. Here's an example: 1x means exactly same speed, then 1.2x means 20% faster, and 0.7x means 30% slower. If it's 150% faster, then it's 2.5x.

 

[amaze1499]

the 32-core M3 Ultra chip achieved a multi-core CPU score of 27,749, which makes it around 8% faster than the 16-core M4 Max chip

Does that mean an additional 16 cores only result in an 8% performance increase overall everywhere? So thinking about getting two more cores and paying a significant price for that upgrade is actually not translating into real difference in time and efficiency?

All I know for now is, that it’s getting more and more confusing to distinguish what upgrade is actually worth it. Either I’ll buy the basic version of the top of the pops. Apples product line so cluttered, one can’t say for sure whether the buying decision is a good one. Reference points for comparison have gotten way too narrow. The only obvious reason to upgrade the cores is to be able to get more RAM.

What does it mean to have an M4 in an iPad or a MacBook Pro or a Mac studio?

 

[picpicmac]

Some explaining is needed for $4k starting price.

As if there are already not plenty of explanations available?

Short answer: those TSMC wafers and packaging are not cheap.

 

[Chancha]

Did both models increase, or just Ultra equipped Studios?

It is unknown. maximum 480W power consumption is pulled from the Studio's official specs page, where this bit applies to both Max and Ultra models.

In the M1 and M2 Studios, the Max variants got a heatsink / heatpipe system without copper, but the PSU I think Apple didn't bother to put in one with a lesser rated wattage. But now they have a 480W one, perhaps the M4 Max variant PSU can stay at 370W .

 

[picpicmac]

This of course will lead so many who are just doing a number competition to (intentionally?) misunderstand what role each product has.

One buys the Ultra SoC for the following:
1) supports more displays;
2) double the video encoders;
3) more RAM.

The internet is full of people who simply want bragging rights (insert here long sociological discussions about male preening and posturing), and that is usually done based on My-Number-Is-Bigger-Than-Your-Number propositions.