KingOfPain
Site Champ
- Posts
- 270
- Reaction score
- 357
M4 Rumors (requests for).
- Thread starter Jimmyjames
- Start date
Confirming that I was just saying Apple seems to have no ambitions of going into the big-iron market.
They do seem to be supporting developers doing smaller training jobs on their Macs, so if better GEMM in the GPU doesn't add too much area or power, it's probably something they're interested in.
The original article on Bloomberg ( https://www.bloomberg.com/news/articles/2024-04-11/apple-aapl-readies-m4-chip-mac-line-including-new-macbook-air-and-mac-pro?leadSource=uverify wall ) actually predicts 4 tiers, like we have now. 9to5Mac just didn't do a good job encapsulating it:
"[1] The Donan chip is coming to the entry-level MacBook Pro, the new MacBook Airs and a low-end version of the Mac mini, while [2] the Brava chips will run the high-end MacBook Pros and a pricier version of the Mac mini. For the Mac Studio, Apple is testing versions with both [3] a still-unreleased M3-era chip and a variation of the M4 Brava processor. The highest-end Apple desktop, the Mac Pro, is set to get [4] the new Hidra chip."
Last edited:
That makes more sense
. It does still seem like they’re changing up the product tiers further with the high end Macbook Pro and mini getting the same chip and a special Mac Pro chip - all of which sounds potentially good! Could still see the proposed merging of the Pro/base die in my speculative post based on that … but now we might getting a high Brava variant for Studio and low Brava variant for the MacBook Pro/Mini. Other possibilities abound of course. If true, looking forward to it! Naturally, as always, the usual caveats apply to leaks.Thanks for clarifying, couldn’t read the original.
I can read the power on newsletter:
www.bloomberg.com
The two interesting tidbits here that we may not get high end M3 Ultras* - another possible reason why the interconnect is missing this generation - and the prospective timetable of releases with the idea that Apple is indeed going to try to move to yearly release schedules when possible but skip certain models/variants.
So in his schema Donan is the base chip and Brava is the combined Pro/Max die** but with a high end version to replace the Max and a low end version to replace the Pro. But it’s interesting that he just states the Studio has some kind of unspecified high end M4 chip but the Mac Pro gets the Ultra “Hidra” chip. It just feels like if his code names/die schema are accurate, then the old base/Pro/Max/Ultra levels may not map cleanly to the M4 generation. We’ll see of course. The first clue will be the A18/pro in the fall.
Edit: *I thought they would’ve wanted to get their ray tracing capability into their workstations ASAP (and Gurman may be wrong), but it’s possible they don’t think there’s enough upgrade potential (odd) and the combination of AI and ray tracing capabilities coming with the M4 makes for a better upgrade.
**this could also reflect a change back to the pre-M3 Pro/Max die where the M1/2 Pro was a chopped design of the Max rather than a core binned variant within a die but they might still be referring to them by the same code name if one is the same as the other with the bottom chopped off.
The AI Device Revolution Isn’t Going to Kill the Smartphone
The technology industry is betting on AI to create alternatives to the iPhone. But testing one high-profile attempt — the Humane Ai Pin — reveals that smartphones aren’t going anywhere. Also: A Mac makeover is underway; Beeper gets acquired for $125 million; an Apple store attempts to unionize...
www.bloomberg.com
The two interesting tidbits here that we may not get high end M3 Ultras* - another possible reason why the interconnect is missing this generation - and the prospective timetable of releases with the idea that Apple is indeed going to try to move to yearly release schedules when possible but skip certain models/variants.
So in his schema Donan is the base chip and Brava is the combined Pro/Max die** but with a high end version to replace the Max and a low end version to replace the Pro. But it’s interesting that he just states the Studio has some kind of unspecified high end M4 chip but the Mac Pro gets the Ultra “Hidra” chip. It just feels like if his code names/die schema are accurate, then the old base/Pro/Max/Ultra levels may not map cleanly to the M4 generation. We’ll see of course. The first clue will be the A18/pro in the fall.
Edit: *I thought they would’ve wanted to get their ray tracing capability into their workstations ASAP (and Gurman may be wrong), but it’s possible they don’t think there’s enough upgrade potential (odd) and the combination of AI and ray tracing capabilities coming with the M4 makes for a better upgrade.
**this could also reflect a change back to the pre-M3 Pro/Max die where the M1/2 Pro was a chopped design of the Max rather than a core binned variant within a die but they might still be referring to them by the same code name if one is the same as the other with the bottom chopped off.
Last edited:
B01L
SlackMaster
The latest from everyone's favorite YouTuber...! ;^p
Jimmyjames
Site Champ
- Posts
- 678
- Reaction score
- 766
We need to have a little talk about the word “confirmed”.
B01L
SlackMaster
You mean cOnFiRmEd...?!?
Yeah, I thought to myself "wait a minute..." when he said confirmed...
But I do think a desktop-specific Mn Ultra (and Mn Extreme) sounds like a good idea...
Indeed I give his video an eye roll because he just can’t help but go overboard on “confirmed” vs “speculation/supported/leaked” but desktop specific dies ARE a good idea (and something I’ve been talking about for awhile so it’d be churlish of me to dismiss his ideas on it). I don’t think he’s got it quite right but then Gurman was a bit vague on how the different chips stack up. Like I agree that the Hidra is almost certainly a fusion chip and his analysis of the name from mythology is spot on. However the high end Brava might not map well to the current Ultra - just losing efficiency cores and neural engine and extra media engines (though seriously those aren’t useful I thought they were?) might not be enough to bring the monolithic die size down and we “know” the high end Brava and low end Brava share a common design as they have a common name. So is the unnamed high end M4 chip for the Studio in Gurman’s list the same as the high end Brava? If so then how does the low end Brava and Donan relate to the base M4/M4 pro/M4 Max? If the high end Brava is not Studio-specific and is also shared with the MacBook Pro then does the upper Studio get demoted? Or is there another chip in here? Gurman talks about code names and chips but doesn’t really relate the two together and things get a bit murky.
Also “best consumer Nvidia chip” and shows a 4070
- I mean from a perspective sure but the M4 Hidra better be beating the x070 whenever it is released if it comes - that should be a given. I’ll grant you that in raw flops Apple’s chips are a touch low but we’ve discussed on these forums how Apple makes up for that for most GPU work that gets done (especially rasterization) and, given the price, in actual rendering and so forth they had better be beating a 4070 or 5070 in performance if they release such a monster. The Hidra better be better than the x080 if not outright competitive with the x090.
Last edited:
Interesting speculative analysis on Macrumors with regard to LPDDR 6, base RAM, and bandwidth:
forums.macrumors.com
Of course impossible to know if it’s right but it seems well thought out - especially the ending comment that “Apple will jump to 12GB and then keep it for waaaaaay too long”.
What do you all think?
LPDDR6: New Beginning of AS: 12GB Standard
Question: Have you wonder why Apple is rumored to launch M4 series by the end of the year? This round Apple will launch M4 from top to bottom including Mac Mini within a quarter period, why? Mark Gurman can't tell you, but I can. The key ingredient missing is LPDDR6. :cool: Before I...
forums.macrumors.com
Of course impossible to know if it’s right but it seems well thought out - especially the ending comment that “Apple will jump to 12GB and then keep it for waaaaaay too long”.
What do you all think?
B01L
SlackMaster
Donan = M4
Low-end Brava = M4 Pro
High-end Brava = M4 Max
Hidra = M4 Ultra & M4 Extreme
Regarding desktop specific SoCs; losing efficiency cores & extra media engines makes sense, but cutting Neural Engines does not...
Apple's AI push will most likely focus more on the Neural Engine arrays, and I believe I had read somewhere about Apple actually doubling the number of Neural Engine cores in the M4-series of SoCs...?
Regarding GPU power, I am hoping the target for the M4 Extreme is to match and/or beat a brace of 5090 GPUs...?
Low-end Brava = M4 Pro
High-end Brava = M4 Max
Hidra = M4 Ultra & M4 Extreme
Regarding desktop specific SoCs; losing efficiency cores & extra media engines makes sense, but cutting Neural Engines does not...
Apple's AI push will most likely focus more on the Neural Engine arrays, and I believe I had read somewhere about Apple actually doubling the number of Neural Engine cores in the M4-series of SoCs...?
Regarding GPU power, I am hoping the target for the M4 Extreme is to match and/or beat a brace of 5090 GPUs...?
A definite possibility but then the M4 Studio is getting demoted with no Ultra chip since Gurman claims the Hidra is Mac Pro specific. It's possible with the rework the old base/Pro/Max/Ultra/(Extreme - in parentheses since it never got developed) levels just don't really apply to the M4, those relationships will all be different - even if the names are still used, what they mean and which computers get which dies might be substantially different. Then again, it's possible Gurman is just missing a codename either at the low-end or high-end to make it all work as before or is wrong about the Hidra being Pro specific or something else.
Correct me if I'm wrong, but even in the current Ultra one of the two Neural Engines is disabled, is it not? It's possible two neural engines can't really work together, but maybe that'll change? I'm pretty sure he doesn't mean making the Neural Engine smaller, he means ensuring that a doubled die doesn't have two or extraneous Neural Engines, but then doesn't really work unless the two dies aren't symmetric so ... yeah not sure what he's thinking there. Unless Im getting confused I'm pretty sure the M2 Ultra only has two ...
My impression was that with media engines, the more the merrier for people who do video work - like being able to encode/decode x different streams simultaneously. But that's far enough outside my wheelhouse that perhaps I misunderstood.
We can hope ... but I think that might be a bit much (depending on the task). Matching a 5090 or at least sitting between a 5080 and an 5090 for most tasks but with oodles of memory to make those tasks requiring massive VRAM pools run faster than a pair of 5090s, that's definitely possible.
Jimmyjames
Site Champ
- Posts
- 678
- Reaction score
- 766
More is definitely better when it comes to media engines. It allows you to either encode multiple files simultaneously, or since Sonoma on the Ultra, split one encode between the engines.
B01L
SlackMaster
Tech specs shows the M2 Ultra Mac Studio with a 32-core Neural Engine, and the M2 Max Mac Studio with 16-core Neural Engine...
I believe it was the original M1 Max SoC that had a die shot/floor plan image showing what appeared to be two 16-core Neural Engines; with one labeled as a Neural Engine and the other not labeled at all...?
Forgive me if the below post is a bit inarticulate/confused, I'm a bit tired. But one way to square all this (and this may have been what @B01L was going for in his schema above) is if Gurman is partially wrong and the codename Hidra refers to both the desktop base die which will go into the Studio and the fused die which will go into the Pro. We're starting to get into Kremlinology here, but Gurman never named the "high end" M4 chip going into the high end Studio (the low end will presumably get the high end MacBook Pro chip as it currently does). MaxTech took that "high end" M4 to mean the high-end Brava but that doesn't necessarily work all that well.
So let's say Hidra refers to not one but two chips, the base and fused die. But honestly MaxTech's cutting of features isn't far enough in my opinion if they want to make an economical monolithic die which means the new "Ultra" is going sit somewhere in between the current Max/Ultra schema. Right now two Maxes is basically close enough to the reticle limit that an Ultra would be too expensive to mass produce. The current chip we know closest to TSMC's reticle limit is the H100 and supposedly it costs a few thousand to manufacture which could include other costs like memory but die fab is going to be the lion's share. I talk about that in an earlier post. I'm basing some of this off of 3rd party estimates that I have no way of verifying, but moving forward, assuming I'm right then they're going to have slim the current Ultra down even more than he suggests - especially if they want to add more IO to the base die to benefit the Pro's internal expansion. That does still mean a demotion of capabilities for the high-end Studio (though not communicating across an interconnect would bring its own benefits), though not as bad as the Studio maxing out at the Max, and a big increase in capabilities for the Mac Pro. Not sure how pricing will change ... but again this what I'm talking about that the current schema may not neatly map onto the schema going forward if these leaks are true.
I don't think losing efficiency cores makes much sense. These occupy very little space and offer similar perf/mm2 performance as the P-clusters. Also, they are great for running auxiliary tasks, freeing P-cores for higher-priority work.
thenewperson
Member
- Posts
- 17
- Reaction score
- 15
Is LPDDR6 even ready to be used in anything launching in October? This same guy has been over on the Anandtech forums peddling this same thing and no one's buying it much. I think they'll just go to LPDDR5X as expected.Interesting speculative analysis on Macrumors with regard to LPDDR 6, base RAM, and bandwidth:
LPDDR6: New Beginning of AS: 12GB Standard
Question: Have you wonder why Apple is rumored to launch M4 series by the end of the year? This round Apple will launch M4 from top to bottom including Mac Mini within a quarter period, why? Mark Gurman can't tell you, but I can. The key ingredient missing is LPDDR6. :cool: Before I...
Of course impossible to know if it’s right but it seems well thought out - especially the ending comment that “Apple will jump to 12GB and then keep it for waaaaaay too long”.
What do you all think?
casperes1996
Power User
- Posts
- 186
- Reaction score
- 173
What is the timeframe for LPDDR6? If it's expected to be ready by early-mid next year I could see M4 having a memory controller that can use either and then higher end Macs getting LDPRR6 later with base M4 (maybe Pro too) launching earlier with just LPDDR5X. They've done similar in the past after all where only some chips in a generation got LPDDR5 and some had LPDDR4X. Or maybe it was 4 and 4X... Can't quite remember
oh I thought it was going to be ready? but I'll admit I didn't check, maybe my timeline is off. A pity if not.
They did that with the M1 and I believe it was LPDDR4X and 5 though of course the MR poster is positing 6 to help justify an expansion to 12GB base RAM in the M4 base though LPDDR6 isn't strictly necessary for that.
thenewperson
Member
- Posts
- 17
- Reaction score
- 15
Oh yeah, the M1 (LPDDR4X) vs M1 Pro/Max (LPDDR5) split. I could see that. There are reports like this which show that even though LPDDR6 hasn't been certified there seems to be accelerated interest in it so maybe TigerRick is making some points
