7-week wait for 14" M1 Max 24-core...does Apple wait for binned chips to occur naturally?

[Fomalhaut]

I'm still waiting to pull the trigger on a new MBP and have decided on the 14" M1 Max with 24-core GPU.

I've noticed that there is quite a long wait time for these - if I ordered today (8 Dec), the estimated delivery date is 21-31 January. That is 2-3.5 weeks longer than other BTO options that I've tried.

Do you think Apple waits for "natural binning" to sell machines with the 24-core M1 Max? i.e. they create these 24-core M1 Maxes from the 32-core versions that have manufacturing faults?

Or do they create some of these by purposely disabling otherwise functional cores, but limit the number of these "hobbled" SoCs based on the volume of the 32-core orders they have? i.e. if they can sell the more expensive SoC, they will, but if there is an excess of 32-core manufacturing capacity, they select a proportion of these "good" chips to turn into 24-core models?

 

[mr_roboto]

I'm still waiting to pull the trigger on a new MBP and have decided on the 14" M1 Max with 24-core GPU.

I've noticed that there is quite a long wait time for these - if I ordered today (8 Dec), the estimated delivery date is 21-31 January. That is 2-3.5 weeks longer than other BTO options that I've tried.

Do you think Apple waits for "natural binning" to sell machines with the 24-core M1 Max? i.e. they create these 24-core M1 Maxes from the 32-core versions that have manufacturing faults?

Or do they create some of these by purposely disabling otherwise functional cores, but limit the number of these "hobbled" SoCs based on the volume of the 32-core orders they have? i.e. if they can sell the more expensive SoC, they will, but if there is an excess of 32-core manufacturing capacity, they select a proportion of these "good" chips to turn into 24-core models?

Prior to starting up production, they had some estimate of how many M1 Max die should yield as 32-core GPU parts, and how many of the remainder could be saved as 24-core parts. Some pricing genius then tried to set the MBP price points to match 24-core demand to supply. If the price delta's really small, nobody will order the 24-core. If it's too large, everyone will order the 24-core. Somewhere in the middle things are "just right".

But it's slightly more subtle than that. Maybe it turns out preproduction yield estimates (based on defect density) were off. Or short term yield variance means there's a temporary glut or shortage of 24-core parts. Or their pricing genius made a mistake, customers are pinching pennies more than anticipated, and demand for the 24-core is too high.

The usual solution to this category of problem is the last thing you mention. When they set prices Apple probably aimed at perpetual shortage of 24-core parts, planning to make up the shortfall by using some 32-core parts as 24-core parts. Obviously they don't want to go too far down this path - they'd be leaving money on the table - but a minor perpetual shortage is preferable to ever naturally having too many 24-core parts, because those "true" 24-core parts can't be converted to 32-core. So the pricing genius needs to really finesse things. Probably has to convince management to sign off on the size of the error bars in their analysis.

I do not know for certain whether Apple is actually doing this, but it's standard practice for any chip supplier trying to maximize income per wafer when improving yield by selling some parts with reduced core counts / cache size / whatever.

 

[Gnattu]

Prior to starting up production, they had some estimate of how many M1 Max die should yield as 32-core GPU parts, and how many of the remainder could be saved as 24-core parts. Some pricing genius then tried to set the MBP price points to match 24-core demand to supply. If the price delta's really small, nobody will order the 24-core. If it's too large, everyone will order the 24-core. Somewhere in the middle things are "just right".

But it's slightly more subtle than that. Maybe it turns out preproduction yield estimates (based on defect density) were off. Or short term yield variance means there's a temporary glut or shortage of 24-core parts. Or their pricing genius made a mistake, customers are pinching pennies more than anticipated, and demand for the 24-core is too high.

The usual solution to this category of problem is the last thing you mention. When they set prices Apple probably aimed at perpetual shortage of 24-core parts, planning to make up the shortfall by using some 32-core parts as 24-core parts. Obviously they don't want to go too far down this path - they'd be leaving money on the table - but a minor perpetual shortage is preferable to ever naturally having too many 24-core parts, because those "true" 24-core parts can't be converted to 32-core. So the pricing genius needs to really finesse things. Probably has to convince management to sign off on the size of the error bars in their analysis.

I do not know for certain whether Apple is actually doing this, but it's standard practice for any chip supplier trying to maximize income per wafer when improving yield by selling some parts with reduced core counts / cache size / whatever.

Apple's MBP requires a new SKU for each RAM/SSD configuration and that stress the supply chain. There are 80 SKUs in total for the new MacBook Pros, the RAM and storage config does affect the BTO time, and you may need to wait a whole month more only by choosing an SKU with specific RAM/SSD size. I think the numerous amount of SKUs is a bigger factor than the SoC itself.

 

[UBS28]

Ofcourse Apple is waiting for binned versions of the 32-core version. Apple is not going to sell you a fully functioning 32-core version for less money without any reason.

 

[anthonymoody]

Just don't choose 64GB over 32 as that pushes the delivery estimate well into February.

 

[Fomalhaut]

Ofcourse Apple is waiting for binned versions of the 32-core version. Apple is not going to sell you a fully functioning 32-core version for less money without any reason.

I think this will depend on how many 32-core M1 Maxes they can sell. Presumably, this is a different production line to the M1 Pros (at least partially), so if they have the capacity but are not selling enough, then it is conceivable that they would deliberately create *some* 24-core M1 Maxes from fully functional 32-core units, just to satisfy customer demand for the 24-core variant.

 

[Fomalhaut]

Just don't choose 64GB over 32 as that pushes the delivery estimate well into February.

I would be getting 32GB, but this is good to know. Is this due to scarcity of RAM modules, or just because they produce fewer 64GB variants based on predicated sales?

 

[JPack]

It's extremely unlikely to be binning because as everyone knows, Apple can disable cores.

Apple priced their products in a way that they would not rely on natural binning to fulfill SKUs. Why would Apple let TSMC determine how quickly Apple can sell 24-core M1 Max to consumers? Apple has such a high margin, they don't need to do this.

Everybody knows the silicon shortage is not because of leading edge nodes. It's shortage of mature nodes, the MOSFETs, MLCCs, and PMICs that are causing problems. If Apple has only limited number of power components for MacBooks, they would put them in highest margin and highest volume SKUs, i.e. M1 Max 32-core and M1 Pro 14-core. Those are also pre-configured SKUs and they know they can move products. For 24-core, it's BTO, Apple can't stock it and so makes sense for it be back of the line.

 

[mr_roboto]

I think this will depend on how many 32-core M1 Maxes they can sell. Presumably, this is a different production line to the M1 Pros (at least partially), so if they have the capacity but are not selling enough, then it is conceivable that they would deliberately create *some* 24-core M1 Maxes from fully functional 32-core units, just to satisfy customer demand for the 24-core variant.

Same production line (s), different mask design for M1 Max and M1 Pro.

Chip fabs and their Apple-scale customers think in terms of wafers. There's a lot of manufacturing steps involved in processing each wafer, and many of them take substantial time on one machine. To keep throughput high, the fab has many copies of each type of machine so the processing steps can be run in pipelined fashion. It can take weeks or months for an individual wafer to make its way through all the stations which take it from raw silicon to finished and ready for test, so you'll often see a fab's capacity defined as the number of wafer starts it can handle each month, or WSPM for short.

Fabs are incredibly expensive to build, so new ones (particularly cutting edge nodes like TSMC 5nm) need to run at near 100% capacity for the first few years to guarantee the fab at least pays for itself. Since very few customers have WSPM demands which happen to add up to an integer number of fabs, TSMC designs in the flexibility to run multiple wafer designs in the same fab at the same time.

Thanks to all that, it's likely that M1, M1 Max, M1 Pro, and A14 wafers all get made in the same set of TSMC facilities on the same lines. It's on Apple to adjust the mix of wafer starts based on demand for each product.

 

[wilberforce]

Apple doesn't make any chips. They order them from TSMC.
I expect the order consists of: X number of chip A, Y number of chip B, Z number of chip C, and deliver them by XXX date(s) to Foxconn. (I imagine more complicated than that, but you get the idea).
TSMC does whatever they need to do to fulfill the order, and are not going to wait for chips to be defective to bin them.

I suspect the reason for the delay on this model is simply because the demand for this model is higher than anticipated, and it will take months to adjust.

 

[Fomalhaut]

Apple doesn't make any chips. They order them from TSMC.
I expect the order consists of: X number of chip A, Y number of chip B, Z number of chip C, and deliver them by XXX date(s) to Foxconn. (I imagine more complicated that that, but you get the idea).
TSMC does whatever they need to do to fulfill the order, and are not going to wait for chips to be defective to bin them.

I suspect the reason for the delay on this model is simply because the demand for this model is higher than anticipated, and it will take months to adjust.

Thanks for that. It's what I anticipated.

It does seem that the 14" M1 Max 24-core will be popular because it is a sweet-spot for portability, performance, and price. From reviews I've seen, there is only a very marginal improvement going to the 32-core 14" M1 Max, and probably a hit on battery life.

 

[curmudgeonette]

Do you think Apple waits for "natural binning" to sell machines with the 24-core M1 Max? i.e. they create these 24-core M1 Maxes from the 32-core versions that have manufacturing faults?

There can be other "bins" that we don't yet know about. It is likely that Apple is stockpiling other configurations. We'll eventually see them in the Mini, a big iMac, and the Mac Pro. That's a reason to delay introduction of additional models - the delay lets Apple find out the actual mix of chips. Perhaps the Mini will max out to only 28 cores. Maybe there'll be a Mac Pro with quad 16-core Max's (for the extra memory capacity but no extra GPU.)

 

[Fomalhaut]

Even in the one day since I posted this, the time has slipped back 4 days and into early February.

I've just placed an order because, at this rate, I could be waiting a long time if I delayed my decision any longer.

 

[Alameda]

I'm still waiting to pull the trigger on a new MBP and have decided on the 14" M1 Max with 24-core GPU.

I've noticed that there is quite a long wait time for these - if I ordered today (8 Dec), the estimated delivery date is 21-31 January. That is 2-3.5 weeks longer than other BTO options that I've tried.

Do you think Apple waits for "natural binning" to sell machines with the 24-core M1 Max? i.e. they create these 24-core M1 Maxes from the 32-core versions that have manufacturing faults?

Or do they create some of these by purposely disabling otherwise functional cores, but limit the number of these "hobbled" SoCs based on the volume of the 32-core orders they have? i.e. if they can sell the more expensive SoC, they will, but if there is an excess of 32-core manufacturing capacity, they select a proportion of these "good" chips to turn into 24-core models?

There are process variations which can occur, and the chip has to make timing across all of them. So it is possible to offer the highest speeds from a percentage of what is manufactured.

The second thing is that wafer manufacturing is a photo lithographic process. If the most minute particle of dust floats between the light and the wafer, it creates a failure in a specific spot on the chip. It is possible that if the failure is in one specific core, that core could be disabled. It boils down to what the yield rates are on these latest-generation fabs and what capabilities Apple designed into the design for disabling any of the eight cores. I have no idea. But the other possibility is that a part with more cores needs more pins, which means different package. Or, even if it’s the same package, power consumption will be higher with more cores enabled and that impacts the package.

My real point is that we just don’t know and we can only speculate. Package engineering is very complex, especially with a new process like this.