December 2, 2020

Intel’s Alder Lake May Ship in a Wide Range of Confusing Configurations

The reports that Intel’s Alder Lake will be a big.Little (or a big.Bigger CPU, if...

The reports that Intel’s Alder Lake will be a big.Little (or a big.Bigger CPU, if you use Intel’s parlance) got a boost with a list of coreboot patches that show the various configurations Intel is contemplating. Just because a part appears on this list doesn’t mean Intel will launch a SKU corresponding to it, mind you — these are just the configurations that are being contemplated.

There are 12 possible configurations for Alder Lake-S chips, helpfully summarized in the chart below by THG:

Image by THG

There’s also a 2-0-1 configuration I clipped off the above. This configuration of hardware is dramatically different from anything we’ve ever seen ARM ship. There’s an 8+8 configuration, which makes sense, but there’s no way to determine why Intel would ship a 6+8 core, or what the benefits of 8+2 versus 8+4 are. Keep in mind, these are all desktop chips — we’re not seeing mobile and desktop mixed together.

The fact that there’s an 8+8 configuration at the top of the stack implies that this is the fastest part. The 8,4,2,0 fallback makes sense, then — except it also implies that having no small cores on the CPU is a significant enough loss of performance to create a meaningful gap between parts. Does this mean we’ll see more of an emphasis on core count and less on clock speed with these chips, compared to previous launches? Which is faster — an 8+0 CPU or a 6+4? What about 8+2 versus 6+6?

THG also discusses a second set of results for Alder Lake-P, an apparently lower-power Atom CPU, but I’ll let you read about it over there if you have a mind to. The simplest explanation for this list of parts is that someone enumerated all the options Intel thought it might launch for simplicity’s sake, without bothering to select only the SKUs the company plans to launch ~12 months from now. Another is that small cores add very little to overall performance and we should assume that an 8+0 core is faster than a 6+8 core. This, however, would raise the question of why Intel is bothering to build an 8C small-core cluster at all.

A third option is that while the small cores are expected to be useful, they may not be useful to all groups of users. Maybe when it comes to gaming, the OS is instructed to run all workloads solely on the high-end cores, to ensure that maximum performance is always available. This would explain why an 8+0 CPU might rank higher than a 6+8, even if the 6+8 offered better scaling in some multi-threaded tests. Gaming is the area where Intel CPUs are performing the best these days, and the company may have aligned its SKUs accordingly.

Lots of questions but not a lot of answers yet. According to Intel’s statements regarding 10nm, Cannon Lake shipped on 10nm, while Ice Lake is 10nm+. Alder Lake is expected to debut on Intel’s 10nm++ process, which should have better characteristics for desktop performance and offer higher clocks than what we saw with Ice Lake. But this core configuration question is interesting — it’s obvious that Intel wants to hit higher core counts and boost efficiency, but how they’ll use this new configuration to do it is a very open question we don’t know much about yet.

Now Read:

Source link