Interesting.
Future Mac Pro may use Apple Silicon & PCI-E GPUs in parallel | AppleInsider
Despite Apple Silicon currently working solely with its own on-board GPU cores, Apple is researching how to support more options, like PCI-E GPUs, all working in tandem.appleinsider.com
Either overt clickbait or someone seriously did not actually read the patent at all for comprehension.
'slot' is not a euphemism for PCI-e slot in this document.
In the overview of related are on page 1 ( of first patent linked 20230050061 )
" ... As process technologies shrink and GPUs become more powerful they may contain large number of shader cores. Software or firmware may provide units of work to be performed , referred to as "kicks" . Data master circuitry (...) may distribute work from these kicks to multiple replicated shader cores over a communication fabric. ..."
That communication fabric is inside the chip package. ( see figure 26 of patent).
On page 4 , section [0048] , 230
" ... These hardware slots may also be referred to herein as "dSlots" . Each subunit may include multiple hardware slots. 230 Sub units may also be refered to herein as "mGPUs ... "
On page 4 , section [0112] "Slot Reclaim "
"... In some embodiments , control circuitry is configured to allow a logical slot to reclaim a hardware slot that is assigned to another logical slot ...
... only hardware slots in the emptied state are eligible to be reclaimed by another logical slot"
What they are talking about there is that there are "slots" into which can insert work into inside the GPU subunits. mGPU isn't 'multiple GPUs". The 'GPU' here is closer to being what the Apple marketing literature calls a GPU core.
The appleinsider article pulls figure 6 out from the patent into a mostly made up alternative universe context. Where each vertical 'swim lane' in that diagram starts with a "Control Stream Processor" . You can think of that as what the marketing calls a GPU core. Once know that a mGPU is a sub-unit can also put this into a the proper contet that these are subcomponents of a greater whole.
What this patent is far more so about is how Apple can provide a bit of an abstract via a logical (i.e., virtual) kick unit mappings. For example can make it so that app1 , app2 , app3 get their own logical unit 'slots' to put work into. The kick manager here is working to assign it out a large number of mGPU units in some coherent , productive way. This job gets harder as grow the number of GPU cores ( and respective subunits inside those 'cores' ; ALUs , matrix , etc. ) larger and larger. It looks a little like how register renaming works. the opcodes refer to register3 (a logical/virtual name) , but dynamically that is 'renamed'/'mapped' to an internal register inside to processor with another physical number. Virtual workload slots are dynamically 'renamed/'mapped' to internal physical workload slots. When the internal workload is down then have to remap the results back to the logical/virtual address.
This is more so about 'really big', single GPU spread out over a chip die (or chip package) . Not Multiple extremely discrete GPUs.
The article mentions several other Apple patents today. I couldn't find one that has anything to do with discrete GPU topics/coverage. If he has got some specific numbers others, maybe there is something there that is closer to the 'story' being told here. ( third Affinity patent. same out of context hocus pocus where patent is talking about same mGPU (subunits) and Kick slot manager issues. )
Last edited:
