But these won't be simple modifications either! So we will need to run some experiments, see what kind of perf gains we can get and weight the dev effort against it. The specific problem discussed here (which is "why doesn't the M1 Ultra render exactly twice as fast as the M1 Max") does not belong to the unrealistic category. And, in yet other cases, the optimizations are deemed unrealistic given the existing architecture of Redshift. In some other cases, the optimizations might require more work and are pushed further down the line. In some cases those communications result in discussions about possible optimizations that are fairly easy/straightforward - so we go ahead and do them immediately. For this reason, we try to keep close communications with the hardware vendors. What I can say (and this is not specifically about Apple or the M1 Ultra but a general comment about our dev philosophy) is that we want Redshift to run as efficiently as possible on all the GPUs it supports today - or the ones it will support in the future. I'm afraid it will be very hard (or, rather, impossible) to comment on such deep technical details due to a number of NDAs.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |