#
✕ nick:chadmed
#asahi · 2021-10-23 11:22
chadmedno problem lol
#asahi · 2021-10-23 11:22
chadmedwhat _is_ magic in chip design is how intel has managed to get 6-way decode on their upcoming alder lake chips
#asahi · 2021-10-23 11:21
chadmedmort_: its almost entirely impossible to do that with an amd64 chip
#asahi · 2021-10-23 11:20
chadmedits a segment almost entirely under the yoke of microsoft and intel, who set virtually all the standards that the devices in this segment are built around
#asahi · 2021-10-23 11:20
chadmedand therein lies why the consumer desktop segment has been so hard to crack for arm (or any x86 competitor from the release of NT onwards)
#asahi · 2021-10-23 11:19
chadmedwindows is completely worthless* ftfy
#asahi · 2021-10-23 11:17
chadmedthings seems to have been heading this way for a while. the consumer desktop market is really the last one in which arm hasnt made any significant inroads and the biggest problem for it has always been the chicken-and-egg paradox. apple can resolve this by being both the chicken and the egg in thaat they control the software and hardware
#asahi · 2021-10-23 11:15
chadmedjust what having infinite money and a perpetual arm architectural license can do
#asahi · 2021-10-23 11:15
chadmedyeah theyre not. theyre well designed chips and a textbook implementation of a highly integrated heterogeneous soc, but magic they are not
#asahi · 2021-10-23 11:13
chadmedyeah though im having more and more trouble keeping it casual. if australian unis had better computer science degrees i 100% would have done one already
#asahi · 2021-10-23 11:11
chadmed ahahahah nope im a medical researcher right now by trade :P
#asahi · 2021-10-23 11:10
chadmedsome golden samples with very very good b-die DIMMs managed to hold 3400CL16 but this was rare until zen 3
#asahi · 2021-10-23 11:09
chadmedbenchmarked memory access times were also absolutely horrible, and the memory controller in those microarchitectures could not train and hold anything above 3200Mbps on the memory bus, even with loosened timings
#asahi · 2021-10-23 11:08
chadmedno design details im afraid, but well documented experimentation from press outlets (aand my own) shows that zen, zen+ and zen 2 cores are all extremely starved for data and that tuning tertiary memory timings provided a significantly higher performance boost than did a core frequency overcloc
#asahi · 2021-10-23 11:07
chadmedyessir
#asahi · 2021-10-23 11:06
chadmedi do believe the IO die introduced some, but it was miniscule compared to the penalty incurred by the imc design itself\
#asahi · 2021-10-23 11:05
chadmedi dont think it was the IO die as much as it was their shithouse IMC. zen/zen+ had utterly atrocious memory latency too
#asahi · 2021-10-23 11:03
chadmedunless amd has a patent on the very concept of offloading IO to a separate die to optimise yields then frankly i have no idea why you wouldnt do that for every design going forward. of course its likely that these chips have been in at development since before amd proved the concept.
#asahi · 2021-10-23 11:01
chadmedyeah but the soc itself is like 60bn xsistors
#asahi · 2021-10-23 10:59
chadmedsurprised they didnt go MCM from the get go for these things, its not like the concept is experimental or anything
#asahi · 2021-10-23 10:56
chadmedtheyll probably have a way. even if the current m1 max is low volume once they start chucking it in more devices theyll need a way to minimise the amount of wasted wafer
#asahi · 2021-10-23 10:54
chadmedand yeah apple has infinite money
#asahi · 2021-10-23 10:54
chadmedwell its basically the nvidia gt200 strategy except this time it worked because this chips arent crap
#asahi · 2021-10-23 10:51
chadmedmort_: you have a point in that theyre not selling chips, i believe the original context was me explaining to citizen1 why these chips arent exactly "magic" -- apple doesnt care about minmaxing perf/mm^2 and as such can make huge chips that other companies would be eating a fat loss on
#asahi · 2021-10-23 09:56
chadmedyeah i just meant on the actual chip itself. the dies are quite large
#asahi · 2021-10-23 02:55
chadmedactually i should retraact my initial statement. actuaally bothering to design a high performance consumer-class ARM chip is in and of itself something special unfortunately lmao
#asahi · 2021-10-23 02:52
chadmeds/nod/node
#asahi · 2021-10-23 02:52
chadmedtheres no magic, its just the logical outcome when you pour billions and billions and billions of dollars into minmaxing perf/W in a very well designed microarchitecture implementing a very well designed ISA on the worlds most advanced process nod
#asahi · 2021-10-23 02:51
chadmedcitizen1[m]: apple's not really doing anything special. they have ~10 years of designing architectures that maximise perf/W as much as humanly possible, they have virtually infinite money and the desire to design a high performance heterogeneous soc. as already said too, the chip is absolutely enormous and it is highly unlikely apple makes a profit on them.
#asahi · 2021-10-22 06:37
chadmedthats genuinely impressive wow
#asahi · 2021-10-22 06:23
chadmedyeah we were saying this before. dunno what sort of deal with satan intel did to get to 5-way decode for alder lake
#asahi · 2021-10-22 05:33
chadmedthey hold themselves captive to intel and amd now, but thats only because until recently There [Was] No Alternative(tm)
#asahi · 2021-10-22 05:33
chadmedif itanium came out when it was meant to maybe it would have been competitive. dont get me wrong, im thankful it didnt but a part of me wonders what wouldve happened to the PC had it succeeded as intel and hp planned. would people have held themselves captive to intel and hp and continued to use the platform?
#asahi · 2021-10-22 05:26
chadmedyeah intel almost screwed the pooch there with netburst's marianas trench of a pipeline
#asahi · 2021-10-22 04:21
chadmedsuch that there is no longer any meaning to terms like risc and cisc because the canonical way to design a cpu core has converged so much that architectures that we once called risc and cisc now look virtually identical at the hardware level
#asahi · 2021-10-22 04:19
chadmedi think i may have already mentioned this here a while ago but the progress of hardware design shares many similarities with convergent evolution in biology
#asahi · 2021-10-22 03:10
chadmedright
#asahi · 2021-10-22 03:09
chadmedcrazy how windows has only just recently received support for addressing hardware schedulers in GPUs, and only through dx12u
#asahi · 2021-10-22 03:06
chadmedyeah the last good one was terascale i think, and all radeon cards have hardware scheduling
#asahi · 2021-10-22 03:04
chadmedits been a while since i looked at it
#asahi · 2021-10-22 03:04
chadmedi think power isa does that for some instructions too
#asahi · 2021-10-22 03:00
chadmedohh right so it is just fixed fields in the instruction word ok
#asahi · 2021-10-22 02:57
chadmedyeah thats about as far as my understanding goes lmao, i cant wrap my head around _how_ it does that effectively
#asahi · 2021-10-22 02:52
chadmedive tried multiple times to study VLIW and every time it kinda just makes my brain shut off
#asahi · 2021-10-22 02:45
chadmedand to be fair to them, its not an easy thing to explain. the problems with it are multifaceted and extremely complex and so its easier for people on the left hand side of the dunning-kruger graph to just say "oh yeah its cisc so its bad lol"
#asahi · 2021-10-22 02:45
chadmedi think people just use CISC as a pejorative to express their dislike of amd64, but very few people can actually exlpain *why* amd64 is crap
#asahi · 2021-10-22 02:44
chadmedwell yeah im actually with phire on this one, the terms are utterly meaningless in 2021 and people only really use them because we havent invented new ones to distinguish designs (because there really isnt much to distinguish them anymore)
#asahi · 2021-10-22 02:42
chadmedlike im not even sure if ibm z/architecture is a "real" CISC anymore or if they went the uop route like amd and intel
#asahi · 2021-10-22 02:41
chadmedwell in that case given the convergence of all designs on load/store and the increasing complexity of all instruction sets, do the terms really mean anything in everyday parlance anymore?
#asahi · 2021-10-22 02:40
chadmedi guess if we consider the traditional meaning of the term, the last "real" RISC before RISC-V was probably SPARCv9, for which i have the manual

plus ›