That era of Macs with the pull-down hinged side was peak 'Pro' Mac. Back when they were actually more pleasant to service than any other computer on the market (and everything was modular).
Even most of today's nicer PC case designs are not as much of a joy to work inside—often the drive bays are stuck in random spots and require many screws to get at them.
Wish Apple would've carried that on again with the modern Mac Pro, but there are fewer replaceable parts now.
It really is a tragedy that Apple abandoned the Cheesegrater v1 for the Trashcan.
The modern iteration of the Pro Macs (really, Mac Studio for most people) probably the best that can reasonably be expected of Apple. But the continued feud with NVidia is really unfortunate.
Although I suppose the Mac Pro doesn't really support graphics cards in its PCI-X slots anyhow. Or at least that's what I recall.
I know what you mean, but I had a the cylindrical aluminum trashcan Mac Pro and that was a beast. I sort of which I hadn't traded it in a few years ago when upgrading. the machining of the case somehow was fascinatingly cool, i thought.
Not to disagree about the cheesegrater (I've been a DIY PC guy since graduating from my C=128) but I always loved working on those things and spent a bit of time trying to find any standard *ATX case with that sort of build quality - even at a decent price premium!)
But my last two PC cases over the last ~15 years were nothing terribly fancy and they have drive bays that just pull out on trays in the back. No need to open the case, just undo a chunky thumb screw and pull out the tray.
Just wish the rest of the cases were as solid and attractive as the old Mac Pros.
I have a G5 and indeed the interior is awesome. But I also have a modern ThinkStation and it's actually nice. Not as flashy as the cheese grater, but the build quality seems to be there.
I concur. I have a 350MHz Power Mac G4 that I purchased for only $40 in 2009 from a couple who couldn’t figure out why it stopped working. It turned out the hard disk and the CD-ROM drive didn’t work, and its internal battery ran out; replacing them was easy. It’s one of the easiest computers I’ve ever worked on. I have both Mac OS 9 and Mac OS X Tiger installed on separate hard disks. It even handled the Web decently until around 2013.
I also have a “trash can” Mac Pro that was my daily driver from 2017 (when I purchased a refurbished model) to 2022 when I switched to a Ryzen 9 build. My Mac Pro was a beast, and while it lacked expansion slots, it was still user-serviceable.
I miss the days of user-serviceable Macs, which is why I switched back to PCs after using Macs as daily drivers since I bought an Intel Core Duo MacBook in 2006. I now use a Ryzen 9 build at home and a Framework 13 laptop when I’m away from home.
I am grateful for Jean-Louis Gassée, who helped turn the Macintosh from a closed appliance to an expandable, user-serviceable machine, beginning with the Macintosh II in 1987. The IIfx, the Quadra, the Power Macintosh line, and the pre-ARM Mac Pro line were all part of this legacy.
This thing still looks awesome! I really liked the timelessness of mac design and I kind of miss the awe of earlier macs. Nowadays it's incremental. Back in the day you were wowed by the aurora background.
I was the first to order a Yikes! 400MHz G4 tower from my local Apple reseller when they first came out. I loved this ad campaign! I was also smitten when Apple dropped the speeds on all the SKUs by 50MHz as I had gotten mine before that happened.
The Yikes! G4 was basically the blue and white G3 hardware (chassis, mainboard, everything) just with tiny circuit changes to give 66MHz PCI for the graphics card while all the other G4 towers had a completely different motherboard with AGP graphics. I really liked that machine, but I ended up selling it around 2008 as I no longer used it much.
Motorola struggled to meet demand and Apple had to give up on the 500 MHz model it had promised customers. Apple had to shift its speeds down, and ended up selling 350 MHz and 400 MHz “PCI” models and a 450 MHz “AGP” model.
I actually own a 350MHz “Yikes!” model, though I purchased mine in 2009 for just $40. It’s a very nice machine and is one of the easiest desktops I’ve worked on. Mine runs both Mac OS 9 and Mac OS X Tiger; while I own faster PowerPC Macs for running Tiger, this is my fastest machine running the classic Mac OS; it feels nice booting into Mac OS 9 every now and then and using classic applications.
While that law seems hilarious to us now, and the new restrictions on AI seem equally dumb, we have to admit that the law worked. It did what was intended -- prevented the sale of supercomputers to enemy nations, which in fact held back their weapons and aerospace R&D.
It will be a lot harder this time around to prevent "super AI" from spreading around the world, but it might slow it down just a bit, which again is the goal of the US government.
It's also worth looking at the conditions the law worked under. Up through til the 90s your semiconductors where coming almost exclusively from the West (+Japan, in this case counting the same though) - also the only place you could get all the different high performance products you needed designed at. If you want to avoid both sides of that coin you've got decades of catching up to do.
Of these gaps one is already eroded and the other is significantly smaller in that China already has domestic 32 GB 10+ TFLOP GPUs and supercomputers made of domestic CPUs. Will export restrictions slow things more than doing nothing? Sure, but it may not be particularly meaningful just because it worked well when we had the same idea in the 70s. It may, for example, instead be the kind of thing to put enough market pressure to move the tasks to happen somewhere else.
Good news, everyone! The Mac can still be used as a munition!
I often vacation at an old medieval castle and the Mac just barely fits through the murder hole. I dropped one on some carolers just this past season! It was quite a family bonding moment. (no one was hurt, we yelled GERONIMO! before tossing it)
The CRT on AIO models makes a satisfying sound when it hits the ground.
I had a Centris 610 with an ITAR restricted motherboard. Even the people at the Apple store were jealous. The Centris 650 could run LabVIEW, but the stock 610 could not as it emulated certain instructions which LabVIEW relied on. Apple sold me the 610 with the claim that it would run any software the 650 would (within reason of course). I couldn't program on the 650 because it was in a hot room and I wasn't allowed to spend a lot of time in there. After much arguing they agreed to "fix" the 610. It took six weeks. To get it back I had to sign a document attesting I was a US citizen. If you ask me LabVIEW was broken, but whatever.
It's not just a Mac thing. My vague recollection is that once upon a time the Playstation ran on defective chips which couldn't be manufactured in certain countries. The Barcelona Supercomputing Center made what amounted to a jailbreak so you could use the Playstation as a development platform.
The Centris 610 had the 68LC040, which lacked a floating-point unit. Mac apps were not supposed to depend on the FPU being present, but I guess LabVIEW did.
I assume they just replaced your 68LC040 with a full 68040.
Is it okay to use the unit like "1 GIGAFLOP of computing power" and "thousands of GIGAFLOPs in our pockets"? I thought the base unit was "flops", where the "s" always stood for "second", as in "1 flops" (1 floating point operation per second).
I was at Cisco in the early 1990s working for customer service for the US Federal sales team. The Cold War Era Defense Procurement Allocation System (DPAS) — which is still on the books [https://www.bis.doc.gov/index.php/other-areas/strategic-indu...] — classified routers and switches as "Missile Parts."
We had to expedite any government deliveries purchased under DX-rated and DO-rated orders. And there were a lot of them.
We also faced Department of Commerce and Department of State restrictions on shipping any software that included 56-bit (DoC limited) or 64-bit (DoS limited) encryption. It made it hard to ensure data-in-flight was secure around the world. [A lot of international banks were not happy the USG was blocking 64-bit encryption to them and only allowed on a case-by-case basis.]
I mentioned this in another comment. The Barcelona Supercomputing Center made a jailbreak for the playstation (https://www.bsc.es/projects/deepcomputing/linuxoncell/ 404 now). But no worries, the chips in the Playstation were the rejects. :-/ (PS3 according to my contemporaneous notes.)
Does anyone know how good the PowerPC chips from that era actually were compared to the competing chips from Intel?
I remember reading a lot of marketing claims, but back in those days there wasn’t so many review sites that would test them against each other with a multitude of precise benchmarking suites. Nor were there so many pieces of software which ran on both platforms.
TL;DR: Clock-for-clock the 7400 was arguably faster than the Katmai, the Coppermine was arguably faster than the 7400, and the 7450 was arguably faster than Coppermine.
TFA is specifically discussing FLOPS. The PPC 7400 (first G4) had an AltiVec unit that was considerably more powerful than the Katmai (first P3) SSE engine.
The 7400 altivec was 128-bits wide and had separate permute and arithmetic engines; this means it could retire up to 2 4x32 SIMD floating point operations per cycle (if you scheduled permutes .
The SSE engine on the Katmai was only 64-bits wide, so it took 2 uops to do a 4x32 add and 2uops to do a 4x32 multiply. However, it could do an add and a multiply simultaneously, so if you were alternating adds and multiplies, you could retire one 4x32SIMD instruction per cycle. Coppermine (2nd generation P3, released fall of 1999) used a 128-bit unit, but I'm not sure if it could still retire adds and multiplies in parallel (I think not).
For general (i.e. not just FPU) performance: Coppermine and 7450 introduced on-die L2 cache to the P3 and G4 series respectively. This probably had a bigger performance impact on a broader base of code than other changes made in and around the time, as the FSBs were not keeping up with the CPU clocks.
Something I forgot to mention is that Intel obtained and maintained a process advantage over Motorola around this time; The early G4 PowerPCs were on Motorola's 200nm process, but had to be downclocked to get the yields up. Intel used their existing 240nm process for the Katmai P3, which allowed them to release a 600MHz stepping of the Katmai the same month that Motorola failed to launch a 500MHz stepping of the 7400.
Of course Intel famously then pivoted to the NetBurst architecture, for the P4, which turned out to be an evolutionary dead-end thanks to increased leakage at 90nm. The idea was to tradeoff reduced IPC in exchange for higher clocks, but they were never able to get the ALUs running over 7.6GHz, and they had to reduce IPC so much to get there that any advantages of the high clock speed were lost. Nevertheless the fastest P4s on the 130nm process were significantly faster than the fastest P3s on the 130nm process.
"The Wassenaar Arrangement on Export Controls for Conventional Arms and Dual-Use Goods and Technologies" is still an interesting document for learning where the "fun" areas of technology currently are...
> Robert Beach, a spokesman for the Compaq Computer Corporation in Houston, said: ''We are pleased to see that the regulations have been relaxed. It's good any time these things are done.'' But like Apple, the Compaq official said that ''we don't see any great impact on Compaq in the near term.'' Compaq's product line has migrated to the Intel 80386 processor, which is still restricted. ''We would welcome further relaxation to include 386 machines,'' he added.
We in Ukraine seen these restrictions, as that time we considered "east block".
Rumors said, this was real problem, so local Apple distributors written in documents simplest model, which does not exceed 1GFlops, but in reality delivered other.
I'm a bit surprised that the author went to the trouble of faking the CNN "screenshot", but allows it to look off. It purports to be viewing a page from an old version of MSIE on Windows, but it doesn't pass the sniff test.
* The images aren't loaded. This is naturally expected of an article loaded 20+ years after publication. But if the screenshot was grabbed in the past, you would expect the images to also work.
48 comments
[ 3.3 ms ] story [ 112 ms ] threadEven most of today's nicer PC case designs are not as much of a joy to work inside—often the drive bays are stuck in random spots and require many screws to get at them.
Wish Apple would've carried that on again with the modern Mac Pro, but there are fewer replaceable parts now.
The modern iteration of the Pro Macs (really, Mac Studio for most people) probably the best that can reasonably be expected of Apple. But the continued feud with NVidia is really unfortunate.
Although I suppose the Mac Pro doesn't really support graphics cards in its PCI-X slots anyhow. Or at least that's what I recall.
But my last two PC cases over the last ~15 years were nothing terribly fancy and they have drive bays that just pull out on trays in the back. No need to open the case, just undo a chunky thumb screw and pull out the tray.
Just wish the rest of the cases were as solid and attractive as the old Mac Pros.
I also have a “trash can” Mac Pro that was my daily driver from 2017 (when I purchased a refurbished model) to 2022 when I switched to a Ryzen 9 build. My Mac Pro was a beast, and while it lacked expansion slots, it was still user-serviceable.
I miss the days of user-serviceable Macs, which is why I switched back to PCs after using Macs as daily drivers since I bought an Intel Core Duo MacBook in 2006. I now use a Ryzen 9 build at home and a Framework 13 laptop when I’m away from home.
I am grateful for Jean-Louis Gassée, who helped turn the Macintosh from a closed appliance to an expandable, user-serviceable machine, beginning with the Macintosh II in 1987. The IIfx, the Quadra, the Power Macintosh line, and the pre-ARM Mac Pro line were all part of this legacy.
The Yikes! G4 was basically the blue and white G3 hardware (chassis, mainboard, everything) just with tiny circuit changes to give 66MHz PCI for the graphics card while all the other G4 towers had a completely different motherboard with AGP graphics. I really liked that machine, but I ended up selling it around 2008 as I no longer used it much.
Motorola struggled to meet demand and Apple had to give up on the 500 MHz model it had promised customers. Apple had to shift its speeds down, and ended up selling 350 MHz and 400 MHz “PCI” models and a 450 MHz “AGP” model.
https://en.wikipedia.org/wiki/File:Munitions_T-shirt_(front)...
Who's laughing at the original 1979 restrictions in their original context?
It will be a lot harder this time around to prevent "super AI" from spreading around the world, but it might slow it down just a bit, which again is the goal of the US government.
Of these gaps one is already eroded and the other is significantly smaller in that China already has domestic 32 GB 10+ TFLOP GPUs and supercomputers made of domestic CPUs. Will export restrictions slow things more than doing nothing? Sure, but it may not be particularly meaningful just because it worked well when we had the same idea in the 70s. It may, for example, instead be the kind of thing to put enough market pressure to move the tasks to happen somewhere else.
I often vacation at an old medieval castle and the Mac just barely fits through the murder hole. I dropped one on some carolers just this past season! It was quite a family bonding moment. (no one was hurt, we yelled GERONIMO! before tossing it)
The CRT on AIO models makes a satisfying sound when it hits the ground.
It's not just a Mac thing. My vague recollection is that once upon a time the Playstation ran on defective chips which couldn't be manufactured in certain countries. The Barcelona Supercomputing Center made what amounted to a jailbreak so you could use the Playstation as a development platform.
I assume they just replaced your 68LC040 with a full 68040.
We had to expedite any government deliveries purchased under DX-rated and DO-rated orders. And there were a lot of them.
We also faced Department of Commerce and Department of State restrictions on shipping any software that included 56-bit (DoC limited) or 64-bit (DoS limited) encryption. It made it hard to ensure data-in-flight was secure around the world. [A lot of international banks were not happy the USG was blocking 64-bit encryption to them and only allowed on a case-by-case basis.]
https://en.wikipedia.org/wiki/Export_of_cryptography_from_th...
I remember reading a lot of marketing claims, but back in those days there wasn’t so many review sites that would test them against each other with a multitude of precise benchmarking suites. Nor were there so many pieces of software which ran on both platforms.
TFA is specifically discussing FLOPS. The PPC 7400 (first G4) had an AltiVec unit that was considerably more powerful than the Katmai (first P3) SSE engine.
The 7400 altivec was 128-bits wide and had separate permute and arithmetic engines; this means it could retire up to 2 4x32 SIMD floating point operations per cycle (if you scheduled permutes .
The SSE engine on the Katmai was only 64-bits wide, so it took 2 uops to do a 4x32 add and 2uops to do a 4x32 multiply. However, it could do an add and a multiply simultaneously, so if you were alternating adds and multiplies, you could retire one 4x32SIMD instruction per cycle. Coppermine (2nd generation P3, released fall of 1999) used a 128-bit unit, but I'm not sure if it could still retire adds and multiplies in parallel (I think not).
For general (i.e. not just FPU) performance: Coppermine and 7450 introduced on-die L2 cache to the P3 and G4 series respectively. This probably had a bigger performance impact on a broader base of code than other changes made in and around the time, as the FSBs were not keeping up with the CPU clocks.
Of course Intel famously then pivoted to the NetBurst architecture, for the P4, which turned out to be an evolutionary dead-end thanks to increased leakage at 90nm. The idea was to tradeoff reduced IPC in exchange for higher clocks, but they were never able to get the ALUs running over 7.6GHz, and they had to reduce IPC so much to get there that any advantages of the high clock speed were lost. Nevertheless the fastest P4s on the 130nm process were significantly faster than the fastest P3s on the 130nm process.
https://arstechnica.com/features/2004/08/ppc-1/
Comparison between the Pentium 4 and the Power PC G4e:
https://archive.arstechnica.com/cpu/01q4/p4andg4e2/m-p4andg4...
Stokes had so much material on these deep dives, he collected it into a book, "Inside the Machine":
https://nostarch.com/insidemachine.htm
Chapter 7 is the PowerPC vs Pentium 4 analysis. Curiously, I was going over it again just last night.
https://www.wassenaar.org
From https://www.nytimes.com/1989/07/19/business/in-shift-us-ease...:
> Robert Beach, a spokesman for the Compaq Computer Corporation in Houston, said: ''We are pleased to see that the regulations have been relaxed. It's good any time these things are done.'' But like Apple, the Compaq official said that ''we don't see any great impact on Compaq in the near term.'' Compaq's product line has migrated to the Intel 80386 processor, which is still restricted. ''We would welcome further relaxation to include 386 machines,'' he added.
Rumors said, this was real problem, so local Apple distributors written in documents simplest model, which does not exceed 1GFlops, but in reality delivered other.
* The URL bar displays http://www.cnn.com, but the true URL is http://www.cnn.com/TECH/computing/9909/17/g4.ban.idg/
* The images aren't loaded. This is naturally expected of an article loaded 20+ years after publication. But if the screenshot was grabbed in the past, you would expect the images to also work.
* The font is "Times Roman" used on Apple devices, not the "Times New Roman" on Microsoft ones. Subtle differences in letters like "5" and "S" exist: https://commons.wikimedia.org/wiki/File:Times_Roman_vs_Times...