I remember thinking that Alpha might eventually topple x86. At the time, there was no x86-64, and Linux was still buggy, so the RISC platforms were thriving...and Alpha was at the top. Then Microsoft made NT run on Alpha. All in the early to mid 90's. Then the 1-2 combination of the Itanium distraction followed by the Opteron + 64 bit Linux, and then Alpha died.
Yeah - I had high hopes for the alpha as well. I think it had more to do with Dec not getting price competitive with x86 before x86_64 took off. I don't really think the Itanium factored into it - it always seemed more of an 'also ran' then a real threat.
I think I might still have some Dec Multia's with Alpha 166MHgz out in the garage somewher. Neat lil boxes.
When Compaq bought out DEC they shuttered all Alpha development in favor of Itanium. Continuing to develop a competing ISA would put them in a bad place as an Intel customer.
The way it was done was so abrupt that Windows 2000 nearly released with Alpha version on standard CD (which would be AFAIK 64bit clean, unlike the older ones), but essentially the whole NT/alpha team (which was part of Compaq) was told close to RTM release "sorry, your project was closed".
Later, Compaq had to restart production, and HP even had to introduce updated chip, because some customers, especially VMS, didn't want to move to slower Itanium machines.
I have even heard that part of Itanium's flop was how often older EV6 machines were running circles around new Itanium ones...
"I don't really think the Itanium factored into it"
There's a sibling comment that mentions Compaq putting in the final nail. But, my impression was that more customers might have switched to Alpha servers (from HP-UX, Solaris, etc) if they didn't think Itanium was going to win the market. It also gave more credibility to x86-64 by sort of confirming that old-school RISC was dying. Itanium did end up with a heavy stigma, but there was a time when many thought it would be the future.
I got my first computer around the time that happened, a humble Pentium 1.
It was only much, much later that I learned about Linux, then FreeBSD and the whole history behind UNIX, RISC machines and all that - at a point when it was mostly history.
I even acquired some machines out of pure fascination.
I got an IBM RS/6000, a Sun Ultra45 and a Blade 100.
Still gotta get a MIPS and, especially, an Alpha machine.
These machines seemed so much more serious compared to PCs, I was blown away when I discovered there had been 64 bit architectures in the mid-90s!
To this day, PCs seem like toys to me with their gimmick-y firmware.
It always left me wondering about what could have been.
And Alpha may have been the most promising back in the day!
What's more, there were 64 bit laptops in the mid 90s. Not particularly mobile ones, but if you could tolerate a 7 pound behemoth, you could get a 200 MHz 64 bit SPARC processor with 512MB of RAM in '96/97. The AlphaBook with an Alpha processor that was announced in '95 comes even earlier. https://www.vaxbarn.com/index.php/other-bits/555-tadpole-alp...
One thing that always strikes me about workstations and (especially?) these laptops is that, while the CPUs themselves weren't always much faster than competing Intel ones, they always seemed to have much better I/O options. The SCSI drives and controllers seemed to get much better throughput with lower CPU utilization, which pays dividends to how the overall system performs.
A well designed workstation is more than just the sum of its parts. While these workstations/servers were arguably overpriced and it's not surprising they failed to the cheaper Wintel machines, they just seemed more well rounded...
> The SCSI drives and controllers seemed to get much better throughput with lower CPU utilization, which pays dividends to how the overall system performs.
Indeed - my employer at the time acquired 10 AlphaPC 164SX motherboards with 21164PC CPU's and after adding a nice amount of RAM and a nice fast Ultra2 SCSI drive they felt really fast. I remember running the early GNOME 1.0 desktop on it and it was just so responsive.
There is a deep irony to it. DEC basically created the minicomputer market by eating away at the low end of the mainframe industry. They would later be myopic when the same was done to them. They almost completely missed the PC revolution. They never quite grasped that the future was in commodity silicon and the desktop form factor.
The VAX was a processor-on-a-chip by 1985, several months before the Intel 386 came out. A complete 32-bit machine that could fit in a desktop form factor. It never really happened. In the late 80s they would weakly pursue the high-end workstation market. But a cheap, mass-produced VAX was anathema.
Alpha would follow the same pattern, and by then PCs were already the established architecture. It would have been too late to veer into the consumer market, probably.
If Apple were white-labeling M1's for anyone to use, I'd expect full-blown panic at Intel. Intel can at least breath a little knowing that Apple and AWS are going to keep their ARM CPUs in-house.
Intel also blatantly stole patented technology from DEC. The Pentium onward, IIRC, ripped off entire designs. When Compaq eventually acquired DEC, they cancelled all of the lawsuits and gave Intel complete access to DEC’s patents. They also gave Intel DEC’s raced out ARM chip that they built for Apple.
Also note that it doesn’t run on bare metal: “To date all users run x86 OpenVMS as a Virtual Box or KVM guest. VSI provides an appliance file (.OVA) which the customer downloads”
It's a fork of es40. New build system, more stable networking, bug fixes, but don't expect miracles just yet. 10 year old c++ codebase is not something you magically speed up overnight
I used to have a, I believe, three processor Alpha computer when I worked at Microsoft. I had a MIPs, PPC and X86 based computers as well and all were top end and the Alpha always seemed to outpace them all. I was sad when it was discontinued.
From what I know (I used to be software engineer in the hardware design team) Alpha designers used to use domino logic [1]. It allowed for some quite beefy circuits expressed in smaller space, because domino logic dominantly used nFET transistors which are usually smaller than pFET.
Typical k-input Logical AND-NOT uses k pFET and k nFET transistors, the domino logic uses one pFET and k nFET for similar purposes.
The paper about domino logic synthesis [2] hints at what Apple uses for their CPU cores.
23 comments
[ 2.1 ms ] story [ 56.2 ms ] threadI think I might still have some Dec Multia's with Alpha 166MHgz out in the garage somewher. Neat lil boxes.
All this happened before x86_64 was a thing.
Later, Compaq had to restart production, and HP even had to introduce updated chip, because some customers, especially VMS, didn't want to move to slower Itanium machines.
I have even heard that part of Itanium's flop was how often older EV6 machines were running circles around new Itanium ones...
There's a sibling comment that mentions Compaq putting in the final nail. But, my impression was that more customers might have switched to Alpha servers (from HP-UX, Solaris, etc) if they didn't think Itanium was going to win the market. It also gave more credibility to x86-64 by sort of confirming that old-school RISC was dying. Itanium did end up with a heavy stigma, but there was a time when many thought it would be the future.
http://alasir.com/articles/alpha_history/compaq_epoch.html
It was only much, much later that I learned about Linux, then FreeBSD and the whole history behind UNIX, RISC machines and all that - at a point when it was mostly history.
I even acquired some machines out of pure fascination. I got an IBM RS/6000, a Sun Ultra45 and a Blade 100.
Still gotta get a MIPS and, especially, an Alpha machine.
These machines seemed so much more serious compared to PCs, I was blown away when I discovered there had been 64 bit architectures in the mid-90s! To this day, PCs seem like toys to me with their gimmick-y firmware.
It always left me wondering about what could have been. And Alpha may have been the most promising back in the day!
One thing that always strikes me about workstations and (especially?) these laptops is that, while the CPUs themselves weren't always much faster than competing Intel ones, they always seemed to have much better I/O options. The SCSI drives and controllers seemed to get much better throughput with lower CPU utilization, which pays dividends to how the overall system performs.
A well designed workstation is more than just the sum of its parts. While these workstations/servers were arguably overpriced and it's not surprising they failed to the cheaper Wintel machines, they just seemed more well rounded...
Indeed - my employer at the time acquired 10 AlphaPC 164SX motherboards with 21164PC CPU's and after adding a nice amount of RAM and a nice fast Ultra2 SCSI drive they felt really fast. I remember running the early GNOME 1.0 desktop on it and it was just so responsive.
The VAX was a processor-on-a-chip by 1985, several months before the Intel 386 came out. A complete 32-bit machine that could fit in a desktop form factor. It never really happened. In the late 80s they would weakly pursue the high-end workstation market. But a cheap, mass-produced VAX was anathema.
Alpha would follow the same pattern, and by then PCs were already the established architecture. It would have been too late to veer into the consumer market, probably.
https://github.com/simh/simh
For the VMS-nostalgic, VSI now have OpenVMS running on x64 Intel (and in fact apparently already released it as 9.0!):
https://vmssoftware.com/updates/state-of-the-port/
Code Contributions, bug reports and testing are very welcome
Typical k-input Logical AND-NOT uses k pFET and k nFET transistors, the domino logic uses one pFET and k nFET for similar purposes.
The paper about domino logic synthesis [2] hints at what Apple uses for their CPU cores.
[1] https://en.wikipedia.org/wiki/Domino_logic
[2] https://www.researchgate.net/publication/260714833_1-of-N_Do...