The Alpha 21264 CPU: NT's Greatest RISC (1998)

62 pointsposted 9 hours ago
by Lammy
(halfhill.com)

41 Comments

hinkley

2 hours ago

Digital, HP, and IBM had the foresight to supply NCSA with Windows NT machines with their respective not-Intel chips in them to make sure that they had a web browser on their platforms. I got the honor of using the Alpha machine, and so it got the most QA by far. I recall spending an unnecessary amount of time not only staring at the heat sink but occasionally showing it to the new guy so they could confirm my astonishment.

Having the free hardware mostly worked, but there was a long time before all of the data alignment bugs got sorted out. It got to the point where my CS classes had progressed enough that I started looking for them myself. Literally the first development tasks I got paid to do was code reviewing for word alignment bugs on 64 bit code. It was a long goddamned time ago so it's pretty fuzzy but IIRC not all 3 processors had exactly the same restrictions for all data types. So if it worked on Alpha it usually worked on the other 2 but not 100%.

I didn't have to deal with 64 bit software at work again for another 10 years, at which point people looked at me like I was trying to be edgy when I just shrugged. No, really, I was looking at 64 bit code errors 10 years ago.

rhelz

2 hours ago

Most of the barriers for Alpha adoption were problems inherent in the 32-bit to 64-bit transition. Alignment problems, yes, but another killer problem was that 64-bit pointers were twice as large.

The software I was writing at the time (EDA) took almost exactly twice as much memory. So if we needed more than 4GB, we could go to a 64-bit machine, but unless you bought even more than 8GB, you couldn't really run on bigger problems.

As you note, all those problems had to be sorted out 10 years later, when AMD finally forced Intel's hand into selling 64-bit computers.

fredoralive

4 hours ago

The bit about how the AMD Athlon (K7) uses the same bus, and how they planned to make Slot A Alphas where the only adjustment an Althon motherboard would need is a different BIOS. Imagine what might have been, especially as Alpha Windows 2000 had a built in FX!32. Cheap Alpha systems with a good x86 compatibility story, it could've been a contender.

(Yeah, I know, several stars would've had to align for it to actually work).

krylon

4 hours ago

"Cheap" and "Alpha" would have been hard to pull of simultaneously. But it would have been really cool.

rasz

3 hours ago

There was a problem with that idea, DEC engineers working for AMD made K7 too fast for Alpha to compete.

hinkley

2 hours ago

There was a second problem with that, Digital proved in court that Intel stole their architecture to put into the Pentium.

Part of the settlement was that Intel had to fab for Digital for a long time and pay for a bunch of their patents. The Digital folks made it sound like it was a victory for them but it felt like the beginning of the end, and sure enough it was.

I think Alpha's big problem was people were still migrating to 32 bit code and here was a bunch of smartasses trying to push 64 bit processors 10 years before x86-64 became the flagship for Intel-compatible processors. During the time when 'Internet time' was a thing people said regularly and unironically. 10 years was forever.

rhelz

2 hours ago

"I am old, and have lived through four ages of the west, (the 8, 16, 32, and 64-bit ages), and have seen many defeats, and many fruitless victories."

hinkley

2 hours ago

Some wise person in the 90's looked at the 16-32 bit transition, and side-eyed the 64 bit machines that were already available. He noted how differently we write software on 32 bit hardware (flat memory addressing, virtual memory, etc), and wondered what sorts of things we'd do differently on 64 bit. It took a while, but I think the answer is virtual machines, and multiplexing virtual addresses in a single process. There are some key techniques for highly concurrent garbage collection that use features meant for VMs to accelerate mark and sweep by switching the address space of objects that have been swept, and using some page faulting tricks to handle objects that are being written and swept at the same time.

We won't have to go to 128 bit processors for address space reasons. So I wonder what the justifications will be for doing so? And what features people will actually use it for when they have it? I suppose in some ways SIMD has skipped mostly over 128 to go to 256 and 512 bits though, hasn't it?

But I also know of 2 top-tier JITed languages that can use pointer packing to use 32 bit pointers for heap sizes under 32GB to save tons of memory. So 128 bit words would make pointers obnoxiously expensive.

leoc

5 hours ago

A video of an 1994 presentation put together for Hot Chips VI on the 21164: https://youtube.com/watch?v=OHupqMbLj1g

An April 1992 University Video Communications presentation on the Alpha architecture https://youtube.com/watch?v=klg1FtHADso and then from 38m 19s on the 21064 CPU https://youtube.com/watch?v=klg1FtHADso&t=2299s . From about 2m52s https://youtube.com/watch?v=klg1FtHADso&t=172s to 4m 43s Richard L. Sites gives the Alpha team’s predictions from 1992 for the next 12-25 of CPU development, which seem to have been fairly on the nail.

(Sites hasn’t been idle recently either! He’s responsible for the ultra-low-overhead KUTrace: https://news.ycombinator.com/item?id=40972099 )

jleyank

7 hours ago

I seem to remember that Alpha’s were very fast for the time but their maximum optimization level waived IEEE floating point conformance. This, of course, drove us nuts trying to validate ports of numerically intensive code. Less interesting chips with lower optimization and limited market penetration.

Now, HP’s PA-RISC chips…. Those things were fast and easier to work with. Curiously, with SoftPC they could do windows faster than a 486 could. Slaughtered all sorts of mini-supers they did.

Would have been interesting if alpha survived to compete with SGI’s MIPS.

drob518

6 hours ago

I worked on the first PA-RISC workstations (“snakes”, 1989-1991). PA-RISC started as a fairly pure RISC design and was consequently very simple and predictable (short pipeline, 1 delay slot, no complex instructions). The philosophy of the design team was to make the system fast with high clock speeds, short pipelines, and big caches, all big fundamental variables in the performance equation.

Alpha was much more sophisticated but also a lot more complex. The Alpha memory model, in particular, was quite complex with lots of cache control and barrier primitives, IIRC. But it could fly when you got the stars to align.

Edit: Alpha also came out later and PA-RISC also got more complex in later generations.

spamizbad

4 hours ago

I feel like PA-RISC actually landed with a handful of useful somewhat-complex instructions. It always struck me as the best architecture from that era, making the correct trade-off of avoided microcode but adopted stuff like completers and shift-and-add operations to minimize instruction count and maximize work done per pipeline.

mhh__

6 hours ago

what did they not implement from 754? Given the design of the processor i can imagine them being very aggressive with assumptions around exception handling / traps - iirc this is actually the original reason why Tomasulo algorithm exists because the first out-of-order processors didn't actually make any guarantees about the order that you would observe these side effects.

pm215

5 hours ago

Alpha punts handling of denormals, infinities and NaNs to software emulation, but that wasn't particularly unusual: some sparc CPUs and early implementations of Arm VFP floating point did the same.

Looking at the alpha architecture manual, the fp emulation traps are imprecise, which imposes constraints on codegen to make it work right: the "trap shadow" extends from the potentially trapping insn until a following trap barrier, and in the shadow you mustn't e.g. use a register more than once as a destination, have a branch, or modify registers that are inputs to any insns in the shadow. (The idea is that the hardware will have already executed some of the insns in the shadow by the time it realises it needs to trap, and the handler has to be able to emulate the trapping insn and resume execution at the insn just after that, so it will re-execute all the insns in the shadow.) That's obviously pretty inconvenient for codegen, so I wouldn't be surprised if the compiler provided some kind of fast-math mode where it didn't trap and you just had to avoid generating denormals, infinities, etc.

I think making the fp using code have to be written carefully to work with the software emulation of edge cases is unusual -- I don't think either sparc or arm imposed that requirement, and instead trap precisely, or at least before anything happens where it would matter that the fp insn is emulated late.

p_l

2 hours ago

Thinking longer about Alpha's imprecise faults made me realize that signaling NaNs can be quite useful in such case

cbm-vic-20

5 hours ago

VAX compatibility was very important to DEC- the Alpha has hardware to handle the non-IEEE VAX floating point formats.

bluedino

5 hours ago

600Mhz Alphas existed when there was 195mhz MIPS

hapless

4 hours ago

turns out there is more involved in performance than just clock rate

who knew? (everyone)

bluedino

4 hours ago

And the Alpha was faster

The last line of the parent post makes it sound as if the chips didn't exist at the same time.

MomsAVoxell

4 hours ago

In those days, I was using SGI machines to build web sites for folks .. the Indy was very popular for this purpose. One day I was given a DEC Alpha machine to evaluate and see if it was a worthwhile addition to our inventory.

It came with NT, so there was some friction to just adding it to our services. These days were very frustrating - Microsoft was hell-bent on killing Unix, and later Linux also - and there was a lot of back and forth in our engineering team whether we wanted to invest in this hassle.

We didn't. I had that machine under my desk doing basically nothing for a year, before I sent it back.

If there had been a bit more insight into the nature of things, and if it had been running a Unix variant, we would have given it a better chance.

So then it was even more frustrating when SGI did a deal with the devil in later years, and tried to get its customers switch to NT, also. That killed SGI, in my opinion.

Looking back now, it's kind of incredible the resistance to Unix in those days, and how it was all going to be replaced with some "New Technology". Linux won. SGI didn't. And DEC was an early victim they should have learned from, in my opinion.

pavlov

3 hours ago

From Microsoft’s POV they soundly beat Unix in the 1990s because they were primarily focused on the GUI workstation market.

In 1991 the market for high-end desktop software like engineering, video editing and 3D modeling tools was dominated by Unix and classic Mac. In 1999 all of those applications were on Windows NT.

Vendors like Autodesk and Avid were building for Windows first. New graphics acceleration hardware targeted PC add-on cards rather than being exclusive to a workstation vendor (SGI tried this approach with their NT box and it flopped).

In retrospect it was just commercial Unix that had lost the game, but it wasn’t obvious at the time that Linux could reclaim this market. And Mac OS X was considered by many a doomed project (after all Apple had been promising a new OS for the entire ‘90s, nobody knew how deeply the NeXT acquisition would transform the company).

adrian_b

2 hours ago

Engineering never went completely to Windows NT.

Between 1997 and 2007, I worked at 3 different companies, in 3 different countries (2 in Europe + Israel) as Design Engineer in electronics.

All the serious engineering programs for EDA/CAD were run on Solaris and accessed from Windows with X-terminal programs.

Towards the end of that decade, the Opteron-based servers were both much faster and much cheaper than the Sun servers or the Fujitsu servers, so the EDA/CAD programs were migrated from Solaris to Linux, while the Windows computers continued to run only the X-terminal programs for accessing the servers.

At the beginning of that decade, I also used a Sun workstation, but those disappeared after 1999, because they were much too slow in comparison with a PC with Intel Pentium III or with AMD Athlon.

It is likely that the reason why those EDA/CAD programs did not have Windows versions at that time was that they already required a lot of memory, typically much more than 4 GB, so they migrated from Solaris to Linux only after the availability of x86-64 servers, while having a Windows version was not possible before mid 2006, when Intel joined AMD in providing 64-bit CPUs even for PCs, not only for servers, so the market share of 64-bit PCs became non-negligible.

pixelesque

2 hours ago

Yeah, even in like 1996/1997 for certain industries there were hints as to the way things were going, even if it took 4/5 years for the transition to fully take place.

For example, in 1996/1997, Digital Domain (VFX industry) used a 'render farm' cluster of Carrera Alpha workstations running NT to render the Titanic film, instead of SGIs running IRIX. (SGIs were still often used on the artists workstations though, but progressively that shifted).

By 2001, many of those machines were x86 which were then often as fast as the SGIs and Alphas, even with x86's stack-based floating point architecture which handicapped it a bit, and the significantly higher memory bandwidth and larger caches of the SGIs.

adrian_b

2 hours ago

In 1999, AMD Athlon became the first x86 CPU that was able to do both an addition and a multiplication during one clock cycle, for the 80-bit x87 floating-point numbers.

The previous Intel CPUs of the P6 family, from Pentium Pro to Pentium III, required 2 clock cycles for that, i.e. they reached at most half the throughput of Athlon at the same clock frequency. And Athlon had an even higher clock frequency.

So the launch of Athlon was one of the greatest jumps in floating-point performance per socket in the history of x86 CPUs.

It had a higher clock frequency than any Alpha. IBM POWER CPUs could do much more per clock cycle than Athlon, but their clock frequency was much lower, so Athlon was still faster.

One year and a half later Intel launched Pentium 4, which could match the throughput per clock cycle of Athlon, but only when executing new SSE2 programs, not when executing any legacy program.

rbanffy

8 hours ago

It’s so good to re-read Tom Halfhill’s articles. I wonder where he’s been in the past… checks notes… 30 years I haven’t seen his writings.

GeekyBear

2 hours ago

He was my favorite author at Byte.

stmw

5 hours ago

Fun fact - Linus first trip to the United States was related to Linux-on-Alpha port and sponsored by Digital Equipment Corporation.

dosinga

4 hours ago

> outside of the highly specialized embedded-application market, RISC is in retreat

with hindsight it is funny to realize that RISCs come back would start from inside that highly specialized market in the form of ARM.

trimbo

2 hours ago

And just like the 90s, every major tech company is making their own RISC chip[1]! Apple Silicon, Tensor, Graviton, Axion, Cobalt, Grace....

[1] - Though not exactly since they're all ARM ISA at the foundation.

EvanAnderson

7 hours ago

> David Jessel, the Alpha's senior product manager, claims that Alpha fans have nothing to worry about. "Compaq has been fully supportive of the Alpha and is continuing to invest in it," Jessel said.

That worked out swimmingly for Alpha... >sigh<

I never programmed Alpha assembly, so I don't really have a feel for the architecture. I did deploy some Alpha-based boxes running NT, and they were very nice. They didn't feel pieced-together like x86 servers did.

cbm-vic-20

5 hours ago

On the hardware side, DEC built very-well engineered hardware. The early Alpha machines were very well designed. Even the PC compatible lines of the era were built to a very high quality standard.

phire

5 hours ago

My understanding is that after Compaq sold the Alpha IP to intel, and quite a lot of the DNA ended up in Nehalem and Sandy Bridge.

whaleofatw2022

5 hours ago

Not sure about that bit...

What is well known however, is that a bunch of Alpha engineers wound up at AMD, and a lot of that DNA went into the OG Athlon

scrlk

3 hours ago

IIRC, the Sunway CPUs used in Chinese supercomputers have an Alpha-like design.

theandrewbailey

8 hours ago

From back when IA-64 (A.K.A. Itanium) was supposed to take Intel to the promised land.

pjmlp

5 hours ago

Without AMD, maybe it would have.

shdwslrkr

5 hours ago

No, it wouldn't have.

Without AMD rescuing x86, PowerPC wouldn't have died. MIPS wouldn't have died, and faces with the need to brake the 4Gn barrier on consumer equipment microsoft would have had windows running on three or four competing architectures until 2013 when everything would have switched to ARM.

Itanium would have already been a rotting corpse.

AMD rescued Intel from its own management.

pjmlp

2 hours ago

Sure it would, Windows XP 64 bits was alreary on Itanium, and NT versions for PowerPC and MIPS were already dead by then.

danielktdoranie

2 hours ago

Hardly anyone ran NT on these beautiful chips. Tru64 was the OS of choice for these in 1998. Hence the decision by MS to cease support. Running NT on a Dec Alpha would be like running kerosene in your Ferrari’s engine. If you wanted to run Windows NT you could do so a lot cheaper on an x86 PC.

hinkley

2 hours ago

BYTE magazine has been gone for almost 30 years now. Fuck.

Dr Dobbs has been gone for 12, which is still a long time, but when I first read the intro my brain transposed the two and I had a proper freak out before I realized what I'd done.

robin_reala

6 hours ago

I remember Byte folding, but I’d blanked that it was a full 28 years ago.