Perhaps not super-relevant to lobste.rs as it’s a business story, but I know there’s some love here for the SGI of old.
The business technology giant said Thursday that it will buy computing hardware maker SGI for roughly $275 million. HPE said the deal should close in the first quarter of its fiscal 2017 as long as regulators approve it.
SGI made some of the greatest workstations and NUMA machines of the past. People drooled over the specs. Even a recent one was 2000+ cores, a few TB of RAM, and multi-GB/s interconnect. Bad management led them downhill until Rackable bought them. Now HP buys them.
A little concerned for its future given HP tried to kill two, good, RISC CPU’s and OpenVMS.
Yep, they did - not to mention their machines had real style in the days when Apple was making horrible beige PPC Macs.
They’ve made so many bad decisions in the past (of course, they’re not alone in that regard). @bcantrill mentioned on the BSD Now podcast that Sun bought the Cray Business Systems Division from SGI for less than $10 million and in the first year at Sun it turned over more than $1 billion (this article suggests the price paid was closer to $50 million, but that’s immaterial really). Hell, the E10K was the go-to big system in the late 90s/early 2000s - I remember seeing a few maxed-out models in the DC of a national telco ca. 2001 and thinking “each of those cost over $1 million”.
I was shouting suggestions at them the situation was so bad. They were slow to capitalize on clustering. They could’ve created different tiers of pricing on NUMAlink to market grab on business servers for databases with COTS components and/or workstations. All the UNIX workstations were operating under assumption people would drop $20+k on a box with Moores law & gaming market pushing x86 faster at $1-3k a box. Retarded. Much of their R&D revenue came from government purchases of supercomputers anyway so they’d get it back in volume. Could’ve loaded their own desktops with the emerging GeForces with extra memory. They were ahead a bit on low-latency, FPGA acceleration with RASC plugins for Altix systems. Should’ve done it for desktops or some low-cost server, though, as competitors like Mercury were aiming for PCI cards and shit.
One poster in another forum figured HP is buying it for NUMAlink mainly. I agree. It’s the tech I loved 10 years ago. It’s still totally awesome today albeit competition doing nicely, too. NUMAlink plus HP’s high-end servers would be beastly combination. Plus whatever expertise they have in scaling Linux, compilers, etc on ridiculous numbers of CPU’s. Large, SSI HW + NUMAlink + software to use them well are still defining advantages of SGI boxes. Plus it had a simple, catchy name. People overlook the value of stuff like that.
Don’t shed too many tears…this is HPE buying Rackables, not really Silicon Graphics.
They own the IP and boxes that resulted from the real SGI. I’ve seen a lot of stuff catch up to SGI on desktop and RAS side. I haven’t seen anything beat their NUMA machines overall. Have you seen the SGI UV?
https://www.sgi.com/products/servers/uv/uv_3000_30.html
Their case don’t look as awesome as the old Origin cases but they their specs do. :)
I worked on SGI machines at the beginning of my career, and they were cool, and fast, and XFS was rad, and boy did they ever reboot a lot, but NT4 and commodity 3D hardware rendered them irrelevant long before they died. There’s nothing of the Crimson/Power Challenge/Onyx lineage left. Reams could be written about the undoubtably bad decisions SGI took, but at the end of the day, they were just another victim of the inexorable downward pressure on profits that Intel and Microsoft were able to exert.
I have mixed feelings about this one. In one way, it’s good because it means SGI has recovered somewhat, but on the other hand it reminds me how much they’ve fallen.
When I was at school an SGI Indy was one of those machines I was fascinated by - they even used to advertise them in BYTE. Later, I visited the SGI bus that used to tour around showing off their latest hardware - RE2-equipped Origins amongst them - and was even more impressed. sigh.
Here’s you a highlight from the past. They used to do stuff in real-time on 100-200MHz MIPS that nobody could do.
https://www.youtube.com/watch?v=zuMSk_S2ARI
Keep in mind SGI makes (made?) Itanium workstations, and HP is basically the only one pushing Itanium along. In addition, SGI might have NUMA patents and experience useful for expanding their Superdome line.
Indeed. I think HP realise that Itanium really isn’t going anywhere but with the death of PA-RISC are pretty much committed to supporting it until their x86 replacements are up to scratch (cf. their x86-based Superdomes). Bit odd that they canned the decision to port HP-UX to x86 - heck, even VMS is going to make the move. I wonder if HP-UX will die with Itanium?
No-one makes Itanium workstations any more, I’m afraid.
I’ve heard HP kept Itanium on life support to hurt Oracle in a court case. I remember seeing the documents, but don’t remember what or where they are.
Yeah, Oracle dropped support for Itanium, HP sued and won $3 billion.
I can’t help but wonder if some of this was fueled by HP’s failed bid to buy Sun (at around the time Oracle bought them).
These days SGI is pretty much a cut rate rackmount PC manufacturer. Being acquired by HP makes a good deal of sense in that context.
SGI was the company that knew we needed machines that could do 16cores & tens of GB of RAM in one desktop/server all the way up to 2,048 threads w/ 3+TB RAM in single image with tools handling workload. Their engineers knew it but we didn’t. Their management applied their creations in many of the wrong ways. They missed the key market for that kind of desktop hardware.
Web apps. What SGI was doing every day on “high-end” boxes is pretty-much the minimum specs you need to run these modern web pages and applications with responsiveness that matches my Windows 98 on Pentium 2 experience. We have the IRC app with 100MB of RAM. Apps with 50+ dependencies w/ cut & pasted code + lots of temp files that RAMdrives might speed up. Even more if I get into containers and such where at one point people told me I needed a VM for each version of each app, library, or kernel plus one to debug the hypervisor itself. Plus updates while system was running. All that might eat up 16 cores and 64GB of RAM pretty fast. So, SGI’s forward-thinking engineers got busy on the monsters packing 6TB and such. And “visual workstations” + FPGA’s to accelerate the JavaScript engines for games, CAD, and Adobe Photoshop.
Such forward-thinking architecture. I mean, they’re possibly the only ones that got it right. Only HP’s gifted management sees this. They know more browsers, JS engines, clouds, and Nintendo games are coming out. So, they’re getting ready by dropping several hundred million on the visionary company that stayed ready for it. Smart move.