Slashdot Mirror
Sun Sacks UltraSparc V and 3300 Employees
bender writes "According to this article, Sun Microsystems has cancelled the next generation UltraSparc V processor even though the chip had already taped out. Perhaps this has something to do with the recent partnerships with AMD and Fujitsu?"
This is most unfortunate since the UltraSPARC line was extremely efficient. Under heavy loads even an UltraSPARC II with 128MB of RAM could outperform an Intel chip with ten times the RAM.
From The Register Yesterday: Sun shelves UltraSPARC V in favor of the great unknown
Please, if you're going to enourage your readerbase to read an article, please do the same.
Sun said nothing about laying off the Ultrasparc V or Gemini staff.
"Sun plans to lay off 3,300 employees, but many from the UltraSparc V and Gemini projects will remain at Sun, the spokeswoman said."
You mean here
Another story is here, which explains things a bit more clearly.
I agree -- Sun was doing so well during the dotcom days that they totally lost track of their competitive position in the market.
On the high-end, the death of SPARC was a long time coming, yet Sun continued to plow massive amounts of money into a chip that was not competing with POWER etc.
On the low-end, they didn't do anything about the growth of Linux except diss it. Had they positioned Solaris x86 strongly against Linux back in the RedHat 5/6 days, they would have killed alot of Linux's market growth -- remember back in 1999, UNIX was the "safe choice" and Linux was not. But Solaris x86 was so obviously an orphan product that nobody took it seriously.
Instead they spent a lot of time bashing Microsoft (not their #1 competitor) and farting around with things like StarOffice. And planting trees.
Business. Numbers. Money. People. Computer World.
The vast majority of money put toward a chip is in the design, not the manufacturing.
The intro only talks about it being taped out. That isn't the end of the design effort. In fact, that's when the really expensive validation work begins. Now its true that the amount of people (and thus salaries) goes down, but the really expensive validation phase ususally consumes more than half the R&D of a development. Heck, a single machine configuration to run benchmarks runs in the Millions of dollars (US).
to the Engineer, the glass is neither half full nor half empty. Its just two times too big.
They didn't give up on it... they finished it.
.5Million just for the mask set.
Not quite. Big chips almost never work right the first time. Minor design changes are always required. Best case, Ultrasparc V was months and millions of dollars away from done. Each "spin" throught he fab is
I suspect the situation for Ultrasparc V was worse than that. If they had truly taped out then the chip would already be in the fab. More likely, the database was in condition that it could have been fabed but it was not meeting performance targets.
Did we read the same link?
It starts with
"Santa Clara, Calif.-based Sun has stopped work on the UltraSparc V, a server chip"
The key here is: "Stopped work on"
Ultra-Sparc is alive and well! If anything, Sun seems to be freeing up some engineers to work on the more promising future versions. As long as these extra hands and eyes don't slow things down (now, who's law is that?), this will probably be a good thing.
Sometimes boldness is in fashion. Sometimes only the brave will be bold.
Gah! I work on the Santa Clara campus and watched them dig up & box those trees. They were being moved, not brought in.
It had nothing to do with Sun. The trees were growing in a large vacant acreage next to the campus that is now being developed into a retirement community to pair with the 2500 housing unit development being constructed across Agnew Blvd next to the shopping center.
BTW, they weren't 100ft tall, maybe 75, with trunks about 3-4 ft' in diameter at waist level.
... taped out, an expression that means the design was complete. (In the olden days, when engineers completed a chip design, they sent the computer tape out to other groups.)
Err, what? I thought this bit of jargon came from the process of creating a photomask by manually applying tape to a pane of glass. Am I mistaken?
Java: the COBOL of the new millenium.
so, as somebody who used to work on the US-5...
:)
as i AC-posted above, UltraSparc V did, in fact, tape out (sometime in the last month, or so). supposedly they actually pulled some wafers off the fab line last week.
the project was a couple of years late. it was supposed to tape out last summer, and that was the second estimate after the first one didn't look possible. there were a lot of very, very smart people working on that project, but management was a bit misguided. as one of the mid-level engineers, i'd blame that mostly on the fact that a lot of our lower-level managers were high-level engineers who were yanked up into management positions. but that's probably just my personal bias.
the cool thing, from my geek perspective, about the chip was that it was truely multithreaded - one core that could be run as one pipeline for apps with greater internal parallelism, or as two pipelines for more throughput. unfortunately we ran into a lot of technical problems making the multithreading work efficiently - that was a big part of what i was responsible for while i was there
Sun plans to lay off 3,300 employees, but many from the UltraSparc V and Gemini projects will remain at Sun, the spokeswoman said.
lower level management (project director level) is much more pessimistic; they expect less than half of those laid off will stay with the company.
anyway, i was pretty suprised when they axed the entire project, but i guess with all the talk about "throughput computing" (read: processors composed of lots of simple cores stacked up side-by-side with shared caches) it shouldn't have shocked me.
Sun spent a huge amount of money creating an alibi for McNealy when the company went down the toilet.
Sun's situation had nothing to do with Microsoft, their market is eroding because of Linux and cheap commodity hardware. They would be in serious trouble even if Linux had never been written, the cost of an Intel box plus a traditional Unix license is much less than the cost of the competing Sun box.
Sun has been going 'upmarket' for the past ten years. Read Clayton Christiansen and 'The Innovator's Solution' to understand why that is a long term strategic disaster. The market for large servers was a temporary phenomena that was always going to end up being turned into a commodity. Ten years ago a workstation was essential if you were going to do serious academic research in the comp/sci field. Today an Intel or AMD box is 'good enough' for 98% of users.
The trap that Sun is in is that a commodity Intel box is overkill for the vast majority of applications. The only customers who are buying Solaris are doing so because they have a big investment in legacy Solaris infrastructure.
As for the settlement, I'll believe it when we see the SEC statements. Microsoft has a long history of making settlements where the headline figure is much bigger than the real figure.
Looking for an Information Security student project suggestion?
Try http://dotcrimeManifesto.com/
The Mono JIT on SPARC passes all the tests that the
Mono x86 JIT passes.
A lot of the recent focus has been on taking
advantage of many of the SPARC v9 features
(like branch prediction) and improving the code
generation after the initial feature complete
stage.
Are you talking about the rotating register file? Sparc has a large collection of registers, of which a subset are addressible by the register-register instructions at any given time. You can move the window (which determines which subset is visible) with single user-mode instructions, which typically are used on entering and exiting procedures. They are aligned so that one chunk of the set (8 regs) is shared between a caller and callee, which makes for very efficient parameter passing.
In the days before out-of-order superscalar execution, I ran timing experiments comparing this system with comparable RISC processors without this feature. If you like to write programs with lots of very small nested functions, then the reduction of function call overhead can be significant, as much as 2x improvement. It's much less of an issue on modern CPUs with out-of-order execution, as the stack area used for parameter passing will generally be sitting in the L1 cache, close to the CPU.
I don't see your point about process switching. It's not much of an issue for ordinary systems. A process switch occurs, what, like every 10ms? Saving a few dozen load and store instructions might save you 10-20ns? BFD.
I'm sure someone on /. has said that many times over, but it's not at all the case.
For those who are too lazy to read the article, Sun is NOT killing off the SPARC line, they are NOT discontinuing all their CPU production and they are NOT switching everything to AMD64 chips.
What Sun is doing is finally putting an end to their rather unsuccessful attempts to produce a single-threaded raw number crunching chip. Sun hasn't been successful at this for some time now (certainly since at least the UltraSparc II and probably for a while before then) and the UltraSparc V was just going to be another failure in this regard. No one buys Sun's for their raw number crunching performance anyway (since they stink in this regard), so this is really a pretty bright move by Sun. Really it's something they should have done a while ago.
The plan going forward is for Sun to work to their strengths. Their CPU division will produce highly multithreaded chips that are designed for server work, ie the sort of stuff that people buy Suns for in the first place. Their workstation line will be replaced by AMD64 systems since EVERYONE is moving their workstation line to x86 anyway. The only thing holding people to Sun workstations (and IBM or SGI workstations as well) was the lack of 64-bit capabilities on x86 chips, but that restriction is no more.
Sun will still need some SPARC workstation products for a while going forward to support customers with legacy Solaris software that can't easily be upgraded though. If they are smart, what Sun will do is buy some SPARC64-V chips from Fujitsu. This gives Sun faster chips for much lower cost then developping their own.