Microprocessor Startup Apocalypse?

This past week, I was a participant in the annual Hot Chips symposium at Stanford University. As part of the organizing committee, I help the conference with PR, including press and analysts. The conference is a mix of academic and commercial presentations, but the general leaning of the conference is more commercial – especially after the demise of the Microprocessor Forum.

This year’s attendance was up sharply from 2009 and pretty much matched 2007 and 2008. That’s a healthy sign for the business – companies are spending money to send people to conferences. But there’s a dark side to this year’s conference – paper submissions was down. Many of the papers this year came from stalwarts AMD, IBM, and Intel. There were a few new companies, but one presenter’s company (Tier Logic) had already been declared dead by Microprocessor Report (July 19th edition - $$)

And on the same day the conference kicked off, Jim Turley, editor in chief of Microprocessor Report, released an editorial on the futility of new microprocessor startups. With a title of “How to Blow $100 Million,” the editorial goes on to develop a rule of thumb that a new microprocessor start-up needs about $100 million in order to fully launch the company. After the idea rattled off a few people, the general consensus that if the new microprocessor requires extensive tools support (compilers, debuggers, etc), that $100 million is not ridiculously high.

And there is the problem – how can you raise a $100 million form VCs these days, when the risks are high, the returns questionable, and many other “sexier” markets (like green tech, social media, location aware services) are out there. A look at recent microprocessor start-up is not encouraging: Raza Micro (where a spent a few months) was sold to Net Logic, PA Semi was sold to Apple, Intrinsity was also sold to Apple, Montalvo was shuttered, Stream Processor Inc. (SPI) is no more, and the list goes on.
I find it sad and disheartening that the microprocessor business is undergoing such major consolidations and innovative new architectures will likely not get funded. The next new architecture may come from the emerging powers such as China. We saw the Chine “Godson” processor evolve rapidly until now we see a are talking about a processor with a wide SIMD unit, running over 1GHz that soon could be the foundation of a major supercomputer. They even talk about an x86 emulations software layer supported by special instructions. With the support of the government and access to a competitive foundry, I wouldn’t underestimate where there program could be a few more years.

Intel just announced that Larrabee is officially, really, NOT going to be released as a discrete GPU any time soon. Or for the next 5 years. Or more.

Why, you may ask? Well, because it made a really weak GPU that would probably cost too much to manufacture and have returned bad margins. How could the mighty and wise Intel have spent over three years working on this project and not know it would suck as a GPU? Well, it could be because they were not reading my blog post of December 5, 2007 where I said:

"Oh, and when did we all decide that x86 was the most perfect instruction set for graphics? Hmmmm, it's not. The example where the x86 is the preferred instruction set for either graphics or high performance computing doesn't exist. The x86 instruction is an burden on the chip, not an advantage, for graphics. In fact, graphics instruction sets are hidden by the OpenGL and DirectX API, so the only people who care are the driver programmers and we are free to design whatever instruction set is most efficient for the task, while Intel has handcuffed themselves with a fixed instruction set. Not the smartest move for graphics. Intel's plans may work better for HPC though. Still, there are many alternatives to x86 that are actually better (shocking isn't it)!"

"Larrabee is based on a simplified x86 core with an extended SSE-like SIMD processing unit. In order to be efficient, the extended-SSE unit needs to be packed efficiently. How is Intel going to do that? The magic is all left to the software compiler, which is going have a tough time finding that much parallelism efficiently."

And that was what happened. In addition, with many coherent cores, the coherency traffic and stalls eat up a lot of bandwidth and limited scaling across cores.

My prediction back then gave Intel too much credit:
"But you heard it hear first: Larrabee will not reach the potential performance capability (by a long shot) and will not displace NVIDIA as GPU leader."
Because Larrabee never even got to compete.

Oh and I still stand behind this quote:
"Let me also ask this question: when has Intel EVER produced a high performance graphic chip? (Answer: never)"

A hard lesson learned by the i guys. But I could have saved them all those many millions of dollars. If only they'd have listened. Sigh.

Netbooks or mini-notebooks, no matter the name you call them, they are the hottest topic in the PC market. My company finally jumped into the fray with the ION platform which combines the very low power Atom processor with our very capable 9400 chipset. This is a very capably platform that would only get better when Intel releases a dual-core Atom processor. The aging 945 chipset from Intel is an anchor on the speedy little Atom processor.

From what I've gathered so far, the appeal of mini-notes is the combination of low price (about the cost of an iPOD touch, which makes you wonder about how high Apple's margins must be), connectivity (WiFi), battery life, and portability (small size and light weight).

The first version of the mini-notes, exemplified by the ASUS Eee PC, where an odd hot with Linux running on underclocked Celeron processors (the so-called 900MHz Celeron was underclocked to ~630MHz), small 7" displays and crammed keyboards. Still there was something attactive about a sub-$400 notebook you could take anywhere in a small bag (not quite pocketable). As the catagory has evolved the keyboards are bigger, the displays are bigger, and added more mainstream WinXP operating systems.

After shipping about 10 million units in 2008, the estimates for 2009 are up to 35 million units. On one hand, this is an opportunity to expand the markets for PCs by inproving the portability, but it's also a chance to crush the system prices to bargain basement levels. We might find that these PC's become as disposable as cell phones with a shortened life span. PC life spans had been increasing, but Netbooks could lower it again. So the margins aren't great, but the shorter replacement cycle could be a good thing for PC OEMs.

Now for something close to my heart: microprocessors. I have a question for you: is microprocessor design dying? I ask, because chips like Intel's Atom are perceived as innovative, when, in fact, Atom is a through back design. An analyst was asked by a publication is he thought Atom would be one of the innovations of the year. Atom is dual-issue, in-order design where low power trumps performance. The architecture is stripped down and basic, not innovative. The innovation is in the marketing of it.

Another piece of evidence: MIPS is planning as layoff. It's been pretty obvious that MIPS ran out of ideas and has been coasting on the designs of Cavium and RMI to keep it relevant and setup boxes and PS2 for revenue. It was destined to be squeezed by POWER from above and Arm from below, but MIPS gave in with a whimper, not a real fight.

The big iron guys like IBM and Sun still keep moving along, but the excitement now comes from the promise of many smaller cores working together (what Sun calls Throughput Computing) with processors like Niagara and Larrabee, only there's this problem with creating mainstream software for many-core processors. And if the goal IS more cores/mm2, those cores are going to be rather simple in design (like the x86 cores in Larrabee).

So, we dump the problem onto the programmers and sit back and yell: "What's the problem? Can't you programmers figure out how to split that code over a bunch of cores? What's so hard about that?"

Well for graphics, it isn't too tough, because there's all these pixel to work on in parallel. It's also reasonably well known in data base, web serving, and some super-computer programming. But it's still going to be a challenge for many other applications.

Then again, maybe we have to ask: are processors already good enough for typical client workloads? Therefore, we don't need particularly powerful client microprocessors, just really good Internet connectivity to cloud computing and low power. Gamers and professional content creators will be the lone users of powerful client computers. Like with cars, everybody is worried about gas mileage today and only a few crazies still care about performance.

I'm just asking.

Back again with some time to post. There's been some interesting news I'd like to cover.

First, Apple introduced new products at MacWorld. The most controversial product was the MacBook Air, the thinnest notebook shipping today (maybe, the thinnest ever). I found the Air made too many compromises to be anything but an executive toy. I know something about sub-notebooks having owned a Fujitsu P2040 with the Transmeta Crusoe processor and an having recently purchased an ASUS Eee PC. Both of those laptops made compromises, but Apple chose the style, keyboard, and larger display over flexibility, connectivity, and expandability. While I will wait until I see one in person before casting my final opinion, I can't imagine that the one USB port on the Air is in any way sufficient for external mic/headset, wired Ethernet, external storage, etc. Beside, the Air turns out to be Apple's slowest notebook and is using Intel's older 65nm processor, not the latest 45nm Penryn processor.

The other big news (as far as I'm concerned) is the announcement of VIA's "Isaiah" processor (also known as the CN processor). This is a major leap forward for VIA's Centaur design team. The processor had been announced at the 2004 Fall Microprocessor Forum, back when I was still at Microprocessor Report. Glenn Henry announced the design at the 2004 Fall MPF and gives a detailed description at ExtremeTech.com.

In looking over the wafer pictures, I figure the die is about 65mm2 in size, larger
than I thought it would be in 2004, but then I didn't expect VIA to put a 1MB L2 cache on die. I also expected VIA would put an on-die memory controller in the design, but they did not. I guess VIA prefers a clean division of labor - processors in Texas and chipsets in Taiwan.

Overall the processor looks interesting, splitting the difference between the Core 2 processors (Penryn is 107mm2) and the upcoming UMPC/MID processor Silverthorne (at 25mm2). Performance should be right in the middle of those two Intel processor families. The question is: is there enough room for VIA to operate or is it too tight a squeeze.

As I've thought more about Larrabee, I have some additional observations:
I'd like to clarify my point about the x86 instruction set. It's not innately bad, but it doesn't progress the architecture. The x86 ISA has too few registers, an irregular instruction length, and highly irregular instruction decoding - none of which help with architectural efficiency. It's just all the work and R&D put into software for the x86 architecture that compensates for its shortcomings. But if given a clean slate, few would chose the x86 ISA for maximum performance or efficiency. My point is that x86 is largely irrelevant for a GPU. The x86 will help Intel in the HPC market.

With regard to the wide SIMD (SSE) structure - we'll have to see how effective Intel is in packing useful operations. It seems that the width was chosen in order to hit certain theoretical performance metrics, not for realistic workloads. The same ambition infected the Cell processor and we're still waiting for software to approach those highly promoted peak numbers.

I remain convinced that Intel doesn't have the formula to produce a world class GPU. I see no sign they have the specialized talents to make it succeed. At best, it will be a reasonably sized die with mediocre performance, but it will have Intel's brand name behind it. That will sell some units, but I hope Intel is ready for the realities of GPU ASPs in the mainstream of the market. At least it should be better than the existing Extreme (-ly bad) Graphics they have today.

I also realized that most PC buyers have no idea how bad Intel's integrated graphics really is. The gamer sites never test Intel graphics because they know how bad it is and it's not even worth the effort. But the mainstream consumers don't know how bad it is because NO ONE TELLS THEM! Both ATI and NVIDIA are guilty here because we still have to work with the Intel chipset group and neither company has been willing to take Intel on directly. Shame on us. It's also not a topic that consumer publications make a big point about, as many in the industry consider it common knowledge - its not new news to them. So Intel gets a free ride. I was waiting for AMD to push it's platform strategy and call out Intel graphics, but I haven't seem it yet. Maybe when its new DX10 chipset ships in Q1.

With regard to Larrabee's supposed process advantage, I find it difficult to image Intel will use the leading edge process for Larrabee. I expect the chip and its successors will be 6 months to a year behind the leading edge of the process. If that's the case, then TSMC will be very competitive. In addition, our design process is great at turning around designs quickly, Intel's process is geared more toward die optimizations with its custom circuit design. Intel will likely make Larrabee coherent so that multi-chip scaling will allow Intel to make the one die and still create multiple graphics card solutions. We tend toward multiple design solutions with one die optimized for a specific price point. AMD may also be moving to the same idea as Intel. We'll have to follow that development closely.


So, if anyone else would like to comment, please add your voice. I do moderate the comments, but I promise to publish any reasonable response.

My last post did generate a response, so check it out below and my response to the comment.

I should add that I was talking from my perspective. NVIDIA does take the Larrabee program very seriously. I see it more as a research project gone amok.

What I also find amusing is that Intel, for many years, tried to run away from the x86 instruction set with i960, i860, i432, EPIC (Itanium), and xScale, and now x86 is the answer to all applications. Fred Weber when he was CTO of AMD was the first I know to active promote the "x86 everywhere" strategy. While AMD still embraces this notion, I haven't seen them approach graphics directly with the x86 instruction set.
I think that AMD's Fusion is a wiser move overall.

It's a bit late (actually you may note I tend to blog late at night anyway), but I've been wanting to take on a topic that's of interest to me from both an NVIDIA point of view and my Microprocessor Report analyst background. It's the chances that Intel's Larrabee will be a successful GPU.

I would like to offer the opinion that the deck is stacked against it, despite Intel's hype machine pushing it. For one, the design looks more like the IBM Cell processor or the Sun Niagara (UltraSPARC T2) than a graphics chips. The Cell was proved to be ineffectual as a GPU (which explains why we got the graphics chip in the PS3) and Niagara is focused on Web services (lots of threads to service).

Let me also ask this question: when has Intel EVER produced a high performance graphic chip? (Answer: never)

Oh, and when did we all decide that x86 was the most perfect instruction set for graphics? Hmmmm, it's not. The example where the x86 is the preferred instruction set for either graphics or high performance computing doesn't exist. The x86 instruction is an burden on the chip, not an advantage, for graphics. In fact, graphics instruction sets are hidden by the OpenGL and DirectX API, so the only people who care are the driver programmers and we are free to design whatever instruction set is most efficient for the task, while Intel has handcuffed themselves with a fixed instruction set. Not the smartest move for graphics. Intel's plans may work better for HPC though. Still, there are many alternatives to x86 that are actually better (shocking isn't it)!

Larrabee is based on a simplified x86 core with an extended SSE-like SIMD processing unit. In order to be efficient, the extended-SSE unit needs to be packed efficiently. How is Intel going to do that? The magic is all left to the software compiler, which is going have a tough time finding that much parallelism efficiently.

Larrabee will have impressive theoretical specs that will wow the press, but it's going to be very hard to use the chip up to its potential, which sounds exactly like Cell. And Cell has not lived up to its hype. So my corollary is Larrabee = Cell; Cell not = to hype; Larrabee will not = hype.

Now we wait until sometime in mid-2008 to see some silicon. But you heard it hear first: Larrabee will not reach the potential performance capability (by a long shot) and will not displace NVIDIA as GPU leader.

Hi Ho Everybody, I'm back. I haven't had a lot of time with the Eee PC, but it's still reasonably nice. I'll like it a lot better when ASUS increases the screen resolution. It would also be nice to move up to a Core 2 processor with real power management, but these are the trade-offs you make when its only $400.

I've been in the middle of thinking about AMD, Intel, and NVIDIA as we all do this dance around the ring - its like a three-way fight where we take turns helping and hurting each other. I'd call us the Good, the Bad, and the Ugly (I'd let you all decide who is who, but I know my choices).

On the topic of ASUS, I'm beginning to wonder about the quality of their motherboards. Both my stepson and I have problems with the motherboards not booting properly and these are two completely different motherboards. How hard can that be to get right?

One project I recently completed was taking video at the memorial for the mother of my wife's brother in-law. I used Pinnacle Studio 10.5 and found it pretty easy to use and create a good basic video with fades and titles. I want to do more work with it over the holidays.