Willamette Willys

February 17, 2000:

I listened rather carefully to the Intel presentations, read some more accounts, and got a few things clarified.

1. The 1.5Ghz Willamette was apparently not air-cooled, though other Willamettes there were. The 1.5Ghz just ran up the frequency demo; other, presumably lower-speed Willamettes actually did real work. These are definitely .18 micron silicon chips.
2. The double-speed ALU (and per Albert Yu, it’s only part of the ALU) is really a “double-pumped” CPU circuit, much DDR is double-pumped. These will handle only very low latency integer operations, not all integer operations.
3. Willamette is much more deeply pipelined than the P6; twenty stages as opposed to ten.
4. There’s some indications that Willamette will rely more on SIMD2 instructions for high FPU performance than its predecessors. In other words, if your application or game isn’t optimized for SIMD2, it could run rather poorly.

Given items two and three, a .18 micron Willamette getting up to around 1.5Ghz eventually is possible. However, a .18 micron Willamette is going to be like a Katmai P3. Intel will be pushing the limits on that chip, and you’ll be better off with a .13 micron Willamette.

Overclockability? This is an interesting issue. Intel is saying Willamette will run at a 400Mhz bus. If that’s the bottom line for bus speed, you’d have to get above that with RAMBUS, which might not be too easy. If early versions of Willamette run at less than that, then you’ll have a possible OCing opportunity.

Someone pointed out that I had been unfair to Intel by not pointing out that the 1 Ghz processor demonstrated last year wasn’t air-cooled, and the Willamette demonstrated was. I guess the person wanted to point out that Willamette was further along that processor was, but now that we’ve found out the 1.5Ghz Willy wasn’t air-cooled, either, I guess it doesn’t much matter.

My point was and remains that Intel wanted to leave the impression that 1.5Ghz was coming Real Soon Now, and it isn’t. Willamette for all practical purposes will be a 2001 chip, and the version you’d want will be the .13 micron variety. That could well be a formidable chip; it’s just not going to be a presence for quite a while.

No matter what you do, the next step beyond Coppermine/current Athlon is going to be an expensive one. You have to figure on having to replace CPU/mobo/memory in one shot at a minimum. Looking to 2001 systems and things like Serial ATA and new PCI standards, you’re effectively looking at replacing hard drives, and maybe some peripherals, too. Video cards go without saying. So the next practical step for many of you with Intel-based systems is to do a relatively cheap upgrade within the next six months with a Celeron2 or Coppermine; then save up for a major upgrade late in 2001.

I’ve left out talking about the Athlon because AMD hasn’t spoken too much about what it plans to do in 2001. We haven’t heard about .13 micron conversion plans, for instance. It looks like AMD will not go to any new 32-bit processors, but move from the Athlon to Sledgehammer, and it’s far too early to even speculate about that.

For many of you, the biggest upgrade you can make will not be your hardware, but your connection. If you are still using 56K with a reasonably modern computer system, that modem is your biggest bottleneck. Take a periodic look into your options, and consider diverting some of that hardware money into a much faster connection.

February 16, 2000:

No doubt you’ve heard about the 1.5Ghz Intel Willamette demonstration.

Before you sell your AMD stock, though, a couple points:

1. Last year, a 1Ghz Coppermine was demonstrated at this conference. Have you seen one yet?
2. Albert Yu did not say the chip would debut at 1.5Ghz, just in excess of 1Ghz.
3. More tellingly, Willamette sales this year were projected to be just in the hundreds of thousands. Sounds like a “virtual vaporchip” for the year 2000 to me.
4. We don’t know whether the chip was made using .18 micron or .13 micron technology. Given this ALU running at twice the speed of the processor, seems like 3Ghz is an awful lot to expect from .18 micron circuitry, notched poly or not.

For people popping off 1.5Ghz in first silicon, Intel sure doesn’t seem to be in any rush to get to 1Ghz; they are still saying 2Q
for 900Mhz, and 3Q for the big gig.

It looks to me like this is the start of the “Waiting for Willy” campaign. Don’t buy one of those Athlons with full-speed cache
running DDR, wait for Willy! Even if Willy won’t be out until 2001. Even if Willy won’t hit 1.5Ghz until well into 2001. Just
don’t buy from the Other Guys.

Well, Willy looks like a 2001 chip, probably a late 2001 chip for overclockers, so don’t let it deter you too much.

February 14, 2000:

Today, at Intel’s Developer’s Forum, we heard that Intel is sticking to RAMBUS for Willamette.

Unless they change their minds, and fast, Intel has just guaranteed the continued success of the Athlon through the year 2001.

Nobody outside of Intel and RAMBUS wants RAMBUS. Not even the inventor’s mother likes it. With one exception (Samsung), not even the memory manufacturers, the people who paid good licensing money to make RAMBUS, want to be bothered with it anymore. Nonetheless, Intel has learned nothing from the fiasco of the last six months.

DDR will be the next wave of the future, first in 200/266Mhz, later in 400Mhz versions. The latter will do as well or better than anything Intel has planned for RDRAM in a Willamette. But Intel says DDR is only going to be used for servers.

If memory manufacturers won’t make RAMBUS, how much do you think the price is going to drop? If Intel insists on making you pay an arm and a leg for memory on a Willamette system, with no BX board to fall back on, isn’t that going to make an Athlon look tremendously good in comparison?

I’m really beginning to wonder if anybody is running the asylum over at Intel. We’ll hear a lot more about Willamette tomorrow, but the tidbits we’ve gotten so far besides this bombshell are not reassuring. First, we get told the Willamette core really isn’t all that different. Then we get told that it may debut at 1.4Ghz.

1.4Ghz is awfully high for a .18 micron chip to run. If Willamette is such a tremendous design, why would it need to debut at such a high speed? More importantly, what do you do for a followup? Seems to me you can’t get it much faster without going to .13 micron migration. Intel hasn’t done .18 micron migration all too well.

When you hear on top of that that the PIII will be around through 2001, sounds like any .13 micron Willamette is going to be a special edition verging-on-vaporware chip for quite a while. I don’t know what AMD’s .13 micron plans are, but given how fast AMD has been moving the past year, I suspect they can easily keep up.

And this is the best-case scenario! If you’re a bit more skeptical, then Intel will be trying to stop Athlon sales with its mouth rather than products. This can only last so long. Even the stock analysts are beginning to realize Intel is playing Oz and becoming the leading vaporchip manufacturer.

Intel may get away with a “Waiting for Willamette” during the summer, but the audience will just get more and more skeptical, and if Willamette doesn’t blow the doors off Athlon come the fall, Intel’s going to have some real problems.

Email Ed