Sempron, The Product

Enough with the name. What about the product?

The moles talking to XBitLabs confirmed that the product being talked about is in fact the long-known about Paris/Victoria.

In a nutshell, socket 754 Hammer with memory controller, single-channel memory, no x86-64, only 256K cache.

Essentially AMD’s version of the Celeron line.

Pros

If some Joe Sixpack friend of mine wanted a cheap computer, and these were available today and were fairly inexpensive, sure, I’d use one today. It should offer a decent performance increase over Athlon XPs due to the onboard memory controller

Whether I’d use one (or, more accurately, Joe would buy one on his own) early next year may be quite a different matter.

Cons

x86-64 is supposed to be disabled in these chips. This has always struck a lot of people as being ideologically inconsistent. How can AMD proclaim that x86-64 is the future when one of their major product lines doesn’t use it?

Now that Intel has joined the bandwagon and will extend x86-64 capabilitity to Celerons, this now makes very little sense at all.

Current Athlon XPs pretty much whip current Celerons. A Sempron would whip it even worse, in 32-bit.

So what’s a poor major CPU company to do?

One blindingly obvious possibility is for Intel to market their x86-64 in OEM boxes, especially Celeron boxes, as hard as they can. x86-64 capability ought to get them back in the ballgame performance-wise, and even more so marketing-wise.

As we’ve pointed out before, adopting x86-64 technology for Celerons puts Intel in the position of being able to beat AMD with its own stick.

Perhaps more importantly, if new computers with Windows for x86-64 becomes the standard in 2005 because Intel says so, Semprons become the odd man out.

This is a problem I don’t think AMD execs quite understand. You can’t charge extra for something your competition is giving away. Right now, AMD is trying to do that with dual-channel in socket 939. Later, Semprons will indirectly try to do that with x86-64.

If Intel gives away x86-64 in sub-$100 Celerons, the likely customer reaction will be to treat Semprons as being worth less than Celerons, and we’re right back in the cheap processor trap.

AMD’s Dilemna

If you keep x86-64 out of the picture, there probably won’t be much performance difference between a 256K Paris and a 512K Newcastle, probably no more than the difference between a TBredB and a Barton.

Ironically, AMD is penalized for being good. The large L1 cache found in all recent AMD processors, bargain or not, means that less L2 cache makes less of a performance difference than it does with Intel CPUs.

I’m sure that in the minds of some AMD execs, disabling x86-64 was a means to widen the performance gap between Athlon 64s and Semprons to Pentium/Celeron levels, and thus indirectly increase the value of the 64s.

So long as Intel was giving the finger to x86-64, this made some sense, but Intel jumping on that bandwageon blows that strategy to hell. x86-64 no longer will be something you can charge extra for, since the competition isn’t.

On the other hand, though, if AMD does enable x86-64 on Semprons, given the likely performance differences, many AMD buyers will simply move downstream.

They’re damned if they do, they’re damned if they don’t. The only difference is in the damner.

A Solution?

I don’t think AMD has any realistic choice but to eventually enable x86-64 should Intel provide it in Celerons. Best bet is that they’ll see what Intel does with x86-64 and Celerons then change their minds if Intel pushes it. That’s probably a good reason not to be the first kid on your block with one.

AMD apparently plans on bringing these “Semprons” to socket 939, too. It’s hard to see how the “Value” processors described here could be anything else.

This is a bad idea. Leave the Semprons to socket 754. While dual-channel doesn’t offer a huge difference in performance, any performance difference helps.

Sempron memory timings could be slowed down a bit, that would add a few more percentage points.

Perhaps the x86-64 circuitry for Semprons could be slowed down a bit.

Semprons could be priced a little higher, and entry-level Athlon 64s a little lower to make going cheap less attractive.

More cache on Athlon 64s and FXs would help. With Intel planning on more 2Mb cache Prescotts at the end of the year (and already providing 1Mb on all Prescotts); it would make sense for 90nm Athlon 64s to have 1Mb cache, and FXs 2Mb.

While it would be hard to AMD to match the performance difference between Pentiums and Celerons, they probably can do enough to widen the gap between Athlon 64s and Semptrons to make market segmenting work.

The question is “Will they do it before they get beat up, or after?”

Email Ed

Be the first to comment

Leave a Reply