Intel: Prescott Reloaded
The 600-series is the most interesting of the bunch, with EM64T capability, 2Mb cache, and power-consumption tweaks.
Unlike the “Matrix Reloaded,” version 2.0 is a better product than 1.0, and is probably good enough to keep the Dell and Company folks happy.
However, what is good enough for an OEM isn’t necessarily good enough for overclockers. Prescott 2.0 is still awfully hot, and this overclocking attempt shows that while Prescott 2.0 may have moved a couple of circles of hell up, it hasn’t left the realm. Serious overclocking will still require water or better.
What these chips do is get it close to current Hammer overclocking, the pre-strained silicon type. Actually, that would be a good name for these chips in general: Close. Close in performance, close in x86-64 functionality, closer (well, not all that close) in power consumption.
The problem is, unless you have something close to an Intel fetish, it’s “close but no cigar” from the overclocker perspective.
If strained silicon at 90nm gives even a fairly modest boost to maximum speeds, the strained silicon Hammers ought to widen the gap. A 2.9GHz socket 939 Hammer ought to be roughly equivalent to a 4.4-4.6GHz Prescott, a 3GHz should come in at about the 4.6-4.8GHz range.
If you still want one nonetheless, you might find it wise to wait until Smithfield-compatible motherboards are available. No point being obsolesced in three months.
Unfortunately, the AMD news is not about strained silicon. It’s more like, “Intel giveth and AMD taketh away.”
Sempron: Make It Suck More!
XBit Labs reports the following:
Advanced Micro Devices has begun to sell its AMD Sempron processors produced using 90nm process technology . . . . AMD Sempron 2600+, 2700+ and 3000+ chips with built-in PC3200 memory controller are marked as SDA3000AIO2BA, SDA2800AIO3BA and SDA2600AIO2BA . . . . According to various online stores, AMD Sempron processor 3000+ operates at 1.80GHz and has 128KB L2 cache, the Sempron 2800+ functions at 1.60GHz and is equipped with 256KB L2 cache, the Sempron model 2600+ runs at 1.60GHz and builds-in 128KB of L2 cache.
So while Intel is doubling the cache on its 600-series to make Prescotts suck a bit less, AMD is halving the cache on some Semprons to make them suck more.
How much more? Testing will determine that, but probably not a whole lot: a 1.8GHz Sempron with 128K cache is rated as a (Celeron PR based) 3000+ while a 1.8GHz Sempron with 256K is rated just a bit higher at 3100+. Figure another 3-5% performance penalty.
How do you tell the difference between the two types? A 128K cache Sempron has a code like SDA3000AIO2BA. “2” means 128K.
A 256K Sempron has a code like SDA2800AIO3BA. “3” stands for 256K.
Why is AMD doing this? No doubt because Semprons are too good for AMD’s own good. Due to Hammer’s 128K L1 design and onboard memory controller, smaller caches just don’t have the performance penalty they do on the Intel side.
But gee, shouldn’t they have been the first to know that? Why is this such a shock to them?
It’s just another sign of AMD running the company by the seat of their pants. Initially, Semprons were supposed to be trailing-edge technology, staying at 130nm for a while while big brother went to 90nm. Oops, that costs too much, better make 90s right away. Semprons weren’t supposed to have x86-64; oops, Intel is going to put them into Celerons, we’re going to have to change that when they do. Oh, oh, after we do that, these Semprons will be almost as good as the good expensive stuff. Get the knife!
I’m sorry, but these were not events out of the blue, no huge surprises. Frankly, if I can figure out these things ahead of time, people getting paid many hundreds of thousands of dollars veering towards the low millions with stock options ought to also.
They didn’t, so we end up with the spectacle of AMD publicly making their chips worse than they were before. While the “why” is understandable, at the least, it’s terrible PR.
Even worse, AMD will sell both versions at 90nm, and these dumbed-down Semprons will cost only $10 less than their better-endowed brothers, and some less-endowed will be rated (and priced higher) than their slower better-cached brethren.
Look, if you’re concerned that people will buy 256K Semprons rather than 512K Athlon 64s, then you cut all the 90nm Semprons down to size. If that’s your fear, it’s ridiculous to continue selling the same product that scares you.
And if that’s not the reason for doing it, what is?
We’ll just have to see what AMD ends up selling once they get forced into adding AMD64 to Semprons. After this news, you might think such empowered CPUs would all be 128K to “compensate” for such an unfortunate improvement. But that would make sense, and that’s the one thing you can’t count on from this company.
Dazed and confused, it’s the same old story.