![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
lprovenI've had a surprisingly popular response to techie pieces of mine in the past, so once again, I'm posting a piece from CIX that grew a little bit longer than I intended. To nearly 1,400 words, in fact.
I'm not fond of Intel's marketing tactics - I think they play dirty - and I've been using non-Intel x86 chips from preference for nearly a decade now, from a 6x86 to a 6x86 MX to a 6x86 MII to an AMD K6-2 to a dual Athlon XP. (The K6 was recently, plus one in my ex-girlfriend's PC years ago.)
I confess that between those two pairs of AMDs, though, I did go for a PII-450, as for a while there, there just wasn't anything to compete. The PIII was a bit of a con, too, though, until the 133MHz FSB “Coppermine” models appeared. The early PIIIs were just PIIs with improved MMX and were no faster than a PII at the same clock speed.
I find these non-Intel x86 chips absolutely fine. The Cyrix ones gave me very minor problems occasionally: some motherboard BIOSes didn't recognise them correctly, which was usually just a cosmetic problem – the BIOS POST screen would be wrong. Big deal. However, I also found that PageMaker 6.5 (or 7, I forget) wouldn't run on a Cyrix machine. That was the only app I ever had a problem with until Windows Server 2003, which also wouldn't - it couldn't even boot. However, by the time WS2003 came out, the 6x86 MII was /ancient/, so it's kinda fair enough. I think VirtualPC also had some problems but the same applies and I never even tried to run it on the old Cyrix machine. (I had a PIII laptop by then. I would have liked to have gone for a Transmeta Crusoe machine, but the purchase was extremely price-critical and performance was more important to me than light weight and long battery life.)
AMDs have never given me even that much problem. They deliver measurably better performance than Intel chips at much lower clock speeds *and* they're significantly cheaper for equivalent performance. With the K7 family - the Athlons and above - you do need a different motherboard chipset and CPU socket. (K6s and 6x86s used the Pentium's Socket 7.) This does make direct swap impossible and means performance comparisons are a bit harder, but they give much better bang for the buck and the company doesn't behave as badly as Intel. 'Nuff said. I also recommend them to clients for workstations and indeed servers when I can find server manufacturers who can supply AMD-based units - these are lamentably rare, but getting better in these days of the x86-64.
And on 64 bit chips, it's worth noting that the P4s and Xeons with "IA32e" - Intel's marketing name for x86-64 - bolted on are no faster than standard ones. Indeed, the "Prescott" P4 - generally believe to be the first version with IA32e, tho' it was disabled on all the early chips - they're only enabling it now, as the P4 approaches end-of-life - was actually slightly /slower/ than the earlier Northwood P4. In comparison, AMD's Opteron is an all-new chip. (The AThlon64 is a marketing name; it's no relation to the Athlon CPU.) AMD x86-64 CPUs are considerably faster, at the same clock speed, than their 32-bit predecessors, which were already considerably faster than a P4 at the same speed.
The P4 itself was a triumph of hype and marketing over good design. It was solely designed to run at high clock speeds, because these were linked in the buying public's mind with performance - due largely to Intel marketing. Cyrix and AMD were thus forced into responding with PR ratings, a number showing the approximate MHz of an equivalent Intel chip. So my old Cyrix 6x86-166+ actually ran at 133MHz but was equivalent to a Pentium at 166MHz. My dual Athlon machine has 2000+ processors; they actually run at 1666MHz but are roughly equivalent to 2GHz P4s. This meant they could sell effectively against Intel, so Intel moved the goalposts by producing the P4, which could run at nearly twice the MHz of the PIII - although in doing so it delivered little or no extra performance. Indeed, the early PIIs at 233MHz were little or no faster than the P1 at that speed. Intel has been falling behind in CPU development for many years, and has retained its market pre-eminence by clever and extremely aggressive marketing.
So a 1.8GHz P4 actually compared badly to a PIII at slightly over 1GHz. However, that big number *looks* impressive, so people bought them. AMD couldn't compete. A new PR number showing the speed of P4 they were equivalent to might have worked, but then owners of earlier Athlons, whose numbers were scaled against the PIII, would complain. If you upgraded an Athlon 1500 (PIII scale) to an Athlon 3500 (P4 scale) and found it was only 10-20% faster, you'd be pretty pissed off.
What's changed this is the advent of the Opteron and x86-64. It’s moved the goalposts – firstly, because it performs much better on 32-bit code (unlike Intel’s x86-64 chips, which run 32-bit code at the exact same speed as ever, or its IA64 Itanium chips, which run it /substantially/ slower), and secondly, because 64-bit is a new ballgame. For best results, you need new 64-bit apps on a new 64-bit OS.
What do I mean by best results? Well, benchmarks show that for CPU-intensive applications like PDF rendering or image manipulation, 64-bit code on a 64-bit OS is some 25% faster than 32-bit code on a 32-bit OS. You don’t get those benefits if you run 32-bit code on a 64-bit OS.
It’s not so bad for AMD, as its x86-64 bit processors run 32-bit code faster than ever anyway.
But it’s bad for Intel, because its don’t. Indeed, even its non-x86 Itanium processors are slower than its latest x86 ones.
It’s also bad for MS because its 64-bit OSs /still/ aren’t fully finalized and finished and it has few to no 64-bit apps on offer yet.
So, for this new game, AMD dropped its P-ratings and has just started giving its CPUs model numbers.
Intel, aware that it’s in danger of losing the game altogether, has eaten humble pie, cloned its much smaller competitor’s instruction set, and introduced AMD-compatible processors: the P4 and Xeons with “IA32e”. And aware that their performance ratings aren’t looking that impressive any more, especially since they’ve failed to come up with a 4GHz part, and indeed given up on doing so, have also copied AMD and gone over to model numbers.
Meanwhile, the entire future development of the P4 line has been cancelled and development is focusing on the Pentium M instead. This is essentially a PIII with improved branch prediction and a P4 bus. It was designed and optimized for low-power use, i.e. notebooks, but it’s starting to move sideways into desktops. Now what they have to do is re-engineer it to handle x86-64.
Then the next big thing will be multi-core processors. These are single chips with more than one CPU core: two processors on one chip. The RISC chip makers like HP, Sun and IBM have been doing this for years. Here, again, Intel is behind the curve.
AMD is close behind the RISC vendors. Intel is some way off. They’ve demonstrated a machine, but no-one was allowed to look inside and it’s widely considered to have been faked with 2 separate chips.
AMD will be shipping these soon: early next year. Intel will probably follow at the end of the year or in 2006.
It may be that multi-core Itaniums and P4s will come first, if so, the latter will probably only be sold under the Xeon brand, as server processors.
Again, here, Intel has to go back some ways and start designing multi-core Pentium M chips. They won’t ship any time soon.
Summary: unless you like paying over the odds for an underperforming brand name, don’t buy Intel. For desktops and servers, buy AMD; for laptops, buy Transmeta. To be fair, the Pentium M is pretty good, for performance laptops, and if you want a honking great desktop replacement, AMD and Intel are more or less equal - but your battery life will suck and your arms will stretch. But the exercise will be good for you, probably.
I'm not fond of Intel's marketing tactics - I think they play dirty - and I've been using non-Intel x86 chips from preference for nearly a decade now, from a 6x86 to a 6x86 MX to a 6x86 MII to an AMD K6-2 to a dual Athlon XP. (The K6 was recently, plus one in my ex-girlfriend's PC years ago.)
I confess that between those two pairs of AMDs, though, I did go for a PII-450, as for a while there, there just wasn't anything to compete. The PIII was a bit of a con, too, though, until the 133MHz FSB “Coppermine” models appeared. The early PIIIs were just PIIs with improved MMX and were no faster than a PII at the same clock speed.
I find these non-Intel x86 chips absolutely fine. The Cyrix ones gave me very minor problems occasionally: some motherboard BIOSes didn't recognise them correctly, which was usually just a cosmetic problem – the BIOS POST screen would be wrong. Big deal. However, I also found that PageMaker 6.5 (or 7, I forget) wouldn't run on a Cyrix machine. That was the only app I ever had a problem with until Windows Server 2003, which also wouldn't - it couldn't even boot. However, by the time WS2003 came out, the 6x86 MII was /ancient/, so it's kinda fair enough. I think VirtualPC also had some problems but the same applies and I never even tried to run it on the old Cyrix machine. (I had a PIII laptop by then. I would have liked to have gone for a Transmeta Crusoe machine, but the purchase was extremely price-critical and performance was more important to me than light weight and long battery life.)
AMDs have never given me even that much problem. They deliver measurably better performance than Intel chips at much lower clock speeds *and* they're significantly cheaper for equivalent performance. With the K7 family - the Athlons and above - you do need a different motherboard chipset and CPU socket. (K6s and 6x86s used the Pentium's Socket 7.) This does make direct swap impossible and means performance comparisons are a bit harder, but they give much better bang for the buck and the company doesn't behave as badly as Intel. 'Nuff said. I also recommend them to clients for workstations and indeed servers when I can find server manufacturers who can supply AMD-based units - these are lamentably rare, but getting better in these days of the x86-64.
And on 64 bit chips, it's worth noting that the P4s and Xeons with "IA32e" - Intel's marketing name for x86-64 - bolted on are no faster than standard ones. Indeed, the "Prescott" P4 - generally believe to be the first version with IA32e, tho' it was disabled on all the early chips - they're only enabling it now, as the P4 approaches end-of-life - was actually slightly /slower/ than the earlier Northwood P4. In comparison, AMD's Opteron is an all-new chip. (The AThlon64 is a marketing name; it's no relation to the Athlon CPU.) AMD x86-64 CPUs are considerably faster, at the same clock speed, than their 32-bit predecessors, which were already considerably faster than a P4 at the same speed.
The P4 itself was a triumph of hype and marketing over good design. It was solely designed to run at high clock speeds, because these were linked in the buying public's mind with performance - due largely to Intel marketing. Cyrix and AMD were thus forced into responding with PR ratings, a number showing the approximate MHz of an equivalent Intel chip. So my old Cyrix 6x86-166+ actually ran at 133MHz but was equivalent to a Pentium at 166MHz. My dual Athlon machine has 2000+ processors; they actually run at 1666MHz but are roughly equivalent to 2GHz P4s. This meant they could sell effectively against Intel, so Intel moved the goalposts by producing the P4, which could run at nearly twice the MHz of the PIII - although in doing so it delivered little or no extra performance. Indeed, the early PIIs at 233MHz were little or no faster than the P1 at that speed. Intel has been falling behind in CPU development for many years, and has retained its market pre-eminence by clever and extremely aggressive marketing.
So a 1.8GHz P4 actually compared badly to a PIII at slightly over 1GHz. However, that big number *looks* impressive, so people bought them. AMD couldn't compete. A new PR number showing the speed of P4 they were equivalent to might have worked, but then owners of earlier Athlons, whose numbers were scaled against the PIII, would complain. If you upgraded an Athlon 1500 (PIII scale) to an Athlon 3500 (P4 scale) and found it was only 10-20% faster, you'd be pretty pissed off.
What's changed this is the advent of the Opteron and x86-64. It’s moved the goalposts – firstly, because it performs much better on 32-bit code (unlike Intel’s x86-64 chips, which run 32-bit code at the exact same speed as ever, or its IA64 Itanium chips, which run it /substantially/ slower), and secondly, because 64-bit is a new ballgame. For best results, you need new 64-bit apps on a new 64-bit OS.
What do I mean by best results? Well, benchmarks show that for CPU-intensive applications like PDF rendering or image manipulation, 64-bit code on a 64-bit OS is some 25% faster than 32-bit code on a 32-bit OS. You don’t get those benefits if you run 32-bit code on a 64-bit OS.
It’s not so bad for AMD, as its x86-64 bit processors run 32-bit code faster than ever anyway.
But it’s bad for Intel, because its don’t. Indeed, even its non-x86 Itanium processors are slower than its latest x86 ones.
It’s also bad for MS because its 64-bit OSs /still/ aren’t fully finalized and finished and it has few to no 64-bit apps on offer yet.
So, for this new game, AMD dropped its P-ratings and has just started giving its CPUs model numbers.
Intel, aware that it’s in danger of losing the game altogether, has eaten humble pie, cloned its much smaller competitor’s instruction set, and introduced AMD-compatible processors: the P4 and Xeons with “IA32e”. And aware that their performance ratings aren’t looking that impressive any more, especially since they’ve failed to come up with a 4GHz part, and indeed given up on doing so, have also copied AMD and gone over to model numbers.
Meanwhile, the entire future development of the P4 line has been cancelled and development is focusing on the Pentium M instead. This is essentially a PIII with improved branch prediction and a P4 bus. It was designed and optimized for low-power use, i.e. notebooks, but it’s starting to move sideways into desktops. Now what they have to do is re-engineer it to handle x86-64.
Then the next big thing will be multi-core processors. These are single chips with more than one CPU core: two processors on one chip. The RISC chip makers like HP, Sun and IBM have been doing this for years. Here, again, Intel is behind the curve.
AMD is close behind the RISC vendors. Intel is some way off. They’ve demonstrated a machine, but no-one was allowed to look inside and it’s widely considered to have been faked with 2 separate chips.
AMD will be shipping these soon: early next year. Intel will probably follow at the end of the year or in 2006.
It may be that multi-core Itaniums and P4s will come first, if so, the latter will probably only be sold under the Xeon brand, as server processors.
Again, here, Intel has to go back some ways and start designing multi-core Pentium M chips. They won’t ship any time soon.
Summary: unless you like paying over the odds for an underperforming brand name, don’t buy Intel. For desktops and servers, buy AMD; for laptops, buy Transmeta. To be fair, the Pentium M is pretty good, for performance laptops, and if you want a honking great desktop replacement, AMD and Intel are more or less equal - but your battery life will suck and your arms will stretch. But the exercise will be good for you, probably.
