Has anyone benched one of these energy efficient CPUs? Actually, I am interested in see the benchmarks and the OCabilitytogether. They actually built the fab over 6 years ago, the fab opened in 2001, they did a good job in the last 1/2 of 2005 managing the output.
From personal experience, if a chip runs undervolted and doesn't make execution errors, it performs as well as a full voltage version.
For example: my bro' and I both had a Celeron 300A (yes, the Intel legend); we both ran them at 450, and while mine operaed well at 1.9V (instead of nominal 2V), my brother's operated at 1.7V - meaning 50% faster, with 15% less volts than nominal.
Both benched equally well; we had the same motherboards (Abit BH6 rev 1.01). So it's not too far fetched to think that 35W X2 should run as well as 65W ones.
Thanks for pointing out the actual start of the 90nm production; I wasn't sure, so I played it safe. If I remember well, Thunderbird was first manufactured using 130nm... Then a switch to 90nm (integrating L2 cache on chip) that lasted until today.
Thunderbird was not exactly a winner for AMD, they lost the most money in the year Thunderbird peaked (over 1 billion to be exact), granted this was during the steepest part of the downturn, as well as a albatrose of a buisness called flash, but still -- it wasn't the star performer that K8 has been and continues to be.
AMD had troubles the first year mainly because they could push out the CPU, but no-one wanted to develop a chipset, or develop a mobo: remember, amongst the first K7 mobos were unbranded Asus ones. Only when Via (which had won a case against Intel) developed the KT133/A and some mobo makers started using it, did K7 flood the market. And then, all benchies were won by AMD, with a 20% performance lead clock for clock (over P3). AMD's core design was a winner; their initial push to market was truly a burst, though.
K8 added 64-bit, 2 stages to the pipelines, and an integrated memory controller; but the overall core design, including the crossbar system, was already part of the K7.
They get plenty of help from IBM, over 1/3 their fab is funded by german taxpayers and 1/4 will ultimately be funded by New York taxpayers. Though I understand your point, AMD has done a lot with fewer resources.
Stop me if I'm wrong, but Intel does have a few fabs in tax-free areas, like in Mexico - I don't think the German government would finance AMD directly, but exonerate them from taxes and lend them interest-free money, that is highly possible. But I think Intel manages to get a slice of the cake too - it just looks like a drop in a barrel compared with a drop in a small glass...
Recall, K8 was late -- late by means of it was not delivered when AMD said it would be delivered.
True - Intel has a bad habit of botching its top-notch products.
- the 286 memory addressing bug
- the Pentium 60 and its FPU error
- the Pentium 200 and its sub-par performance
- the Pentium Pro and its lack of availability and lackluster 16-bit performance
- Itanium
- the original Celeron 300 (no L2 cache)
- the 430TX and its inability to cache more than 64 Mb of RAM
- the Memory Translation Hub, when they didn't want to deliver a native SDRAM chipset to replace the venerable 440BX
- the 1.13GHz P-3
- Netburst.
AMD had... well...
- K5, which was merely a tweaked 486 (it cost much less, though)
- K6 and its fantastic 16/32-bit integer performance but lackluster FPU/MMX (it sucked, but was still better than Cyrix's abysmal performance)
- reluctance to provide chipsets: 750 and 760, while performing well, were hard to come by
- the integrated memory controller in K8 requires you to change your CPU if you change your RAM type (okay, I'm pushing it)
Still, it is true that AMD is usually late to deliver, but they do deliver - little to no product recall. From what I could get, there only was one case of AMD product recall: a K8 revision could, under high stress caused by an artificial infinite loop, lead to execution failure - meaning it wouldn't ever happen if you backed up your entire DVD collection using insane encoding parameters in DivX.
Compared to Intel's 1.13GHz P3 unable to compile a Linux kernel, that's not much.