From what I understood, AMD plays on it being number 2 to Intel, so 3200+ is supposed to equal an Intel processor at 3.2 ghz, while 3800+ is supposed to equal an Intel at 3.8 ghz. Their Opteron line I read is faster then the Intel Xeon line, and doesn't use the Intel comparison number.
That was the original plan I believe to their marketing scheme. It was an attempt to squash idiotic clockspeed comparisons across different architectures and manufacturers. This should all be thrown out the window now that Core architecture pervades the Intel CPU line. Think of AMDs naming scheme as leftover from the Netburst era.
At OP: For native dual cores, it's not fair to try to think of them as two single core CPUs in a single package. They represent a new architecture. Just check Tom's CPU charts to compare anything you're looking at. After all, performance and how you're paying for it are what really matter, right?
AMD started using PR numbers about the time the Wiliamette cores came out. At that time, an xp1500 would slaughter a willy @ 1.5 ghz. Even when Intel went to northwood A, the xp number outperformed Intel's offering @ the coresponding frequency.
What the xp numbers did equate to was a comparison to an athlon @ 1 ghz. To achieve an xp rating, a new chip had to complete a set of benchmarks. The only one I still remember is winbench. There were a few times when AMD fudged a little (the xp 2000 was more like an xp 1925), but for the most part, with the xp chips, the comparison was close.
When the A64s came out, AMD updated thier benchmarks, and picked 2800 as an arbitrary starting point. All other single core chips were designated by a comparison to the A64 2800.
I suspect that the X2 chips use the X2 3800 as the same arbitrary comparitor. Unfortunetly, it seems that AMD is bumping some chips a little.
Not that it really matters, It's good enough to know that a higher number is a better performer.
BTW, if you were to take a joesixpack's A64 3800, and replace it with an x2 3800, (only correcting for the added core) you would find that they were comparable, with the x2 having the upper hand. Dual cores are very usefull when you have a bunch of crap in the background.
AMD's K5 endeavor did not work out very well, thus they bought out an up and coming CPU company, as such, they bought their design as well as their patent portfolio and engineering talent to boot.
This group began work and designed the K6, which would lead naturally to K7 then K8 (today's Athlon 64 line of CPUs). These guys, the people that they bought, were quite smart --- CPUs, with the release of the Pentium Pro, began heavy use of what is called Instruction Level Parallelism (ILP) via a class of microprocessors commonly referred to as superscalar OOOe CPUs. In essence, these new class of CPUs (started, again with the Pentium Pro) could fetch, exectue, and retire more than one instruction per clock.
Thus opened up a new avenue for AMD.... since they busted their chops to get to 1 GHz and Intel out produced them just a week or two later, it was clear they would not win a speed race (clock speed), the bright fella's purchased from Nexus set out to increase computational power not by getting clock speed to go fast, but to simply execute more instructions at a time in one clock. This is ILP at it's best and you will often hear the acronym IPC (instructions per clock). With a high IPC, a low clocked CPU can still perform as many or more instructions in a given unit of time as a faster clocked CPU with a low IPC efficiency.
As such, though AMD's CPUs had a high IPC, the general Joe Consumer often associated speed with clock speed. AMD had to break into this mindset so they came up with a PR number. Originally it was intended to be this is the 'Pentium" clock speed equivalent computational power this processor can do. So even though the processor clocked at 1.2 GHz, the PR number would be something much higher (cannot remember exactly, but say 1.2 GHz and a PR number of 2000+).
The problem is that today, their PR number has not scaled nor matches what a processor (P4, or C2D) actually does today. Thus the PR number has lost it's true 'it is this Intel equivalent' meaning --- I find it a little deceptive, but also a little logical that AMD does not drop it.
When challenged that a 4200+ is no where near the actual speed of a 4.2 GHz P4, AMD stated it was irrelevant, that the customer appreciated the model number because it was a quick glance look to see which processor was better between to AMD processors. I.e. a 4200+ is computationally better than a 4400+. This makes sense to me.... and I wish Intel would streamline their nomenclatures/model branding scheme to be as consistent.
EDIT: Also, AMD fudges numbers around a bit --- and the scale is not always consistent. For example, jumps of 200 in the model number say 3800+ to 4000+ or 4200+ to 4400+ would occur if a) there was a 200 Mhz difference in clock speed OR if there was an extra 512 KB of L2 Cache. Since AMD dropped the large L2 cache, and has recently gone to 1/2 multipliers, it will be a littel confusing to see that a 100 MHz increase in clock equates to a 200+ increase in model number. ...
In terms of the dual core vs single core, you have asked this question before --- it is true, AMD up' model's their dual core at the same equivalanet clock and cache from the single core.... go figure, as the single threaded performance will be no different than the lower single core model number.... do you feel cheated? :wink:
The X2 3800+ is clocked at 2.0 GHz. The single core 3800+ is clocked at 2.4GHz (a X2 3800+ is in fact 2 single-core 3200+ glued together).
They imply that for everyday use, a dual core is 15-20% faster than a same-frequency single core (they could have claimed it was twice as fast, easy)
Which is, to be fair, quite true: you won't notice the added power in non-threadsafe, 100% CPU loading apps with real time priority, but removing the 15% CPU load created by the OS, the sound codec mixer, the antivirus and other pest control softwares... will lead to your heavy app running 15% faster on a same clock dual core chip (it monopolizes one core, the crap runs on the other).
So, do you really feel cheated? I find my 2 GHz dual core system to be faster than the 2GHz single core one, but then I do some multitasking.
Of course the real power of dual core is the ability to pack both cores to the gills with computing demands and see it run 2 softwares at full speed concurrently.
The first PR rating was used by Cyrix in the 150/166+ time (it reflected a Pentium's integer unit capabilities). AMD tentatively used it with some K5/K6, but soon dropped it due to its lack of meaning. After that they only used actual frequency numbers, until the Athlon XP came out (it was a heavily refactored K7), which was actually based on the Athlon 1000 relative speed (which happened to rank very closely with Intel's P3 1 GHz).
The XP's extra speed boost came from an increase of FSB (100/133 to 166/200 MHz), deeper pipelines, refactored branch prediction unit and better/bigger L2 cache, which resulted on a 20% speed boost over an equivalent Thunderbird-class K7.
As far as I know, current PR rating is based on the maximum theoretical megaflops the CPU can do. It is a combination of max memory transfert speed, L2 cache size and CPU core frequency.