First let me say that I'm a Rambus and Intel long, and I've never been a fan of AMD. But I'm not here about either AMD or Intel.
I'd also like to add, very quickly, that the personnel announcement struck me as "really strange". I wonder what happened ? But that's not why I'm here either.
I'm here to talk about Rambus, and not about the current legal turmoil, but about the technology.
I know that many of you think that "Rambust" is a big POS, and some of you wouldn't be that nice.
Many of you simply don't believe that RDRAM (Rambus memory) offers any benefits, and that is what I AM here about.
I actually agree with the position stated in the previous paragraph ..... with respect to current processors (Intel and AMD both).
However, for what it's worth, the first benchmarks of the Pentium 4 with Rambus memory and the Intel 850 have been released. The 850 (Tehama) is the first FULLY optimized, synchronous RDRAM controller. The benchmarks were released by sources that are anti-Intel and anti-Rambus, although they are not from a production system and will await independent verification, which does NOT yet exist. Some of you will want to ignore them, so be it, I'm only presenting this for what it's worth, you can take it or leave it (at your own peril, perhaps). The problem with ALL previous benchmarks is that NO system existed that could "push" RDRAM (Rambus) memory. The Pentium 4 and it's Tehama 850 chipset is the first chip, ever, that is capable of using the full memory bandwidth of RDRAM memory.
Here are two quotes regarding these benchmarks:
****** First quote:
"The most startling numbers are what can only be described as phenomenal memory performance from the Rambus
memory in the test system. SiSoft Sandra memory benchmarks show the P4 delivering an ALU memory bandwidth of 1407 Mb/sec, and an FPU figure of 1520 Mb/sec. By comparison a 1GHz PIII looks very ordinary with 325 and 345Mb/sec respectively.
We'd advise a degree of caution here as this is only one benchmark, but if Rambus can really deliver this kind of performance, maybe we should start liking it a bit more."
******* Second Quote from a different source:
We all nearly fell out of our collective chairs when we saw the results showing the VAST memory bandwidth that the Pentium 4 + Rambus solution provides. This is truly a remarkable number. There has been a lot of talk lately about Memory bandwidth. It's common knowledge that SDRAM just isn't going to be able to keep up with the ever-escalating clock speeds and system bus speeds that have been coming out of both the Intel and AMD camps. Rambus has really been in question over the last several months. Its high cost and questionable performance have really prevented
it from being considered a viable option for the average user. I think this benchmark shows that in accordance with the 400 MHz system bus that Willy provides, Rambus is capable of pushing some incredible numbers."
*******
To summarize the numbers, a 1 GHz Athlon was capable of moving 435 to 520 million bytes per second, the Willamette (Pentium IV) SLOWED DOWN TO ONLY 1 GHz was moving 1,410 to 1,510 million bytes per second. Rambus was really delivering the goods on this test. They were running at TRIPLE to QUADRUPLE the speed of what are, at the moment, the fastest processors that money can buy.
Please note that this was ONLY a memory bandwidth test and it does NOT imply that Willamette will, overall, perform that much faster than current processors. But it says VOLUMES about what Rambus memory is CAPABLE of doing. And while it's not here now, and may not be here even later this year when Willamette begins shipping, there is no question that as things get faster and faster and faster, the time will come when this performance is not only desirable but is desperately needed to keep up with ever faster processors and system bus speeds.
You can see the entire articles here:
2cpu.com
theregister.co.uk
So why am I here ?
It's not to bash AMD or pump Intel.
Rather, it's because many of you have become fanatically anti-rambus with no really good information on which to base such feelings other than that Rambus was allied with Intel, and Intel is "the enemy".
Well, Rambus is not Intel, and Intel's CPU chips are not the only ones getting faster. At some point (REAL soon) AMD's processors are going to outrun the memory bandwidth of SDRAM. Even if DDR EVER ships (I have my doubts, but let's not get into that just now), at some later point, AMD CPUs will outrun it also.
What I suggest to you is that the conclusion that RDRAM (Rambus memory) offers no real benefit was right --- in the context of sub 1 GHz systems with asynchronous memory controllers not optimized for RDRAM (including the Intel 820 and 840). However, I am further suggesting that in a absolute sense -- with processor speeds expected to reach 4-6 GHz by the middle of the decade (that's for real, and presumably Moore's law applies to BOTH Intel and AMD) -- the "first brush" conclusion that many of you have jumped to regarding Rambus memory may need to be re-examined.
I'd only like to suggest that you consider that possibility and keep an open mind towards Rambus memory in the future. |
