> Yea, that sounds really great, but it is primarily due to the 4x faster BUS of the P4, not the Rambus memory. To confirm
> this, note that the Athlon handily beat the P3 in this memory bandwidth test, because its bus is 50% faster than
> the P3.
You cannot deny Barry his glee here. The difference in bandwidth here between the P4 and Athlon is much greater than the difference in bandwidth between the Athlon and PIII. And the memory plays a large (probably the main) factor in this.
What Barry is probably ignoring, though, is that these are scores of dual channel, double data rate DRDRAM. And it is being compared to single channel, single data rate SDRAM. If the comparison was against dual channel, double data rate SDRAM, then I suspect he would be much less apt to post the link on SI.
But even that is somewhat irrelevant, since we likely will not see any dual channel (slash interleaved) implementation of SDRAM in the reasonably near future. It's just a too expensive option, and AMD doesn't do expensive. So only Rambus memory will have the benefit of dual channels (to its credit, DRDRAM is reportedly a little cheaper to which to do this sort of parallelism).
But even that is totally irrelevant, since it is uncertain whether dual channel anything is acceptable to be priced for the mainstream market. Regular consumers might have to either work with (A) single channel DRDRAM, or (B) no DRDRAM at all. And both options will severely constrain DRDRAM's penetration into the market. Now, it's great if DRDRAM ends up dominating the workstation/server market, but I will declare that this may disappoint some (most?) analysts, since various Intel and Rambus folk have in the past strongly pushed that DRDRAM would take over as the mainstream memory type very rapidly.
> But you apparently didn't bother to read the lackluster results using real software benchmarks which showed the
> P4-1GHz about the same speed as an Athlon 600 MHz. I agree that Rambus has great bandwidth, but bandwidth
> doesn't affect PC performance much -- hence the poor benchmark results.
It, of course, depends on the type of benchmark. You can go to any website to see the difference in which benchmarks like big memory bandwidth. Most benchmarks didn't seem to get an effect, but in many of the benchmarks in which the PIII underperformed Athlon vastly (workstation level stuff), the gap was greatly decreased or even reversed.
I remember being surprised, though, when PC100 SDRAM on 440BX beat the crap out of PC800 DRDRAM in some select benchmarks, like various D3D games. The fact that the DRDRAM was constrained by a 1.066GB/s chipset was, of course, an important factor, but it did confirm that once you take away most of the bandwidth advantage of DRDRAM, SDRAM becomes a very competitive factor (with a 25% lower peak bandwidth, SDRAM held its own for the most part). This is why double data rate SDRAM is such an interesting possibility, since it has a higher peak bandwidth than DRDRAM (unless, of course, you're comparing dual channel DRDRAM against single channel SDRAM -- of course, that is an acceptable comparison since it involves real world existing parts, but it is also important to remember the cost effects, blah, blah, blah, yadda, yadda, yadda....)
-JC |
