Now, Per, what part of the clear claim, "Dhrystone 2.1 benchmark program runs exactly 8.1 times faster than the original 80C51 at the same frequency," is unclear? It's not unreasonable, and it's not unrealistic. It's just not particularly impressive, in light of the performance of one-clockers.

As it happens, however, "one-clockers" operate at nominally 12 times as fast as the original 80C51. While 8.1 times the performance is respectable, it's at much (orders of magnitude) greater cost than simply buying a commercially available one-clocker, so one has to have other justification.

As everyone interested in these matters knows, these claims are somewhat "soft" in that, due to slightly different instruction timing from core to core.

My colleagues and I have "examined" a number of soft-cores, and found that few of them come with totally realistic claims.

These limitations arise from the methods used to increase the performance of the core in test environments. Most common is the practice of using a very large, very costly FPGA as a trial environment, allowing the entire ram and rom content to reside on-chip, which is a very costly way to provide ram and rom, costing perhaps only one order of magnitude as much as using external resources but which generally improves performance. Another is the consumption a considerable portion of the FPGA resources to implement the core in the most performance-efficient way possible, making it difficult to utilize the remaining resources effectively. One wouldn't consider a soft-core in FPGA if one could purchase an adequate MCU with the peripheral features one requires on-chip. Careful hand-routing is a given, of course.

My gripe with the "marketing literature" provided is that the claims are incomplete. The vendor claims

SPARTAN-3E-5 using 1029 slices without debug support or 1172 slices with it, and yielding 75 MHz performance.

They don't tell us which size Spartan-3E device they use to arrive at this figure. Further, the 75 MHz performance it offers doesn't compete well with the various one-clockers on the market, though I'd guess it results in a cost an order of magnitude greater than the commercially available MCU's. Now, keep in mind that it isn't THIS core that provides the claim of 300 MHz performance. It seems to top out at about 200 MHz. If you divide that 200-210 MHz by 12, it suggests one would attain nominally similar performance with a one-clocker at about 17.5 MHz. I was giving THIS core, whose vendor made no such claim, credit for the 300 MHz performance claimed by the other core-vendor. It's simple arithmetic, and a gross oversimplification, but I'd guess even you, having a different viewpoint, see what's indicated.

RE