Speculate on Hammer scores

AN INQUIRER reader has put together available AMD roadmaps and quotes from Jerry Sanders III and come up with some guesstimate figures he thinks are feasible for the X86-64 platform.

Our reader, CD, also refers to some AMD slideware that appeared on a Czech site recently. (See Clawhammer performance projected).

Our reader says:

"Because these numbers I have are based on spec 2k, that is all I will use. That said, I will also assume that all clock speed advances translate linearly to spec score increases. The numbers I use are as follows:

Hammer 3400 = 1350 = ?MHz
Palamino 2000 = 730 = 1666MHz
TBird 1400 = 550 = 1400MHz
Zilog Z80 = .0718 = 2.13MHz

With that, we can calculate that:
TBird 1000 = 393
Palimino 1600 = 613 =1400MHz

And Palamino/TBird ratio@1400MHz = 1.15, .90 inverted (1/x)

That said, using Jerry Sanders III's quote of three times tbird 1G performance, or ~=2.70GHz Palomino performance.

Using that, there are 1950PR ratings (1300MHz*100/66) over a 1600PR rated 1.4G paly.

Adding the two, you get 3550PR, or shockingly close to the quoted 3400.

If you don't assume the growth is linear, and subtract a little, you can hit the 3400 number dead on. Eerie, eh? More fudging on the linear gains front, and a 393 rated 1G TBird is about one third of a Hammer rated at 1350 (1/3.45 actually), the numbers add up again.

So what?

If you read a bit into that Sanders quote, you can see that the Hammer has a 33 per cent gain in IPC/Power. That means, a 3400 PR in Hammer speak is the same as a 2550 Palomino.

That makes it 633MHz faster than a Palomino 1400-1600PR (2550-1600 / 100 *66 = 633), or 2033MHz. If you again make an assumption that it will have a faster 100MHz multiple bus, that puts it at 2GHz.

So what?

Nothing. That is my semi-educated guess. µ