Still Too Few Horses For Hammer

This website (at least for the moment, save the benchmarks if you ever want to look at them later) purportedly has a few benchmarks of an Athlon64 3400+.

The key numbers to look at are the Sandra CPU benchmark numbers:

Dhrystone ALU


Whetstone FPU/iSSE2


Unfortunately, this is what an overclocked PIV running at 3.4GHz can do today:


Big difference, isn’t there?

Big Engine, Small Engine

Of course, this doesn’t tell the whole story. From other application benchmarking, the Hammer architecture can certainly meet and beat PIVs running 50% and at least can keep it close with a 60% handicap.

The on-board memory controller on Hammer is probably the biggest performance booster for a CPU we’ve seen in years, more than hyperthreading, a lot more than DDR or dual DDR. We feel tempted to bend our rules and give it product of the year six months ahead of time. Unfortunately, it’s laggard engine makes the whole Hammer a shoo-in Most Brilliant Kludge of the Year.

If it had feelings, the Hammer memory controller would feel like the rest of the CPU is an anchor it has to drag along when it races against a PIV. It’s amazing that it can stay close, but close only counts in horseshoes.

It’s just not a dead heat, but AMD apparently is going to claim that it is, even though it will usually finish 5-10 yards behind the PIV in a hundred-yard dash.

Of course, there’s x86-64, but that looks more like the next 3DNow at the moment. I don’t see a flood of commitments by application and game companies to x86-64 porting, do you?

Spare me talk of Linux. When you’re talking desktops and Joe Sixpack, Windows and Windows apps rules. Talk of x86-64/Linux conquest of the desktop world is just a act of mutual delusion.

AMD must do better here. They can’t keep spotting Intel 70% on clock cycles in the long run. Given the performance boost it gives Hammer, don’t you think Intel might well incorporate the same thing in their next major redesign (Nehalem)? Then what does AMD do?

A Positive Suggestion for AMD

AMD is going to have PR nightmares over PR. They’re going to be sorely tempted due to financial reasons to boost PR a bit more than they should.

Right now, it looks like their PR ratings will be a little too high for their claims in 32-bit mode, but a little too low for likely performance in 64-bit mode. This is likely to get AMD the worst of both worlds.

They’ll get blasted for BSing when the chip first comes out, then they’ll get blasted again if they ever revise their PR numbers to reflect x86-64 performance.

I have a much better idea.

AMD ought to revamp its PR system to:

  • Be a little more conservative on the 32-bit end and
  • Also indicate equivalent performance in 64-bit mode.

    So instead of calling this 2GHz Athlon64 a 3400+, call it something like a 3237.

    The “32” would stand for a conservative PR estimate of 3200+ in 32-bit mode; the “37” would stand for a conservative 3700+ performance in 64-bit mode.

    Marketeers could really sink their teeth into that one. A CPU that can get faster as it gets older! That’s a better pitch for a premium price than a higher single PR number the media is likely to deride.

    This could also boost sales of x86-64 software as in, “Gee, if I buy this, my machine will go faster.”

    OK, it isn’t a strategy Jesus would come up with, but it’s better, clearer, and more honest than what we’re likely to get, and probably will work better for AMD, too.


  • Be the first to comment

    Leave a Reply