AMDZone has some benchmarking of an Opteron/nForce3 platform.
After looking at the numbers a bit, it’s probably safe to say that this combo should be able to match a modern PIV system running 60-65% faster. This means it will get close enough to the expected PR rating for these chips for it to be at least pretty close to the PIV GHzage.
One problem, though.
This measurement is for a CPU with 1Mb cache running dual-channel DDR. The desktop Hammer you are likely to buy will have 512K cache with single-channel DDR.
That will make a difference, probably at least 5% overall, maybe as much as 15% for certain programs.
There’s some indications AMD’s PR ratings will differ for the Athlon64 Jr. chips (i.e., a PR rating of 3100+ for a 1.8GHz Opteron, a PR rating of 3200+ for a 2GHz Athlon64 Jr.)
It’s more likely than not that the 3200+ rating will be a bit fishier than the 3100+.
But that’s not AMD’s real problem.
The important measurement in the benchmarking is not how the Opteron fares against the PIV but rather against the Athlon 3200+. For all intents and purposes, anyone overclocking a TBredB is coming in at around Athlon 3200+ levels.
Is the performance difference you see enough for you to switch, or even keep you from buying an Athlon system today? Especially given what looks to be cheap but very competitive barebones nForces coming.
If these single-channel nForces pan out, one could get a Barton, a mobo, and 512Mb of decent PC3200 RAM and have change left from $300. Go for a TBredB and 256Mb RAM instead, and you could have change left from $200.
In comparison, you’ll be lucky to spend no more than $650 for an Opteron/nForce3/dual DDR platform. Yes, it will be faster, but not a whole lot faster.
Yes, the Opteron system offers the prospect of 64-bit applications and games, but how many of them are going to be around three or six or nine months from now?
Something the Opterons don’t seem to offer is overclocking. The 1.8GHz Opteron used by AMDZone isn’t able to overclock even 10%, and fails rather often even at that speed.
We’ve always had the feeling that 130nm Hammers were going to be AMD’s Willamettes, and the more details we get, the more we think we’ll be right.
We didn’t like Willamette, either, and said to wait for Northwoods. This proved to be a good idea.
We envision much the same for the Hammers. We’ve said before, and we’ll say it again, it looks like the best time to switch platforms to Hammer will be at 90nm, starting about a year from now. The chips should be faster, they should overclock more, and the second-generation motherboards available only around then will have shifted from AGP to PCI Express for video.
Of course, we were sure Intel would be around long enough to come out with Northwood. We’re not so sure about AMD.
You can say, “Well, IBM is behind AMD now.” Unfortunately, I remember the last CPU company IBM was behind, a company called Cyrix. Didn’t help them much, now did it?
Granted, AMD looks to have much more going for it than Cyrix ever did (though both did have a speed ramp problem at the time IBM got involved).
Is this FUD? Of course this is FUD. FUD has an undeservedly bad name. There’s absolutely nothing wrong with fear, uncertainty and doubt when there is an excellent reason (AMD running out of money) to have fear, uncertainty and doubt.
I would be very happy if IBM said it was going to buy AMD, or said that they’d keep them afloat one way or the other. Then I’d find their future pretty secure, and the FUD would go away.
I’d be even more upbeat if IBM also made a statement similar to what they did with Apple when they said they’d have a 3GHz G5 in a year. Say the same thing about Hammer, and I might turn into Chris Tom. 🙂
Right now, though, it looks like AMD will lose $200 million a quarter until maybe next spring. They’ll have an aging, cheapening Athlon line, and not even AMD expects many Hammer sales in 2003.
Provided AMD can get to 90nm fairly quickly in 2004, and have a normal speed ramp for those chips, 2004 could be a big turnaround year.
The big questions for the next twelve months, the ones that dwarfs everything else, is “Can they last that long by themselves? If they can’t, will someone pick them up and back them up if they can’t?”
If IBM does, fine.
What happens if IBM doesn’t?