The Inquirer reports that an Athlon 64 2800+ will be showing up shortly.
While details on the chip weren’t available, it’s probably pretty safe to assume that it’s a 1.8GHz processor with 512K cache, and it will cost about $175.
The Good News
If you need to put together a Joe Sixpack system for friends and family, with little/no overclocking, and the budget is such that the alternative is a PIV 2.8C Northwood, a system built around this chip isn’t a bad idea at all.
If you’re hurting now, and you just can’t wait for Intel and AMD to get their act together sometime in 2005, again, this is a good option to consider.
If the notion of replacing the system with something cutting-edge once Intel and/or AMD get their act together sometime in 2005 fazes you not, one more time, take a look.
However, for most people reading this, there’s one big bottleneck.
To Lock By Leaving Unlocked
Current Hammer chipsets do not have a PCI/AGP lock. This effectively limits the degree to which you can overclock any Hammer.
If you don’t know or understand why this is so (and if you’ve only been doing this a couple years or less, you may well not), you need to get up to speed by reading this now.
While the lack of a PCI/AGP lock hasn’t been too big a deal so far on Hammer systems since the CPU has been the bottleneck, it will become the principal bottleneck with any Hammer system running with less than a 10X multiplier.
Let me show you why:
If you have an Athlon 3000+ running at 10X200, if you tried running it at 10X240, or 2400MHz (which is probably a bit much for current A64s, though likely not for the next A64 stepping coming soon, you’d have an FSB speed of 240MHz, and a PCI speed of 40MHz. While some equipment can’t handle that, there’s a decent chance most will.
However, if you have an Athlon 2800+ running at 9X200, if you tried running it at 2400MHz, you’d have to set a speed of 9X267. This would give you a PCI speed of 44.5MHz, and an AGP speed of 89MHz. Most equipment will fail to function at this speed. Not all, just most.
BTW, if you didn’t know, you can’t increase the multiplier on an A64 above the default.
What sort of equipment would be suspectible to going down? Video cards, hard drives (which sometimes scramble themselves when pushed like this), any PCI devices (or devices built-into the mobo which operate on the PCI clock, like Ethernet connections).
For the vast majority of people trying to overclock a 2800+, PCI speed will be the bottleneck preventing them from running their CPU to its maximum potential.
What To Do?
We know that current Hammer chipsets do not have a PCI/AGP lock. What we don’t know is whether or not future Hammer chipsets will have one.
The next potential candidate is the nForce3-250 mobo. More will come throughout the year.
For those who want to crank up a 2800+ as high as it will go, a PCI/AGP lock is a required feature in any mobo. Your attitude should be “No feature, no purchase.” Be careful, though, some manufacturers have claimed having the feature, and it just wasn’t there. Don’t trust, verify.
We’ve spoken often in the past about socket 754 being doomed to the bargain basement next fall.
However, since AMD doesn’t seem capable of sticking to any plan, apparently doesn’t want to make the equipment we really want to buy (socket 939) any time soon, and thinks it can herd all you sheep into buying socket 754, it’s impossible to predict what the dazed and confused will end up finally doing.
How much will CT technology affect AMD’s plan? Who knows? AMD plans on putting out a cheapy Hammer without x86-64 (Paris). Will they do so knowing that Intel can put out x86-64 Celerons in the blink of an eye? Again, who knows?
Will AMD tell all the people who buy socket 754 now six or nine months later “That was then, this is now” when pimping socket 939 boards? Or will they make faster socket 754 chips?
Do you feel like betting hundreds of dollars on the right decision from a company that probably has no idea what it’s going to end up doing?
Intel doesn’t work this way. You generally know what Intel’s going to do a year in advance (hint, hint, corporate clients like that) so you can figure out when the best time is to buy, and they more or less do what they say.
Why can’t AMD do that?
Here’s what I suspect is really happening. If you recall, there wasn’t supposed to be any dual-channel desktop systems originally. Only Opterons were supposed to have dual-channel memory. AMD changed its mind at the last moment on that and made the Opteron That Wasn’t An Opteron (aka FX). My suspicion is that the AMD execs are still fighting over that decision because the ones that backed single-channel only don’t want to be proven wrong, and they’re stalling and delaying and overpricing socket 939 just to proven “right.” Do I have a shred of concrete proof? No, but that’s the most plausible explanation for what we’ve seen so far, and it’s the kind of infighting you often see at the top.
Regardless of the “why,” I don’t think you should reward this kind of behavior. Why buy something when there’s a good chance you’ll be left hanging in six months?
Mind you, all AMD has to do is say “We’re going to continue to make leading edge CPUs for socket 754.” Not this “we won’t, but buy it anyway” or “we won’t until we change our minds.”
If AMD says “We can’t tell you,” tell them, “Then I can’t buy it. Keep your secrets and your products to yourself.”
If every person considering or tempted by an A64 system would send AMD an email to that effect, maybe, just maybe, AMD will get the message that they’re accountable for what they say to their customers, and if they refuse to be, you’ll refuse to be their customer.
In reality, that’s effectively what most of you are doing anyway, but let them know that, too.
You have the power. Not them. Let them know it.