Went with Joe to a computer show in the New York metropolitan area.
Haven’t been to one for a while, noticed a few changes. There were a lot more people there than a few months ago, actually pretty crowded. I think that isn’t a sign of economic revival but rather more people more intent on spending their money carefully.
Even more surprising was a company I had never seen or ever expected to see at such a fair: Gateway. It will be interesting to see if they keep coming, and if others join them.
Anyway, my mission was to find an AXIA chip. To do that, I go to a number of vendors, and ask to look at their tray or OEM chips.
This is often an awkward moment, because the vendor usually has no idea of what you’re doing, but suspects it’s nothing good.
I try to do this when the vendor isn’t particularly busy, and I try to do this as quickly as possible and take up as little of the vendor’s time as possible.
Got a little reluctance from a couple this time, so I just repeated once or twice. “Could I please see the chips? I need to look at them.”
Then again, if I have to repeat myself, my tone is not exactly supplicating. Probably sounds more like, “Show me the chips, or else” than “Please let me see your chips, oh, pretty, pretty please?” but not “Show me the chips or die.”
Tone is important. 🙂
I got to look at about thirty CPUs; 1Ghz or better, and came to a few quick conclusions:
A 1Ghz AXIA is probably going to be hard to find. I saw a bunch with codes I had never heard of before, like ARGA. The 1Ghz chips nowadays seem to start with the letters AQF or ARG, not the AXI we’re looking for.
I saw a ton of AVIAs. This looks like a new code from AMD. A couple people have reported them to me in the last few days, but there was a lot of them at the shows. Don’t know how well they’ll overclock yet (and if you’ve tried one out, please tell me), but my guess is a little less well than the AXIAs.
My suspicion is the AXIAs are meant for the 1300s/1333s, and show up as overflow in the lower speeds. This usually happens early in the production run.
Anyway, I did find one guy with three 1133 AXIAs, all week 8. At $209 plus tax, not quite the bargain it would have been had it been a 1Ghz, but less than a 1200. So I bought it.
How fast? Well, that depends . . . .
We brought the chip back to Joe’s house, popped in onto a water-cooled testing rig with our KK266, and I started testing.
This may sound strange, but I very much liked how the IWill board blew up.:)
With all the reports of some people doing better with lower voltages, I was testing all sorts of Mhz/voltage combinations that would have been insane a few months ago.
With the IWill, if I got pretty ridiculous, it would POST, then blow up. Then I’d just reboot and set it to something more reasonable, and it would boot up right away.
If I got really ridiculous (like trying to run over 1500Mhz at 1.6V); it wouldn’t boot, but resetting CMOS fixed things up right away; something I could not say about some of the other KT boards I’ve worked with.
I don’t mind boards telling me “No,” I do mind them resenting me afterwards. 🙂
I don’t know quite yet whether or not it’s the CPU or the motherboard, but stability degraded much more slowly with this combo than in my earlier KT tests. Practically all the testing I’ve done with KT boards before have been pretty abrupt: set it to X, it works fine; set it to X + 20, it won’t even POST.
I found I could get to the Windows screen at 1500Mhz all the time. Sometimes it would blow up at that point, sometimes it wouldn’t.
I did manage to get a SiSoft Sandra CPU Benchmark reading at 1500Mhz with little difficulty, but neither 3DMark2000 nor Prime95 were stable at that speed.
So I started toning it down a tiny bit, using 3DMark2000 as the lead benchmark, since most people seem to report that blows up before Prime95. (Has that been your experience? If you’ve used the two to determine stability, tell me whether that’s been your experience or not.)
It seemed to handle that OK at speeds up to 1480Mhz. Was getting just, just shy of 10,000 at 800X600, 16-bit, with a GTS card.
I went back to Prime95 as a doublecheck, thinking that if it passed 3DMark2000, Prime95 wouldn’t be such a big deal.
Nope. Blew up, and blew up fast at 1480Mhz. Blew up just about as fast at 1450Mhz.
Finally got it running OK at 1430Mhz.
Then I tried some more esoteric, expensive diagnostics. That was rather interesting. I tried running these at 1500Mhz. I can’t say these diagnostics had much of a pure CPU test; it blew through those right away.
Where the diagnostics earned their money was in a more subtle test of something else, memory. Since I was running a CPU test, not an FSB or memory test, the memory settings were kept well below what we knew the RAM was capable of to eliminate that as a possible cause of failure.
These diagnostics run extensive tests in a memory section: whether it be the CPU’s cache memory or regular RAM. It showed errors for cache memory, and an occasional error in base RAM.
I think they’re both CPU errors, even the one testing the regular RAM. Why? We’ve tested the RAM before, and it passed these tests with flying colors at higher speeds. The memory test requires the CPU to tell the RAM to run certain patterns. Mess up sending the patterns, and you get a memory error, even though the memory only did what you told it to do.
Things were getting a little warm in there. The system was equipped with what Joe called a “cute” water-cooling device. He’ll describe it himself in a future review, but if the average water-cooling rig is a saloon, this was TGIFridays. 🙂
I’ll see how it does with a more “manly” water-cooling device, but I think the “cute” device did pretty well. Maximum temperatures were hitting 44C, or about 110F, which isn’t too bad for running close to 1500Mhz.
How Can We Make It Faster?
I suspect a wee bit more voltage might get this stable at 1500Mhz.
What I suspect even more (and there’s some database entries to confirm this) is that some Peltier cooling would let this CPU shred 1500Mhz. I’m seeing 1600-1700Mhz reports here and there from folks doing that, and from what I’ve seen so far, I can believe it.
The next step for this processor is to put it in the A7V133, and see how it does. BTW, this board continues its pattern of perfection marred by sporadic disasters. Had a spontaneous reboot a couple days ago while underclocking it (in Win98), but after that, two days of flawless performance.
Sometime later, we’ll probably give one of these AXIAs the full Peltier test, and see how it does.