I acquired an Abit AT8 32X many months ago. It was on sale, and I simply couldn't resist the price. For a brand-new flagship ATI-chipset motherboard, how could you say no to less than $100.00?
The AT8 32X sat in my closet for a long time, primarily because I didn't have enough parts (nor the inclination to spend money to get these parts) to do a proper build. I had a CPU and a cooler, but I needed a video card, 2GB of RAM, a PSU, a case, a HDD, an optical drive, and the wherewithal to mix all these ingredients together and hopefully come up with something tasty.
I patiently waited for good prices on desirable parts, picking up a piece here and a component there, until I'd assembled a decent collection of kit. Now I concede that this rig is no longer "top-of-the-line," but that was never the point of the exercise. It was a new kind of adventure.
After all, it was my very first taste of an ATI chipset.
I was keen to explore just how differently ATIs overclock compared to the nVidia chipsets I'd been used to, or even the odd VIA chipset.
As any overclocker worth his salt should, I spent plenty of time reading up as much as possible on the ATI Radeon 3200 XPress Chipset. It was ATI's equivalent to nVidia's superb nForce SLI X16, so I had high hopes for it.
So here is the composition of the new ATI rig, which I'd dubbed "Darth Scourge" before starting out on the build:
CPU: Opteron 170 LCB9E 0704TPMW (later swapped for an Opteron 170 LCBQE 0722RPAW)
CPU Cooler: Coolermaster GeminII (with two Logisys 120mm x 25mm red LED fans)
Motheboard: Abit AT8 32X
RAM: 2x 1GB G.Skill F1-3200PHU2-2GBNS
HDD: Seagate SATA3.0 320GB
Optical Drive: LiteOn DVDR/W
Video: Sapphire Radeon X1950XT (256MB)
PSU: Thermaltake ToughPower 700W
Case: Enermax Uber Chakra
After building the rig, I ran the rudimentary post-build tests (the "smoke test," MemTest86+ for the RAM @ stock speeds, fan controller functionality tests, etc.) and dove into the BIOS to familiarize myself with it.
I was a little disappointed in the BIOS, to be honest. I wasn't expecting the mind-numbing complexity of a DFI's BIOS, but I thought that there would be more facility for settings adjustment than was actually present.
Anyway, I proceeded on my overclocking journey after I was satisfied the motherboard and all the other components were sufficiently healthy. As always I broke up the overclock into separate phases: First was a Max HTT Frequency test (to see where the motherboard's limits were), then the Max CPU Frequency test (to see how high a CPU can go), and then finally a max RAM Frequency test. After maximum values for these subsystems are found, the overclock is "finalized" with consolidated settings from these three initial phases.
I was a little disappointed that the HTT clock wouldn't budge past 305MHz. Now, with an Opteron's 10x CPU multiplier, that's enough to hit the magic 3.0GHz frequency, but the HTT clock max was low for a top-flight motherboard.
The CPU Max test gave its own share of disappointments and, frankly, problems. First of all, I was somewhat shocked that LCB9E 0704TPMW wouldn't scale past 2.6ishGHz without a struggle. I thought maybe I just had a low-ceiling CPU, so I swapped it out with an LCBQE 0722RPAW. LCBQE struggled to get to 2.6GHz, so I started thinking that perhaps it was the board/chipset that was holding the system back from scaling higher than 2.6GHz.
I played with some voltages; increasing VCore did not gain more than a few MHz in either CPU's case, which really confirmed to me that it wasn't the CPU being stubborn: It was the motherboard/chipset. I played with chipset voltages then, and finally the CPU clock continued upwards. The final max stable for LCBQE 0722RPAW was 2.9GHz.
The biggest source of frustration, though, was RAM overclocking. First of all, this motherboard/chipset (I still don't know if the problem is with the physical board or the chipset's abilities) just simply did not like certain RAM dividers. Would you believe it failed to POST on the lowest RAM divider? (DDR200) I've NEVER ever seen that happen on any motherboard.
I've also never seen a motherboard fail at increasing RAM OCing headroom by lowering CAS Latency. The AT8 32X failed POST when I changed the CAS Latency to 3, from stock 2.5. And lowering Command Rate to 2T also did not free up MHz as it almost always does.
My final RAM max overclock was therefore a rather paltry and disappointing DDR460 (230MHz @ 2.5-4-4-8-1T).
The rig is presently being tested through a second Orthos Blend run. The first Blend run went for 24Hrs on Priority 7, so I know it's stable. CPU Speed was at 2.75GHz, but the RAM was running at a measly DDR376. That's slower than stock! I tried running the RAM @ DDR456, but it couldn't run Orthos Blend nor SuperPi 32M without failing. Clearly, running RAM @ DDR376 was problematic.
I decided to slow the CPU speed down to 2.72GHz, and then used a lower RAM divider so that it can run @ DDR454 (2.5-4-4-8-2T). So far the Orthos Blend test is at the 10Hr 30minute mark; if it makes it to 12Hrs, I'll be happy to call it stable enough. Only after it passes 3D stability testing (loops of 3DMark06 and 3DMark05) will I call it actually stable for my purposes.
In all honesty, the AT8 32X has been the most frustrating motherboard I've ever worked with. The chipset seems to be very very sensitive and requires lots of tweaking just to exceed what I think are rather low limits compared to results I would get from an nForce4 motherboard. Despite my confessed frustrations, though, I won't slam either the motherboard nor the chipset. I'll give both the benefit of the doubt; perhaps it was just my inexperience that prevented me from having an easier time in extracting performance.
On the other hand, the net is filled with stories more horrid than mine as far as overclocking with this AT8 32X, so...
As always, thanks for reading. I appreciate your time and attention. Comments and thoughts and discussion are always welcome here!