Overclocking Legends: The Athlon XP 1700+ DLT3C

Alongside the vendor series, I’m starting a second one on the hardware that made overclocking what it was. These are the parts everyone wanted, the ones with forum threads hundreds of pages long, the ones you’d still recognize from a few letters printed on the chip. I’m starting with mine.

How I got here

My first computer was an IBM Aptiva with an AMD K6-2 at 300MHz. There was no overclocking it. There was no BIOS option for anything interesting, and I’m not sure I’d have known what to do with one anyway. It ran Windows 98 and it did its job, slowly.

My first real build was an Athlon XP 1700+ on an ECS K7S5A. That board was no legend, but it’s what I had. It had a black PCB when everything else was green, the SiS 735 chipset, and both SDRAM and DDR slots, so you could bring your old PC133 sticks along and upgrade to DDR later. PC Perspective gave it 9 out of 10 and called it “an absolutely perfect budget motherboard” for about $75. The one thing it couldn’t do well was the one thing we wanted: the front-side bus options were limited. Even our own reviewer remembered it fondly years later as “the infamous K7S5A.”

So that’s where I started. It didn’t take long to learn about overclocking, and it didn’t take much longer to learn that the chip matters, but the board decides how far you get. Which is how I ended up with an Abit NF7.

A rating, not a speed

First, a quick refresher for anyone who wasn’t there. AMD’s Athlon XP didn’t sell on megahertz. A “1700+” ran at 1.47GHz, and the number was AMD’s way of saying it kept up with a 1.7GHz Pentium 4. Intel had the clock speed marketing war won, so AMD changed the scoreboard.

The original 1700+ in 2001 was a Palomino. In 2002 AMD moved to the Thoroughbred core, and the second revision, the Thoroughbred B, was the one that clocked. AMD used it for its top 2400+ and 2600+ chips, but it kept selling the cheap 1700+ too, and by 2003 a lot of those 1700+ chips were the same Thoroughbred B silicon, just sold as the slowest model on the shelf.

Reading the box

This is where it became a treasure hunt. You couldn’t see the core on the shelf, but you could read the markings on the chip, and sometimes through the box window.

• DLT3C: the end of the part number, AXDA1700DLT3C. The “L” meant a 1.50 volt core instead of the usual 1.60 or 1.65.
• JIUHB: the stepping code. JIUHB meant Thoroughbred B.

A 1700+ marked DLT3C and JIUHB was the jackpot: the good core, at low voltage, at the bottom of the price list. Run it at 2.2GHz or higher with a voltage bump and decent cooling, and you had something close to a 2700+ or better. The forums filled up with people posting their week codes and their best clocks, and the record attempts went well beyond that: one Russian site reported a Thoroughbred B 1700+ at 2,710MHz at 2.15 volts.

Not every chip was a winner. The same site tested a batch of DLT3C chips that turned out to be the older stepping, and none of them reached 2GHz. That was half the fun. You bought a chip, you read the code, and you hoped. Mine wasn’t a JIUHB, for the record. I always wanted one.

The board that made it sing

If the 1700+ was the engine, the nForce2 board was the transmission. The classic pairing was a JIUHB 1700+ on an Abit NF7-S, and the NF7 family could push the front-side bus to 200MHz and beyond with the right BIOS. That’s the board I ended up on. Our own Command Per Clock guide used an Abit NF7 v2.0 as the test board, and called it “quite the overclocker with the right tweaks in the BIOS.” Which BIOS you flashed was a whole religion of its own. There’s more on Abit in the vendor series.

Where it is now

AMD followed up with the Barton core in 2003 and then moved on to the Athlon 64, and Socket A wound down soon after. The “+” rating stuck around for years, though, all the way through the Athlon 64 and its X2 successors.

You can still find these chips cheap on the used market, and retro builders still go hunting for JIUHB codes. And when it was time to move on to the Athlon 64, our own Athlon 64 Overclocking Guide picked up right where Socket A left off.

Next up: Winbond BH-5, the memory chip that ate voltage for breakfast.

— Luke