Slashdot Mirror
AMD Debuts Ryzen 5 2500X and Ryzen 3 2300X For Prebuilt PCs (techreport.com)
AMD announced two new second-generation Ryzen CPUs this morning. From a report: The Ryzen 5 2500X and Ryzen 3 2300X bring Precision Boost 2 and XFR 2 to quad-core Ryzens without integrated graphics, but there's a catch: these chips appear to be available exclusively to system integrators and OEMs for use in prebuilt systems. AMD is debuting the Ryzen 5 2500X in cooperation with Acer in the form of the Nitro 50 desktop PC. AMD says the Ryzen 5 2500X and Ryzen 3 2300X each use a single enabled core complex (or CCX) from the two available on Pinnacle Ridge Zeppelin dies to get their four cores. Recall that the Ryzen 5 1500X instead used two cores from each CCX to get its core count. A consequence of this architectural change versus the Ryzen 5 1500X is that the Ryzen 5 2500X now has 8 MB of L3 cache, down from 16 MB. That puts both the Ryzen 5 2500X and Ryzen 3 2300X on par with the Ryzen 3 1300X and Ryzen 3 1200 on a cache-capacity basis.
The entire product line is one chip design. They burn out fuses in the chips to disable features to make the lower end ones.
When you make chips, you produce a big wafer that gets cut into individual chips. There will be defects in the wafer. You can generally predict roughly how many defects there will be, but not where they will be. So you design your chips so that portions of them can be disabled.
You get a perfect chip with no flaws? That gets sold as the top of the line part. You get a chip with defective cores? That one has cores disabled and gets sold as a lower end chip. Maybe the cores are fine, but there's flaws in the L2 cache? Disable part of the cache and put it in a lower end bin. Or maybe everything works, but the chip starts to get unreliable if you clock it too fast. Again, into a lower end bin.
Chips get designed this way because it means you can still sell most of the chips that have flaws in them. You'll still get some unusable chips, but most flaws can be worked around to produce a product you can sell.
If the market wants a lot of low end chips, sometimes you end up disabling working features to meet demand. It's way, way cheaper to do that than it is to set up a separate manufacturing line just for the lower end chips. This is more likely to happen later in the lifecycle of a particular chip, after manufacturing has been running for a while and the flaws have been worked out. It's usually not worth setting up another manufacturing line to make lower end chips when this happens.