When AMD launched the RX 560 GPU, as an upgrade to the older RX 460, it upgraded the specs of the card slightly. Unlike the RX 580 and RX 570, which retained their GPU cores, texture units, and ROP configurations and simply targeted significantly higher clock speeds, the RX 560 actually changed the core configuration from 896:56:16 (RX 460) to 1024:64:16. Now, AMD has apparently changed its RX 560 configuration without giving any sign they actually did so. The RX 560 is now selling in an 896-core configuration; we’re assuming it’s the same 896:56:16 configuration as the RX 460.
To be perfectly clear: We don’t approve of or condone this practice. It’s not unusual for GPUs to be sold in different memory configurations, but some companies, like VisionTek, don’t even list the core count on their RX 560s on their own product pages. A sufficiently high base clock and turbo boost could compensate for the loss of the texture mapping units and GPU cores, but that doesn’t seem to be what’s happened here.
It would be one thing if these GPUs were confined to the bottom of the market, or restricted to 2GB configurations, but they aren’t. Browsing Newegg shows plenty of Radeon RX 560s for sale in both 896 and 1,024-core configurations, including an Asus RX 560 with 896 cores for a whopping $140. (There’s an Asus RX 560 that shows on Newegg’s product page as costing $166, but when you click through it’s priced at $129).
With the above in mind, we also have to acknowledge this is far from the first time that AMD or Nvidia have sold multiple GPUs under a single SKU, with one part being much faster than the other. Several years ago, it was common to see low-end GPUs equipped with a GDDR5 and a DDR3 option, with the GDDR5 card almost invariably performing better. Nvidia’s GT 630 came in a 96:16:4 configuration and a 384:16:8 configuration. Both companies have, at various times, launched OEM or even standard editions of GPUs with varying core counts. And that’s not even considering that in some cases, the relative difference between two GPUs with a similar model number is vastly different. The Radeon HD 7790 was far, far faster than the Radeon 7770 than a simple 20-point difference in their model numbers would account for. Data below courtesy of Anandtech’s Bench comparison tool, there are additional results at this link that aren’t shown in the graph below. The 7770 is in blue, the 7790 is in red.
The moral of the story here is both AMD and Nvidia have played this kind of card before. It’s not clear why AMD did so; the RX 560 has been in-market for months, it’s widely available in both configurations, and the yields AMD gets on the card should be mature, since it’s based on the RX 460 and didn’t pick up enough clock speed to require a serious binning difference.
The situation in this case is also worse because the GTX 1050 and the RX 560 are neck and neck with each other over an entire suite of games. TechSpot did a 30-game roundupand found the GTX 1050 had just a 1fps advantage over the 1024-core version of AMD’s chip — but check out the spread between them depending on the title.
If you pull 128 cores off the RX 560, chances are that tilts things a bit more towards the GTX 1050, and some of the RX 560s are priced so high, they’re competing against the 1050 Ti (which wins that comparison decisively). Given these circumstances, we really can’t recommend anyone buy the neutered version of the RX 560. Nvidia competes effectively in this price bracket.
And while we’re sure no one’s going to listen, it’s high time GPU manufacturers stopped pulling this crap. Call it the RX 558 if you have to. Call it the RX 560 896-Core Edition. Call it the RX 560 14CU. But while the performance gap between the 896 and 1,024 core-versions isn’t huge, all the initial results on Google were done with the 1,024-core variant. Silently adding a second option might be pretty common in the grand scheme of things, but it’s not consumer-friendly.