With Zen delayed, AMD could tap Carrizo for 2016 desktop hardware

 

 When AMD launched Carrizo earlier this year, it kept the part mobile-only. Carrizo’s entire design was tuned for mobile hardware, and AMD did a number of briefings on how the chip’s power consumption improvements and overall performance per watt were strongest in low power envelopes. It’s therefore a touch surprising to hear that AMD is reportedly planning to bring a desktop version of the chip to market.
Zen’s delay, however, may have left AMD with little choice, since it can’t afford to go an entire year without at least some new products in the market. The company is reportedly prepping a set of both mobile and desktop cores under the Bristol Ridge moniker, built for socket FP4 and AM4 respectively. FP4 would continue to support both DDR3 and DDR4, while AM4 would supposedly be DDR4-only. We know Carrizo has a DDR4 controller on-board, since the embedded version of the product launched with DDR4 support earlier this fall, so it’s not surprising to hear that the same hardware would roll over to desktop.
Here’s the supposed list of desktop SKUs:

There’s almost certainly some errors here, since it’s highly unlikely that AMD would launch a dual-core SKU with a 2.8GHz core clock at 65W. The high-end SKU, if accurate, would essentially match Kaveri’s clock range (3.6GHz – 4GHz) but offer faster DDR4 support for APUs, and a much-improved power envelope (65W vs. 95W). Unfortunately, the performance gain is still going to be thin — 5-10% clock for clock, two years after Kaveri first shipped, simply isn’t going to impress desktop users. Overclocking headroom isn’t going to be anything to write home about either; Carrizo will be stretching as it is to push into 4GHz territory and 65W — just as Kaveri’s power consumption and thermals rose more rapidly above 3.5GHz, we’d expect Carrizo to do something similar. One of Carrizo’s great strengths in mobile was the fact that the improved design let AMD ship 15W parts would more graphics cores enabled. Desktop Kaveri / Godavari never had that problem, robbing the still-hypothetical new chip of one of its competitive advantages against its predecessor.

Could Bristol Ridge be a 14nm chip?

One point raised by Tech Report is that Bristol Ridge could be an early 14nm processor. In theory, this makes some sense — it’s a bad idea to try and pull off a die shrink and a new CPU architecture simultaneously, and the Excavator uarch is well known to AMD. Looking at the range of 15W – 35W parts, however, the data suggests a relatively modest bump of 200-300MHz, and TDP envelopes within the ranges AMD has already set.

The bigger reason I think Bristol Ridge will be a 28nm chip, however, is because die shrinks are both expensive and time consuming. In order to have Bristol Ridge ready for a mid-2016 production window, AMD would’ve needed to tape it out by late summer. Keep in mind that back in May, AMD made it quite clear that Zen would arrive by mid-2016, and that Zen was the future of AMD products.
AMD essentially would have had to design Carrizo for 28nm, tape it out, and immediately begin converting the design to 14nm before it even had silicon back from the foundry. I’m not going to say that’s impossible, but it’s insanely risky and I’ve never heard of a foundry doing it.
It’s exceedingly unlikely that AMD took this road, particularly given the modest gains 14nm would deliver. The Excavator core family is played out, everyone knows it, and the company’s engineers have already performed a near-miracle just squeezing a Bulldozer-derived into a 15W form factor. Pouring money into further Excavator R&D would be throwing good funds after bad, and AMD just doesn’t have the money to waste these days.