This site may earn chapter commissions from the links on this page. Terms of utilise.

Ever since AMD announced Carrizo, we've been cautiously optimistic about the bit's performance and improvements over Kaveri. No i expected miracles from AMD's sixth-generation APU, just the new bit included additional CPU improvements, significantly ameliorate video playback, and it was AMD's commencement fully-integrated SoC. There was a chance that Carrizo could win AMD some respect in the low-power 15W markets where the flake was meant to play.

An absolutely enormous review from Anandtech dashes those hopes. First off, when I say "enormous," I hateful it — the v-way comparison between four Carrizo systems (built past HP, Toshiba, and Lenovo) and one Kaveri organisation totals over 23,000 words. We'll hit some of the loftier points here, only if you're curious to see the subtleties, we highly recommend reading the original piece.

Carrizo vs. Kaveri

Anandtech first compares Carrizo against Kaveri before taking it out against Cadre M. The good news in this match-upwards is that Carrizo oft delivers the improved performance and reduced power consumption AMD promised. In a handful of cases, the gains are considerable — check Cinebench xv'southward single-and-multi-threaded performance graphs below for an example of how much Carrizo can ameliorate over Kaveri in a all-time-example scenario.

CB15MT

Carrizo vs. Kaveri in Cinebench. This is a best-case scenario for the new silicon. Images past Anandtech.

CB15ST

Carrizo vs. Kaveri in Cinebench. This is a best-case scenario for the new silicon. Images by Anandtech.

Even the slowest Carrizo APU is 11% faster than Kaveri in unmarried-threaded style and 15% faster in multi-threaded mode.

Unfortunately, all of the Carrizo systems tested share a common handicap — they all ship in unmarried-channel retentivity configurations. The HP Carrizo systems tin all be upgraded to dual channel performance, at least, but the Toshiba and the supposedly loftier-terminate Lenovo rigs are both stuck in unmarried-channel fashion.

The reason for this is toll. When AMD launched Carrizo and Carrizo-Fifty (refreshed Beema), it talked about how using a unified platform for both chips would allow OEMs to save coin on design costs. Unfortunately, Carrizo-50 is a single-channel pattern. Instead of building motherboards that could handle two channels and simply disabling ane when using Carrizo-L, several manufacturers opted to limit Carrizo to a unmarried channel in the first place.

That unmarried-channel isn't fifty-fifty loaded with high-speed RAM. The Carrizo systems AT evaluated all used DDR3L-1600. That'due south a non-trivial additional functioning hit in GPU workloads; AMD's APUs typically see virtually-linear performance gains from increasing the RAM clock. That's especially true in a single-channel arrangement, where bandwidth is even more limited.

Compared with Kaveri, Carrizo improves in multiple areas, including a huge reduction in video playback power consumption and official support for smoothen 4K video. Unfortunately, that's not the whole story.

Carrizo vs. Core M / Broadwell

The Core M / Broadwell comparisons aren't so neat for AMD. Carrizo sometimes wins against Cadre M (albeit with a much college TDP limit) and in 1 test (POV-RAY) even manages to compete with Broadwell. Mostly, however, it'south stuck between the two grade factors.

Carrizo nonetheless competes improve against Core Grand / Broadwell than Kaveri did, even handicapped by single-channel memory, but here's where the screws actually bite. In theory, Carrizo should be showing up in cheaper SKUs. AMD charges $150 for the Carrizo processors AT compares, versus Intel's $300 for the Core Thousand / Broadwell flavors.

In exercise, OEMs are building unmarried-channel Carrizo systems with $150 APUs and pricing them equally against dual-channel Intel systems with $300 SoCs. This doubly penalizes Carrizo. Not only is information technology going up against Intel processors with far more memory bandwidth, it'southward not being allowed to compete on price.

OEM sabotage

HP's $one,049 Elitebook 840 G3 has an Intel Core i5-6200U and 8GB of DDR4-2133 in a dual-channel configuration. The Dell XPS 13 uses the same CPU, merely offers 8GB of LPDDR3-1866.

HP's AMD-powered Elitebook 745 G3, in contrast, has just 4GB of DDR3L-1600 and the same price tag equally the $1,049 Intel organisation. Given that nosotros know AMD is selling the SoC for $150 less than the official list price on the Intel counterpart, that should raise more than a few eyebrows.

"Demolition" is a strong give-and-take to throw around, simply in my opinion these system prices (and the mostly depression quality of the Carrizo systems themselves) warrant it. Even in a best-case scenario, Carrizo can't exist said to match Core M or Broadwell — but the operation gap between AMD and Intel should exist starting time past a lower system toll. The supposed cost savings AMD offered OEMs past designing Carrizo and Carrizo-L to a common platform aren't being passed on to the consumer, and neither are the cost savings from using an AMD APU.

Anandtech dives into how positioning Carrizo / Carrizo-L as unified low-cost selections may accept encouraged OEMs to handicap the final products out of the gate. As things stand today, without significant discounting, it's hard to recommend Carrizo hardware — not because AMD didn't improve over Kaveri, only because OEM decisions and pricing take crippled the platform.

When AMD briefed us on Carrizo nearly a year ago, information technology made information technology clear that the flake was targeting the $400 to $700 laptop market, with Carrizo-Fifty shipping into the $250 to $400 range. Between $400 and $500, AMD would exist competing against the lower-finish Core i5-5200U from Intel, likewise as various Haswell-era Celerons and lower-end Intel parts. Instead, OEMs are edifice low-quality hardware at premium prices, tossing Carrizo against Intel chips information technology wasn't meant to compete against. This feeds a vicious bike — consumers who buy AMD and are dissatisfied with the result are less inclined to consider AMD in the futurity — for bug that ultimately, are beyond Chimpzilla's control.