AMD’s Bristol Ridge mobile APU lineup leaks on Geekbench

AMD Kaveri analysis header 800x450

AMD is getting ready to launch a new family of APUs on a brand new platform, socket AM4, which is expected to debut at Computex 2016 next month. While details on the chips themselves are rather thin on the ground, recent benchmarks appearing on the Geekbench website spill some of the beans about the mobile versions of these chips and their expected performance, and it looks like we’re going to see some interesting competition coming to the market very soon.

UPDATE: According to documents that I’ve found, Gardenia motherboards have been used by AMD before in their R-Series APUs and in embedded systems for testing Android compatibility. Thus, the family name on the table has been renamed from “Gardenia” to “Bristol Ridge” and the only new motherboard platform is codenamed “Myrtle”. It is possible that mobile Bristol Ridge APUs only make use of DDR3 memory, although DDR4 support is still a possibility.

Aside from the frequencies and Compute Core setups listed here, there’s very little I know about the “Gardenia” family, which are likely only ever going to appear in notebooks. There are some details that one can glean from the APU listings on Geekbench’s leaderboard, which list motherboards from AMD as well as Lenovo being tested with the platform, which means that AMD will launch with Lenovo as a hardware partner. The codename for the platform also appears to be “Myrtle”.

AMD Bristol Ridge Mobile APU Family

AMD Unknown FX-9830P FX-9800P A12-9700P A10-9600P
 CPU core/thread count 2 / 4 4 / 4 4 /4 4 /4 4 /4
 Radeon designation  — Radeon R7 Radeon R7 Radeon R5 Radeon R5
 GPU core count 512 512 384 384
 Compute core count 12 (4+8) 12 (4+8) 10 (4+6) 10 (4+6)
 Base clock 2.0GHz 1.9GHz 1.85GHz 1.8GHz 1.65GHz
 L2 cache 2MB (2x 1MB) 2MB (2x 1MB) 2MB (2x 1MB) 2MB (2x 1MB) 2MB (2x 1MB)

The processor IDs are listed as “AuthenticAMD Family 21 Model 101 Stepping 1”, which is consistent with AMD’s naming scheme for their processors, but all of them are clearly based on Carrizo – the listing for the Athlon X4 845 is “Family 21 Model 96 Stepping 1”. Either these are refreshed processors designed with new chipset capabilities to fill in the gaps left in the market by the lack of Carrizo-based notebooks, or they are new designs based on Carrizo that also have DDR4 memory compatibility. The cache construction is identical, with 2x 96KB of instruction cache, 4x 32KB of data cache, and 2x 1024MB of L2 cache memory, so most of the innards should be untouched.

What’s interesting about these processors is that their clock speeds are really low, atypically low for an AMD chip. For the last few years, the Bulldozer architecture has launched with some really high-clocking chips, and AMD practically perfected clock speed regulation by the time their Godavari APUs rolled out to market. But those are based on Kaveri, not Carrizo, and these chips are clearly going to be very different performers. The top-end APU sits at 2.0GHz on what appears to be two cores, while the A10-9600P drags along at 1.65GHz.


A fifth chip is also listed on Geekbench, and it does appear to have some sort of hyper-threading enabled. Ever since Microsoft rolled out a hotfix for AMD’s launching Bulldozer family, Windows has always seen a Bulldozer module as two separate, physical cores. The Athlon X4 845 is also listed as a quad-core, quad-threaded processor. However, given the chip’s makeup, it probably isn’t Zen. The FX-8800P also registers as a dual-core, four-threaded processor in some systems built by HP and Acer, and AMD’s development motherboard might be referring to the modules in the same way. This might confuse someone thinking that it’s Zen, as I was when I first stumbled across this info.

If you’re in doubt, take a look at the block diagram for Zen. It features only 512KB of cache per core, which means that the arrangement for a quad-core APU based on Zen would be 4x 512KB L2 cache, not 1MB for every two modules. Zen’s basic building blocks also include L3 cache, and this might mean that we’ll see L3 cache on every Zen-based processor AMD ever makes, right down to the dual-core, four-threaded bargain basement chip that will sell for a relative pittance.

Near as I can see, there will also be several models from Lenovo available on launch or within a few weeks of it, and the basic spec details are as follows, ordered by their Geekbench single-core scores:

CPU model Board name System RAM Geekbench Single core Geekbench Multicore
ASUS Intel Core i5-6200U ASUS UX303UA 8GB 2884 5766
AMD Unknown #1 AMD Myrtle 8GB 2584 7985
AMD Unknown #2 AMD Myrtle 8GB 2579 8105
AMD FX-9800P Lenovo Toronto 5B1 8GB 2265 5607
AMD FX-9800P Lenovo Toronto 5B1 8GB 2216 5596
AMD A12-9700P Lenovo Toronto 5B1 8GB 2179 5254
AMD A12-9700P Lenovo Toronto 5B1 6GB 2174 5319
AMD A12-9700P Lenovo Toronto 5B1 6GB 2173 5436
AMD A12-9700P Lenovo Toronto 5B1 8GB 2150 5285
AMD FX-9800P Lenovo Toronto 5B1 8GB 2113 5476
AMD FX-9800P Lenovo Toronto 5B1 8GB 2108 5431
AMD A10-9600P Lenovo VIUU4 4GB 1966 4711
AMD A10-9600P Lenovo VIUU4 4GB 1930 4890
AMD A12-9700P Lenovo Toronto 5B1 6GB 1923 4995
AMD A10-9600P Lenovo VIUU4 4GB 1800 3957
AMD A10-9600P Lenovo Toronto 5B1 6GB 1542 3398
AMD A10-9600P Lenovo Toronto 5B1 6GB 1517 3375

The immediately obvious thing to notice is that there are definitely some systems that should be avoided if you want the best performance out of the chips. The bottom two models are clearly not performing up to the spec that the A10-9600P could reach, and the third-from-last model on a VIUU4 motherboard is possibly thermal throttling as well – the single-core performance is high, but multi-core performance is well behind the other two A10-9600P units on the same motherboard type. My guess is that there are chassis changes that allow for these performance improvements.

Anything from the FX-9800P and upwards seems to be perfectly fine in performance and none of them seem to indicate thermal issues, with the exception of the A12-9700P system sandwiched between the systems with VIUU4 motherboards. It has lower single-core performance than other systems of the same type, but its multi-core performance is much higher. If you want the best, though, prepare to stump up the funds for a FX-9800P system – whichever one you choose will be just fine, and within striking distance of Intel’s Core i7-6700HQ.

For kicks, I added in the scores from the unnamed engineering samples I talked about earlier, as well as the scores from an ASUS system with an Intel Core i5-6200U for comparison. One thing’s for sure, whatever’s inside these puppies is definitely within Core i7 territory, and it’s much faster than anything currently shipping on the Skylake platform, at what would presumably be a similar price point to the Core i5-6200U (currently $281 in 1000 tray unit quantities). Intel’s single-core performance is still on point even in a thermally constrained environment, so it would be interesting to see what kinds of systems end up with that unnamed AMD APU.

AMD will be hosting a web seminar on 18 May 2016 to talk about Polaris, and the company is also reportedly planning a 1 June 2016 launch at Computex 2016 that will detail everything including the upcoming Bristol Ridge APU family for the desktop, as well as the APUs I’ve just uncovered here and more hardware in other areas. It’s an exciting time to be a PC enthusiast.

Source: Geekbench

The top games of 2019, according to us