AMD’s Computex 2017 conference was short, just less than an hour long, and included very few new things that haven’t been announced in the past. The company instead took the opportunity to play up its revamped relationships with their OEM partners, and a good chunk of the conference was spend on third-party companies using or implementing Ryzen processors and Radeon GPUs into their products. Far from being a disappointment, AMD’s conference was littered with a few scraps of new info among the OEM announcements, and we’ll take a quick look at those today while I catch up with the rest of Computex news.

AMD’s EPYC platform gets a launch date

AMD’s EPYC server line up was detailed a bit further in this morning’s conference, and some interesting pictures popped up when AMD was talking about their position in the market when it came to socket options for EPYC. They showed off a single-socket and a dual-socket motherboard. There are some key things to note here.

The first is that these motherboards don’t have much in the way of USB 3.1 or video ports to speak of. The benefits of the Zen platform is that AMD can chop and change what they like for the targeted customer, and for EPYC, this means getting rid of all the unnecessary I/O ports in order to increase the number of lanes available from the CPU to add-in cards. In this case, they’re tearing out the I/O to open up four more lanes; in dual-socket systems there are 128 lanes available from the processors, and 64 lanes available on single-socket systems. Only four lanes are taken up by the chipset that AMD sells on these boards that enables basic storage solutions. That’s a lot more flexibility than Intel allows on most of their server options.

The single socket board option is quite neat. There’s the full complement of RAM slots available, and every EPYC processor can use all eight memory channels. The PCIe slot placement is a little weird, and it looks like there are connectors for PCIe riser cards to connect to the 4x and 8x white slots. Dual M.2 NVMe storage is supported, with the drives laid out in a good location for airflow, and there are only two SATA 6 ports on the board. It looks like AMD’s intention with this reference design is to provide the basics, and get out of the way as much as possible so that you can put in whatever you like that fits your requirements, like a RAID card or a GPU for compute workloads.

We have a launch date for EPYC as well – 20 June. This slightly precedes general availability of the Radeon Vega Frontier Edition on 27 June, so AMD’s corporate and enterprise partners will get to use these products first. AMD is on schedule with this launch, and unusually so, which means that Radeon Vega RX and Threadripper are also on schedule for launch in Q3 2017. That’s going to be quite a busy period for AMD.

Ryzen Mobile makes its first appearance

During the conference, AMD showed the audience, for the first time ever, their new Ryzen Mobile APU. Beneath this small, svelte package is four Ryzen cores in a single CCX, with simultaneous multithreading enabled, as well as a Radeon Vega graphics core. It is the fastest APU AMD has ever made by some margin barring the versions currently shipping in consoles, and AMD will be selling Ryzen mobile to interested third parties for integration into convertible laptops, ultrabooks, mid-range machines, and high-end gaming desktop replacements. This announcement is separate from other Ryzen-based notebooks which I’ll detail later, but this is the first glimpse of the APU that will bring the heat to Intel in the notebook market.

AMD also brought out a 2-in-1 convertible ultrabook running on Ryzen Mobile. The reference design looked very similar to an HP Envy ultrabook with the black-and-gold trim, the squared off hinges, and the lack of any useful ports like USB 3.1 Type-A. This is the first thin-and-light design that AMD has been able to find their way into in almost half a decade, and it should be plenty fast for productivity workloads and some light gaming on the side.

Gaming on an ultrabook has never been realistically possible with Intel’s graphics, and the most you could ever hope to do is to hook up a Thunderbolt 3.0 enclosure for a GPU and hook it up that way. This might finally be the catalyst for AMD to be taken seriously in the notebook market again.

Threadripper in more detail

AMD’s Threadripper family is due to launch quite soon (by July, some sources say), and AMD said that it would be “very competitive” with Intel’s current offerings. Offering up to 16 cores and 32 threads, Threadripper threatens Intel’s bottom line when it comes to the newly announced Core i7-7920X, and it could potentially have a lower retail price than Intel’s offering. In addition, AMD threw down some bombs with the announcement that every Ryzen Threadripper processor has access to all 64 PCIe 3.0 lanes available to the platform, as well as quad-channel ECC-registered memory support as standard across the range. Those are not things that Intel’s Basin Falls platform can boast.

On top of that, an X399 motherboard can hold up to four graphics cards in it, splitting all the lanes properly, without needing an external chip from a third-party. This makes Threadripper quite an attractive budget system for machine learning workloads or chewing through large rendering workloads – pop in a suitable processor, fill up the RAM slots, and then put in four of whatever GPU you prefer using to run through the workloads you need, all without having to purchase an expensive Xeon system that costs much more.

Ryzen Threadripper Core i9-7920X
Chip family Zen Skylake-X
Cores/threads 16/32 12/24
Base clock speed TBD TBD
Boost clock speed TBD TBD
Turbo 3.0 clock speed TBD TBD
L3 cache 32MB 16.5MB
PCIe lanes 64 44
Memory channels 4 4
Max DDR4 speed 2666MHz 2666MHz
Thermal design point (TDP) TBD 140
Launch price $999 (best guess) $1199

If we take the top-end Threadripper at a retail price of $1000, which is in line with what AMD’s pricing is expected to be given that the Ryzen 7 1800X is only $499, Threadripper could well steal the show. With more cores than Intel’s highest-end Core i9 processor (the Core i9-7980XE doesn’t count yet because Intel is launching that part much later), and more cache, it would make for a very capable workstation that can crank out more work than Intel’s chips can. Intel, however, has a clock speed advantage due to a more mature production process, so we could see something around a 500MHz gap when comparing boost clocks between the two.

Processor performance is much more than just those stats though. The performance of the cores themselves, the caches, the latency between cores and caches adjacent to each other, and a myriad of other factors may affect the ranking here. Skylake-X, at the very least, is said to be around 30% faster than the previous generation, but some of that improvement comes from simply having more cores and a higher clock speed. When all is said and done, Threadripper might remain competitive after all.

That’s all highly speculative, of course. AMD hasn’t spoken to me or anyone in the press about Threadripper’s specifications or its performance yet. The only thing we have to go on is a Blender demo AMD ran at the event, where the 16-core Threadripper completed a rendering benchmark in around 13 seconds. That same workload on Broadwell-E and the Ryzen 7 1800X was completed in around 34-36 seconds using the same presets when it was last shown, so the scaling from one on-chip die to two in Threadripper has better scaling than expected.

Finally, the first shot of Threadripper. It is the same package size as EPYC, which might lead one to expect that they share the same socket. I’d like to believe that’s the case, because AMD could then offer two sockets in total for the desktop market, and give consumers on X399 an extremely wide range of options for a workstation processor. The mere thought of being able to slap a 32-core chip into a consumer socket is probably enough to get Intel worried, because the most they’ve ever offered to consumers on the HEDT sockets is the Xeon E5-2699 V4, an underclocked 22-core, 44-thread processor that costs $4000 on its own in a tray (and you would have to buy 1,000 of them to get that price).

Both AMD and Intel came back strong at Computex and took shots across the bow at each other, but both come away better off than they were before – Intel is in a stronger position to leverage higher core counts for enthusiast platforms, and AMD finally gets the chance to make a good impression on owners of older Intel HEDT processors that may be looking to upgrade to a platform that has better multi-threading capability. The rest of the year is going to be pretty exciting.

More stuff like this: