We've known that AMD had some sort of Atom-like CPU in the works for awhile—Dirk Meyer promised to discuss the product in November—but details past the project's existence have been hard to come by. If new information (and leaked slides) are correct, these new processors will be tipping up sooner, rather than later.
According to CHW (Google translation here), AMD will introduce two UVC (Ultra Value Client) SKUs in November. The Athlon-based chips will only be available to direct OEMs, and will both be paired with the AMD 740G platform. If you haven't heard of the 740G, you're not alone; AMD may have announced and launched the chipset, but it hasn't gotten very much press coverage. According to AMD's specification page, the 740G uses a die-shrunk 690G video solution (Radeon 2100) and integrates it with the updated SB700 southbridge. Only MPEG-2 is hardware-accelerated, and HDMI, DVI, and DisplayPort are technically supported (with appropriate connectors). Power consumption figures on the chipset aren't readily available, but it should come in well below the 945GC—virtually anything would.
Intel only uses the 945GC chipset for its low-end line of highly integrated Mini-ITX desktop motherboards, and that's an important distinction to remember. The Atom processors deployed in netbooks are typically N270s with a 2.5W TDP (as compared to 4W on the nettop-bound Atom 230). These netbook processors are paired with the 945GSE platform, which draws dramatically less power.
Clockspeeds, power consumption
It's always important to remember that AMD measures TDP one way, while Intel and VIA measure it another. The TDP values on the Athlon parts below represent the processor's theoretical absolute maximum power draw. Intel and VIA, on the other hand, list the TDP a processor cooler must be designed to dissipate. Neither of these definitions is wrong, though Intel's is arguably of more practical use, but the fact that they are different makes direct comparisons impossible.
As for the processors themselves, they'll ship in two flavors. At the high end (relatively speaking) there's the AMD Athlon X2 3250e, with a 22W TDP, 1MB of L2 cache, and an as-yet-unknown clockspeed (DigiTimes predicts 1.5GHz). That actually makes a fair bit of sense—we know that the 1.9GHz BE-2300 was classified as a 45W part; cutting clockspeed by 22 percent and lowering voltage could potentially hit a 1.5GHz/22W target. As for the 2650e, it's a 15W, single-core CPU at 1.6GHz. We won't know for certain until AMD launches the parts, but predictions are reasonable in both cases.
Given how it measures TDP, AMD's 2650e may draw significantly less than 15W, and AMD could probably trim that figure if they brought out a lower-clocked chip with, say, 256K L2. Even if AMD had such a part, however, it would do them little good without an accompanying chipset. Things may change on 45nm, but it's doubtful that we'll see an AMD netbook before 2009, and it's not a guarantee even then.
There's no doubt that both chips would beat the stuffing out of Atom in raw performance, as AMD's next slide indicates. VIA's Nano proved that a single-core OoOE chip at 1.8GHz was quite capable of whuppin' Intel's Atom 230; a dual-core Athlon X2 at 1.5GHz would crush Intel's chip. The actual Atom 230 processor may use just half the power of the hypothetical 2650e, but the 945GC chipset more than makes up for the Atom's power-sipping. We already know Intel plans to update Atom's desktop platform in 2009, but until it does so, it's vulnerable to a potential AMD counterattack.
AMD's slide doesn't show the Atom 330, but I'm honestly not sure how significant the omission is. Atom 230's performance lagged well behind VIA's in a number of tests, and while a second core would obviously have improved things, I'm not sure it would've been sufficient to completely close the gap. If I had to guess, I'd bet that the dual-core, 1.6GHz Atom 330 would compete with (and possibly surpass) a single-core 1.6GHz Athlon 64, but the dual-core Athlon X2 3250e, I think, would still take first place overall.
When I say guess, however, I mean it—nothing is certain until we've got reference hardware in our hands and on the lab table.Posted on