Bobcat On The Prowl: AMD Dreams Of Intel Throwing In The Towel

Every year at the Intel Developer Forum, as readers of my past years’ conference coverage may already realize, AMD representatives camp out at a nearby hotel and seduce journalists away for real-time competitive briefings on whatever Intel’s announced at the show (along with other topics). This year, I was pleased to receive a demonstration of AMD’s Bobcat core for entry-level and mainstream systems, which I conceptually discussed several weeks ago in the context of the company’s Hot Chips coming-out party. At that time, I passed along AMD’s prognostication that it’d be sampling Bobcat-based CPUs in the fourth quarter of this year, with customer systems best-case in production one quarter later. Based on what I saw earlier this week, AMD’s got a solid shot of hitting its sampling schedules.

The Bobcat CPU core will initially appear in two product forms; an 18W part code-named Zacate targeting $500 notebooks and entry-level desktops, to be followed by a 9W proliferation currently referred to as Ontario for netbooks and thin-and-light notebooks. Both are dual-core CPUs, alongside an APU (i.e. graphics-plus processor) as the first silicon implementations of AMD’s long-touted Fusion vision. Clock speeds are at the moment unpublished; all that AMD will say about Ontario right now is that it disables (and presumably, at some point, deletes from the die) unspecified portions of the Zacate circuitry suite in order to accomplish the target price and power consumption objectives. I’d wager that Ontario will, for example, contain a fewer-core GPU aka APU than does Zacate.

In its demo suite a few blocks away from IDF at the Moscone Center, AMD pitted a Zacate reference design (thanks to Anandtech for the snapshot):

against a Core i5 420M Intel-based counterpart. Here’s another system shot, courtesy of Charlie at SemiAccurate:

AMD representatives politely demurred when I asked to peruse the Windows 7 System Information screen, so alas I’m unable to pass along important details such as clock speed, system memory allocation, HDD size and type, or even the specific Win7 flavor in use. Readers should therefore exhibit appropriate skepticism, as I am until AMD is ready to provide further details, in interpreting the results I saw. Nonetheless, they definitely piqued my interest.

Here’s the curious thing, though…the Core i5 420M isn’t even listed in Intel’s online database, although a Google search confirms that it is a valid product proliferation. And in the midst of doing that search, I uncovered a bit of AMD sleight of hand; although company had officials had claimed that the Intel CPU was a 2.4 GHz variant, that’s not exactly true…the Core i5 420M only runs at that speed in ‘Turbo’ mode, when one of the CPU cores is not in use and sufficient thermal headroom therefore exists for the other core to over-clock to the 2.4 GHz threshold. The conventional clock rate of the Core i5 420M is only 2.13 GHz. Nonetheless, as readers of my past ‘Arrandale’ writeups may recall, the mobile variants of the Core i5 moniker contain two CPU cores along with only 3 Mbytes of L3 shared cache (desktop-intended Core i5 CPUs, on the other hand, tend to be quad-core arrangements with 8 Mbytes of L3), with a graphics-inclusive companion die ‘next door’ in a multi-die, single-chip package configuration.

City of Heroes: Going Rogue averaged around 30 fps on the AMD system, roughly half that average frame rate on the Intel-based platform. I’ll confess that I was satisfied with neither system’s results, given the resolution and quality tradeoffs required to accomplish a perceived smooth playback rate on the AMD design. Nonetheless, the Fusion-integrated graphics proved worthy versus the Intel alternative, at least in this particular hardware and software configuration. And Zacate’s GPU is also DirectX 11 API-compliant, a promotional bullet (albeit with dubious at-the-moment real-life relevance) that Intel is unable to match either with Arrandale or with the coming-soon Sandy Bridge CPUs.

When AMD fired up the Internet Explorer 9 developer preview, however, my smoke-and-mirrors alarm went off at full intensity. The AMD platform achieved a 10x higher score than the Intel alternative in Microsoft’s Psychedelic benchmark. My briefing was early in the morning over breakfast, and I subsequently needed to rush off to an Intel keynote, so I didn’t have time to dig into the questionable results as much as I would have liked. Anand, on the other hand, had more time on his hands and discovered that AMD was using an out-of-date Intel graphics driver suite. Re-running the test with updated Intel software produced near-identical Psychedelic results for it and AMD’s competitive setups.

Anand ran a bunch of other game benchmarks and, fortunately for AMD, concluded that the graphics performance advantage versus Intel generally holds true. As such, especially for potential customers who are into gaming, video editing and playback (where I’m assuming that AMD’s graphics will also reign supreme over Intel counterparts) and the like, I remain bullish about Bobcat’s chances. Versus Atom, Bobcat will have a clear graphics lead for some time to come, and Bobcat’s out-of-order instruction execution capabilities will likely also deliver a clock-for-clock performance advantage over Atom…although Atom’s pricing advantage (or said another way, profit advantage for Intel at a given price point) will likely also exist as a counterbalance. And versus Arrandale, Bobcat may enable AMD to achieve particular performance metrics at lower pricing than Intel, especially if Intel’s approach requires a discrete GPU…although again, Intel’s 32 nm manufacturing process may level the playing field versus Bobcat’s reliance on 40 nm at foundry partner TSMC.

Followup: Although my notes say that the Intel platform was using a Core i5 420M, other journalists’ writeups which I’ve just perused claim that it was a Core i5 520M. Clock speed is the difference; the Core i5 520M runs at 2.4 GHz with a Turbo boost capability of up to 2.933 GHz. It’s possible that the AMD spokesperson mis-spoke during my particular briefing, or that I mis-heard what he said. I’ll ‘ping’ AMD for clarification and update this writeup if I hear back that the Core i5 520M was indeed in use.

Followup II: AMD just got back to me. The Intel CPU was indeed the Core i5 520M, not the Core i5 420M. The company rep regrets the “innocent mistake” on AMD’s part.