Triple Core: Can AMD Even The Score?
In reflecting back on yesterday’s formal unveiling of additional triple-core Phenom microprocessors (whose premier precursors the company unveiled a month ago), I have some thoughts. Then again, based on past coverage history, this is perhaps not a surprise to you!
Two of the three CPUs (the 8450 and 8650) have identical clock speeds (respectively 2.1 and 2.3 GHz) to their predecessors; the third (the 8750) cranks up the core to 2.4 GHz (interestingly, at no claimed thermal design power i.e. TDP tradeoff). So what’s changed? Reflective of the ‘50′ in the device monikers, these parts are based on the B3 product stepping, which fixes the previously discussed TLB cache erratum, and are therefore candidates for retail channel direct sales to consumers.
My AMD technical contact firmly positions the parts (specifically, the low-end model) and their companion integrated graphics-inclusive core logic chipset at the "$500 PC" segment of the market currently dominated by AMD and Intel dual-core products. As such, I don’t agree with other commentators’ suggestions to ‘just get a quad core’, unless you’re building your own system from scratch. PC OEMs will tend to bundle fully functional quad-core AMD Phenom (and, for that matter Intel Core 2) CPUs with other higher-end system building blocks:
- Premium core logic chipsets
- Discrete graphics cards
- Large, fast HDDs
- Abundant DRAM at high-end speed bin SKUs
- etc
thereby resulting in systems that end up vastly overshooting the $500 threshold.
I agree with my AMD contact that $500 and below systems represent the bulk of desktop PC sales at the present time. But how does an AMD triple core-based bill of materials list stack up against an Intel dual-core equivalent, in this particular market segment? As I suspected in advance, and as numerous benchmarking tests published in the past 24-ish hours concur:
it’s pretty much a ‘wash’, at least at the CPU level. The good news for AMD is that it can now compete against Core 2 without completely crashing chip prices, at least on some benchmark suites, whereas it wasn’t previously able to do so with Athlon 64 X2. The bad news for AMD is that in order to do so, it needs to throw an additional CPU core-per-die at the benchmarks in order to keep up, since each core runs at a clock cycle deficit as compared to the Intel competitor. And, of course, the AMD die is fabricated on a 65 nm process, whereas Intel is currently running a ‘fire sale’ on its remaining 65 nm inventory in order to complete the transition of the entirety of its product line to 45 nm.
For heavily multi-threaded applications such as video editing suites, or in heavily loaded multi-tasking environments, a slower-clock triple-core CPU may exhibit an overall benefit versus a faster-clock dual-core equivalent. However, in the bulk of today’s usage environments, characterized by single-threaded conventional applications and light overall loading on average, the additional core will largely go unused and per-core clock speed will therefore gain prominence. Granted, over time (hopefully, given clock speed and core-augmentation trends!) PC software in general will migrate towards (at the per-application level) multi-threading and, (at the operating system level) multi-tasking. And I don’t at all underestimate the ‘bragging rights’ aspect of three cores versus two. But practically speaking, AMD either needs to get its clock speeds up to Intel levels or come up with a vastly more clock-efficient architecture tweak (reflective of Athlon-vs-NetBurst days past) in order to regain its standing as a solid competitor.
Continue reading with ‘AMD: More Persuasive From A System Perspective‘…
