Welcome To Gulftown
Of course, Gulftown is enabled by Intel’s 32nm manufacturing process—the same node we saw debut back in January with the Clarkdale and Arrandale processor families. This time, however, enthusiasts don’t have to be bamboozled by a second, on-package 45nm die handling graphics, memory control, and PCI Express connectivity. The Core i7-980X gets us performance-freaks back to where we want to be—on-die memory controller, PCI Express handled by the well-endowed X58 chipset, and discrete graphics only, please.
With Gulftown, Intel uses its 32nm process to add cores and cache, rather than push integration. As a result, we have a six-core processor with 12MB of shared L3 cache. Architecturally, Gulftown is otherwise the same as Bloomfield. Each core gets 32KB of L1 instruction cache, 32KB of L1 data cache, and a dedicated 256KB L2 cache.
The 12MB shared L3 actually is a potential performance-booster. Because the cache can be dynamically allocated, an application that only utilizes one core can conceptually monopolize the entire cache. According to Intel, there are some gains to be had in gaming, for example, but it’ll be difficult to gauge just how much of the speed-up we see comes from increased core count versus cache, particularly since we’re using very few single-threaded benchmarks any more.
Despite the addition of two cores and 4MB of L3, Gulftown employs a smaller die than its predecessor (248 square millimeters versus Bloomfield’s 263). Transistor count increases from 731 million to 1.17 billion. That’s fairly incredible, considering the Core i7-980X fits within the same 130W thermal envelope as existing Core i7-900-series processors.
Gulftown’s memory controller remains unchanged, still rated for three channels of DDR3-1066 memory. This is actually somewhat interesting, since the 130W Westmere-EP processors that Intel plans to launch alongside Gulftown support DDR3-1333 (and with up to two modules per channel, no less). Nevertheless, we should see similar memory performance, as Bloomfield’s four cores clearly weren’t starved for data anyway.
The other addition worth noting is AES-NI, Intel’s hardware-based instructions for accelerating the cryptography standard. Previously seen only in the company’s Clarkdale-based Core i5s (and unfortunately left out of the other Clarkdales), AES-NI isn’t yet having a massive effect on performance. But as we’ll see in the benchmarks, there’s a ton of potential there.
|2010 Intel Core i7 Processor Family|
|Base Clock||Max. Turbo Clock||Cores / Threads||L3 Cache||Memory||TDP||Price|
|Core i7-980X||3.33 GHz||3.6 GHz||6/12||12MB||3 x DDR3-1066||130W||$999|
|Core i7-975||3.33 GHz||3.6 GHz||4/8||8MB||3 x DDR3-1066||130W||$999|
|Core i7-960||3.2 GHz||3.46 GHz||4/8||8MB||3 x DDR3-1066||130W||$562|
|Core i7-920||2.66 GHz||2.93 GHz||4/8||8MB||3 x DDR3-1066||130W||$284|
|Core i7-870||2.93 GHz||3.6 GHz||4/8||8MB||2 x DDR3-1333||95W||$562|
|Core i7-860||2.8 GHz||3.46 GHz||4/8||8MB||2 x DDR3-1333||95W||$284|
Hyper-Threading And Turbo Boost Persist
Hyper-Threading and Turbo Boost were both interesting new additions to Bloomfield. They naturally are a part of the Gulftown story, too.
We hadn’t seen Hyper-Threading in years prior to Bloomfield. Because the software community has become better about threading since then, though, the feature was more of a boon to Nehalem than it was to Pentium 4. Thus, the same technology that allowed four cores to address eight threads now enables six cores to juggle 12. At the very least, this makes for a cool screen shot, especially from a single-socket desktop.
Turbo Boost carries over as well. We were really starting to get excited about Turbo when the Lynnfield-based quad-core chips emerged with four and five speed bins (133 MHz increments), giving us up to 533 MHz with a single core active. Unfortunately, Gulftown drops us back to Bloomfield’s more conservative binning structure. When one core is active, you’ll see two bins (or 266 MHz) of speed-up, yielding 3.6 GHz. With two or more cores active, you get a one-bin boost to 3.46 GHz.