Dual core at the ready
Just a word or two before we take a look at the new CPU and chipset. Intel provided HEXUS with the opportunity to evaluate a dual-core Pentium processor and 955X Express motherboard at short notice. The purpose of this article is to take a brief look at the CPU's new architecture and evaluate its performance with results based on a pre-production motherboard. We expect the benchmark numbers to change slightly with faster performance BIOSes and tweaked settings. On to the CPU.Intel Pentium Extreme Edition 840
The first thing you'll note is that Intel has dropped the number 4 suffix from the processor's name. It's referred to as a Pentium Extreme Edition 840.
Dual-core setup
The simplest method of delineating just what makes the new Intel Pentium XE 840 tick is to take a look at an overview of how it's put together.
As you can see, the dual cores are located on a single piece of silicon. Each core has access to its own 1MB of L2 cache and its own bus interface. Both CPU bus interfaces then connect up to the chipset's Memory Controller Hub, and hence share the system's bandwidth. Intel terms this dual-core setup as Pentium D, and each independent core is nearly identical to a current Pentium 4 Prescott's. For a closer look, you can see a Pentium 4 Prescott block diagram here and the Pentium D's here. The one big difference between the two processors is a lack of Hyper-Threading technology on all Pentium Ds. The Pentium Extreme Edition 840 has the best of both worlds. Its two HT-enabled cores allows a compatible OS to see up to four distinct CPUs; two physical and two logical.
It's this degree of parallelism that can offer up significant performance increases in dual-core processors running at the same clock frequency as traditional single-core models. In multitasking environments, where two or more applications are running concurrently, each independent core can execute code simultaneously. For example, games playing with media encoding running in the background. Greater processor parallelism, together with Hyper-Threading, is also extremely useful in instances when, say, applications are multi-threaded in nature. You could then have 4 threads running in a concurrent fashion, as shown pictorially below.
Hyper-Threading allows, assuming an HT-aware OS and a multi-threaded application or a multi-tasking environment, the XE 840 to execute integer and floating-point threads on each CPU. The Pentium D, on the other hand, does without Hyper-Threading, and therefore can only process a single thread per CPU. That doesn't mean the Pentium D should be any slower than a single-core HT-equipped Pentium 4, as its dual cores can still execute two threads without having to contend with a HT CPU's need to arbitrate between a shared cache. Further, a single-core HT CPU, under certain circumstances, can only process a single floating-point/integer thread at one time. The overall effectiveness of a dual-core HT-enabled CPU will obviously vary with just how threaded an application is and how well the code can be scheduled to meet the parallel processing power of a multi-core CPU.
Both Pentium D and Extreme Edition 840 CPUs run off an 200MHz system bus speed. Expect this to rise to 266MHz in the near future, matching the current single-core XE CPUs' speed. Both dual-core processors also carry additional 600-series single-core features such as EM64T (Intel's implementation of the AMD64 ISA), no-execute bit (stops code from being run in designated areas), Enhanced Intel SpeedStep Technology, and enhanced thermal monitoring.
Let's put the Intel Pentium Extreme Edition 840 into a table alongside current Pentium 4 models.
| 570J | 3.4GHz XE | 3.73GHz XE | 660 | XE 840 | |
| Frequency | 3.8GHz | 3.4GHz | 3.73GHz | 3.6GHz | 3.2GHz |
| Form factor | LGA775 | S478 | LGA775 | LGA775 | LGA775 |
| Number of cores | 1 | 1 | 1 | 1 | 2 |
Core| Prescott-1M |
Gallatin-2M |
Prescott-2M |
Prescott-2M |
2x Prescott-1M (2MB total) |
|
| Die Size | 112mm² | 237mm² | 135mm² | 135mm² | 206mm² |
| Bus Speed | 200MHz | 200MHz | 266MHz | 200MHz | 200MHz |
| Major pipeline | 32 stage | 20 stage | 32 stage | 32 stage | 32 stage (each) |
| Transistors | 125M | 169M | 170M | 170M | 206M |
| Process | 90nm | 130nm | 90nm | 90nm | 90nm |
| Caches | 16KB L1, 1MiB L2 | 8KB L1, 512KiB L2, 2MB L3 | 8KB L1, 512KiB L2, 2MB L3 | 16KB L1, 2MB L2 | 16KB L1, 2x 1MB per independent processor |
| ISAs and ISs | x86, SSE->SSE3 | x86, SSE->SSE2 | x86, x86-64, SSE->SSE3 | x86, x86-64, SSE->SSE3 | x86, x86-64, SSE->SSE3 |
| Power Techs | TM1, ACPI C0 (throttle only) to C3 | TM1, ACPI C0 (throttle only) to C3 | TM1, ACPI C0 (throttle only) to C3 | TM1->TM2, EIST, ACPI C0 to C3 | TM1->TM2, EIST, ACPI C0 to C3 |
| Operating voltage | 1.3v | 1.55v | 1.3v | 1.3v | 1.3v |
Even ticking along at 3.2GHz and under load, the dual-core XE 840 pumps out around 130w. That figure, though, is tempered by the larger dissipation area, and Intel shipped the XE 840 with a regular LGA775 copper-bottomed cooler. Load temperatures hovered around 65c, and the XE 840 was run through a 3-day testing period without issue. However, the load on each independent core needn't be the same. For example, the operating system may schedule load for one processor whilst keeping the other idle. Additionally, enhanced speedstep technology allows speed and voltage to be altered to best suit the current environment. Any voltage and speed changes will affect both cores simultaneously. The only real area of disappointment, I suppose, is the processor's 200MHz FSB speed. Enthusiasts could always drop the unlocked CPU's multiplier down to 14x and raise bus speed.
