Intel's Pentium XE 965 and 955 Processors

Out with the Smithfield and in with the Presler

Intel's first round of dual-core processor offerings took off in July 2005, besting AMD's efforts by nearly a month. Under normal circumstances, having a market lead is always a formidable advantage, but in the case of the Smithfield core used for the first generation of Pentium D processors (800-series), things didn't turn out all that rosy.

First and foremost, the Smithfield was basically a larger Prescott with dual processing cores on the same die; and we all know how well received the Prescott lineup itself was. A normal Prescott core processor was already running very warm and its performance to price ratio was contested easily by AMD's Athlon 64 lineup. With that said, having two more of the exact same cores put together isn't going to resolve any of the existing concerns other than offering a dual-core variety, which does boost performance but not as much as its competitor. Needless to say, heat again was the Achilles heel of the Pentium D 800 series and its performance didn't scale like that of AMD's Athlon 64 X2 series. The exorbitantly priced Pentium Extreme Edition 840 was further hurt by the fact that it was a few hundred dollars dearer than the Pentium D 840 and was identical to the non Extreme Edition in every way but the exclusion of Hyper-Thread technology. As we've explored before, Hyper-Threading for multi-core processors isn't very elegant unless the software is really aware if it's sending a thread for execution to the logical or physical core depending on the processor's state. Also, despite the fact that Intel was retailing their dual-core processors at a more favorable price than AMD, its overall performance and other operating qualities didn't strike the right chord with the more savvy DIY market.

Riding out for many months settling for second place to AMD for its processor lineup, Intel finally brought out a new series of mainstream dual-core processors based on the 65nm process in early Jan 2006. Codenamed Presler, the new core wasn't just a die-shrink of the existing Smithfield as we reveal to you more on the Presler core's improvements on the following page and how it impacts both Intel and yourself as the consumer. As with standard Intel processor model number nomenclature, the newer processors were christened as the 900-series, indicating that the significant changes are to be expected when compared to the existing 800-series. It was also a good start to shed off the poor image that the 800-series cast during their tenure as premium CPUs from Intel. To kick-start our evaluation of the Presler core, we begin with the best of what Intel has to offer, the Pentium Extreme Edition 955 processor and the very recently but quietly introduced Pentium Extreme Edition 965 processor.

The Presler Core

With the maturity of Intel's 65nm process engineering, it was a golden opportunity for Intel to spruce up its current processor lineup for both the desktop and mobile platforms. Massive efforts went towards the formulation the Yonah core for the laptop market, which is Intel's first dual-core mobile processor that's now known as the Core Duo processor (and incidentally, it's also the design basis of the upcoming Merom and Conroe processors). The desktop side was not to be outdone as well and the new Pentium D 900-series debuted at about the same time frame as Core Duo did. While the new Pentium D based on the Presler core wasn't as groundbreaking as Yonah, it had enough advancements to categorize it as an evolutionary step up.

The old Smithfield based Pentium D 800-series processors had a single large die (read as 'expensive') which fused two cores with 1MB of L2 cache each and were basically two Prescott-class processors put together. Sharing the same roots as its single-core Pentium 4 counterparts, the feature set was identical to them in every way - Netburst architecture, deep 31-stage pipeline (per core), Execute Disable Bit, Intel EM64T and Enhanced Intel SpeedStep Technology (EIST). However, unlike the 600-series Pentium 4 processors which sport a huge 2MB L2 cache, the 800-series Pentium D has a total of 2MB L2 cache; meaning 1MB exclusively for each core only. Based on their then 90nm process technology and the CPU structure of the Smithfield, Intel could only qualify them until 3.2GHz and were still operating off the old 800MHz FSB. The reduced cache quantity available to each core along with the limited FSB greatly hampered the Pentium D's dual-core processing throughput in addition to other shortcomings such as its limited clock speed range and the high cost of manufacturing the Smithfield.

The newer Preseler core of the Pentium D 900-series approaches the dual-core situation in another route. Instead of having one single monolithic die with two cores, Intel went with two separate dies (with one core each) put together on the same packaging (as shown in the above diagram). Since each core or die can exist on their own in the form of the Pentium 4 single-core lineup, Intel can maximize processor production output quantity and minimize chances and cost of discarding the low-yielding or faulty dies.

Each of these dies are manufactured on the 65nm process technology, have a full 2MB L2 cache, support all of the other features found in the previous generation plus Intel Virtualization Technology (VT). While virtualization was always possible via the software stack alone, the inclusion of hardware support in the CPUs would greatly simplify the software side of things as well as increase the experience of using multiple virtual environments concurrently for dedicated usage needs without compromising stability or performance. Lets get back to the processor talk, as we'll be covering more about VT in a separate article. The single-core version goes by the codename of Cedar Mill and you can identify Pentium 4 processors using them via CPU model numbers of the 6x1 (without VT) and 6x2 class (with VT). Having mentioned what Presler is about earlier, it is basically two Cedar Mill cores put together on the same package.

The Presler will have 2 x 2MB L2 cache for a grand total of 4MB per CPU, but like its predecessors, each core can access its own cache exclusively. With more cache per core closer to the processing units than the Pentium D 800-series, the Pentium D 900-series is bound to benefit from this larger pool of ready data to feed the deep processing pipelines. Should there be any branch prediction error, the chances of having a cache hit to retrieve the necessary data increases two-folds. Of course with a larger cache, the Pentium D 900-series has a better prefetch mechanism to fill it up as well.

The L2 cache of a modern processor requires significant die real estate and with Presler having twice that of the Smithfield, the transistor count has shot up to a staggering 376 million (both cores combined) from the 230 million on the Smithfield. Thanks to the 65nm process technology, it has truly enabled Intel to pack more on the processor and surprisingly, costs less to produce than its predecessor! To prove this point, the die size of the Smithfield is 206mm2 while the superior Presler core is only 162mm2 (both cores combined). Given the same 300mm wafers used for production, Intel now can obtain up to 25% more Pentium D processors per wafer. And since the same dies are also used for the top-end Pentium 4 series, Intel is maximizing all its efforts to gain more for reduced costs. As such, the 900-series processors retail at the same price or slightly less than the 800-series. Interestingly, the Thermal Design Power envelop of these new processors remain the same as the older generation despite the new process technology adoption. It is highly likely that the added features and vast increase in transistor count negated any power savings obtainable.

The new Pentium Extreme Edition Processors based on the Pentium D 900-series have yet another two aces up their sleeves � an FSB boost to 1066MHz for a higher throughput and higher clock speeds up to 3.73GHz on the 965 processor model. These are in addition to Hyper-Threading technology to give them a total of 4 logical CPUs. Together with the enhancements made to the Pentium D lineup, the top-of-the-line Pentium XE processors can be a formidable challenge to the AMD's best Athlon 64 X2 and Athlon 64 FX processor series. Laid out below is a feature and characteristics comparison table of the current best-in-class CPUs from both Intel and AMD:-

Test Setup & Benchmarks

To find out if all this feature-talk of the Presler CPUs really made a strong impact, we take Intel's top guns for a spin with AMD's best of the lot. So that encompasses Intel's Pentium XE 965 and 955 against AMD's Athlon 64 FX-60 and FX-57 CPUs along with the Athlon 64 X2 4800+. To have an overview of how much the new Intel processors have progressed from the past, we've also got the Intel Pentium XE 840 and Intel's fastest single-core CPU - the Pentium 4 XE 3.73 GHz. Coincidentally, the latter's clock speeds corresponds with that of the very new Pentium XE 965 processor, so it's going to be interesting to find out to what extent dual-core and more cache has helped improve computing performance.

The components used in the AMD testbed include:-

• ASUS A8V Deluxe (VIA K8T800 Pro chipset)
• AMD Athlon 64 FX-57/FX-60 and Athlon 64 X2 4800+ processor
• 2 x 512MB Corsair DDR400 non-ECC memory modules (CAS 2.5, 3-3-6)
• Seagate Barracuda 7200.7 80GB SATA hard disk drive (one single NTFS partition)
• MSI GeForce FX 5700 256MB (AGP) - with NVIDIA Detonator XP 61.34 (beta)
• Microsoft Windows XP Professional with Service Pack 1

The components used in the Pentium XE 840 and Intel Pentium 4 XE testbed include:-

• Intel D955XBK Desktop Board (Intel 955X Express chipset)
• Intel Pentium Processor Extreme Edition 840 processor (with Hyper-Threading) / Pentium 4 Extreme Edition 3.73 GHz (with Hyper-Threading)
• 2 x 512MB Micron DDR2-533 non-ECC memory modules (CAS 4, 4-4-12)
• Seagate Barracuda 7200.7 80GB SATA hard disk drive (one single NTFS partition)
• MSI GeForce PCX 5750 128MB (PCI Express x16) - with NVIDIA Detonator XP 61.34 (beta)
• Microsoft Windows XP Professional with Service Pack 1

The components used in the Pentium XE 965 and Pentium XE 955 testbed include:-

• ASUS P5WD2-E Premium (Intel 975X Express chipset)
• Intel Pentium Processor Extreme Edition 965 / 955 processors (with Hyper-Threading)
• 2 x 512MB Kingston DDR2-667 non-ECC memory modules (CAS 4. 4-4-10)
• Seagate Barracuda 7200.7 80GB SATA hard disk drive (one single NTFS partition)
• MSI GeForce PCX 5750 128MB (PCI Express x16) - with NVIDIA Detonator XP 61.34 (beta)
• Microsoft Windows XP Professional with Service Pack 1

The benchmarks used in this review include:-

• BAPCo SYSmark 2004
• SPECCPU 2000 v1.2
• Lightwave 3D 7.5
• Futuremark PCMark 2004 Pro
• Cinebench 2003
• WSTREAM
• XMpeg 4.5 (DivX encoding)
• Futuremark 3DMark03 Pro
• Futuremark 3DMark05 Pro
• Unreal Tournament 2004
• AquaMark3

Results - SYSmark 2004

After numerous attempts by Intel over several months, their current flagship Pentium XE 965 finally garnered pole position in SYSMark 2004, but barely. The initial Presler core based Pentium XE 955 was no match for even the Athlon 64 X2 4800+, let alone an FX class dual-core processor. Clock speed has always been Intel's major ace to try and even things up and we can see why Intel quietly slipped in the Pentium XE 965 into their lineup recently, but availability and price is another discussion altogether which we'll tackle at the end of the article.

Since the two different suites of tests determine the overall score encompassing office productivity applications and media content creation applications, let's take a quick look at the breakdown of scores. In the Internet Content Creation test suite, the Athlon 64 FX-60 took the lead a couple of percentage points ahead of the Pentium XE 965 due to its more efficient processing architecture but the tables turned in the Office Productivity suite where the large ever-changing data sets involved highly favored the huge 2MB cache per core of the Pentium XE 955 and 965 processors. You can also see how unfavorably the old Pentium XE 840 fared with its restricted FSB, smaller cache and lower clock speeds.

Results - SPECCPU 2000 v1.2

Thanks to the bolstered specifications of the Pentium XE 965 and 955 processors, they eclipsed the Pentium XE 840 predecessor by a healthy margin of 24% and 16% respectively in both the integer and floating-point departments. The performance of the Athlon 64 series on the other hand was a mixed bag of results with integer performance a tad lower than the new Pentium XE processors, but were greatly disadvantaged in the strong floating-point tests since they do not posses the kind of raw clock speed prowess of Intel CPUs. However, the SPEC base speed tests only evaluate single-thread performance, so scroll down to see what how dual-core performance fared.

In the SPEC rate tests, we got the a chance to evaluate dual-core performance better by having the CPU run two copies of the same workload and the results below rank the Intel Pentium XE 965 just a peg faster than the AMD camp in integer workloads, but was some 20% speedier in the floating-point department for the same reasons as above. The Pentium XE 955 on the other hand was probably in the same performance ballpark as the AMD dual-core processors.

Results – Lightwave 3D 7.5

The Athlon 64 lineup takes a huge win in this rendering benchmark that has benefited tremendously from its architecture's short pipelines and efficiently tweaked memory controller. So much so that even the single-core Athlon 64 FX-57 is able to outrun the Pentium XE 965 when addressing two threads at a go in the more intensive Tracer- Radiosity test.

Results - Futuremark PCMark04

Next, we have a few lighter tests from the PCMark04 test suite to quickly assess the CPU's overall performance and the platform's memory subsystem. As the results below depict, the respective high-end Intel and AMD parts seem to return very similar CPU scores, but overall memory performance goes to the Intel camp despite AMD's well tweaked memory controller. This is because Intel's memory subsystem runs at a much higher clock rate and has an overall greater throughput. However there are certain usage roles and tests like the Lightwave 3D benchmark we've seen earlier that favors a more responsive lower latency memory subsystem found on the AMD processors.

So there are both pros and cons with either memory subsystem, but with AMD following Intel to adopt DDR2 memory on their upcoming AM2 processors, it is possible that future AMD processors might perform closer to what Intel processors are scoring currently for their memory index. The final outcome would be very dependent on how well AMD tweaks its on-die memory controller and we'll find out in less than a month's time.

Results - Cinebench 2003 & XMpeg 4.5

Like the Lightwave 3D application, Cinebench 2003 is another rendering benchmark and Intel barely took the lead here thanks to its Hyper-Threading technology giving it up to four logical CPUs for processing. Had it not been for that, AMD could have easily swept the lead position as the results do indicate that it is hot on Intel's tail. Interestingly, the former best from Intel, the Pentium XE 840 and Pentium XE 955 don't fare well in both their performance and price-to-performance ratio as the newer and dearer Pentium XE 965.

Video encoding tests such as that we performed were all in favor of CPU clock speeds and it was no surprise to have the Intel pack leading in this comparison.

Results - Futuremark 3DMark03 and 3DMark05

Heading into some gaming related benchmarks, we dished out a couple of synthetic tests from Futuremark to see how the CPUs fare in tackling games. 3DMark03's CPU score is very depictive of single-threaded gaming performance and as was customary, AMD's Athlon 64 lineup clinched top spot. 3DMark05 introduced elements to take advantage of the multi-threading capability of modern CPUs and as such we found Intel's Pentium Extreme Edition processors' quad logical cores to stand in good light ahead of the Athlon 64 X2 and Athlon 64 FX-60 chips. However, AMD's highly clocked single-core FX-57 processor still outpaced all others in this test and held the gaming edge. It won't be long before AMD validates a dual-core version of the FX-57 and that should really trounce Intel's offerings by a larger margin. We'll reserve further comments till it does show up.

Results - Unreal Tournament 2004 and AquaMark 3

Throwing in a couple of real-world benchmarks, we see that AMD's Athlon 64 X2 and FX series are still the undisputed leaders despite Intel's much newer offerings.

Conclusion

Intel's dual-core processor lineup as of last year had left savvy users wanting more and their alternative was AMD's Athlon 64 X2 which had proved itself as a very competent alternative despite its higher price. This year, Intel promptly switched over their 65nm process technology for their latest processor lineup and taking advantage of their technology leadership, they had given their desktop dual-core CPUs a much-needed boost for competitiveness. The new Pentium D 900-series processors now feature up to 4MB of L2 cache (double its predecessors), are available at higher clock speeds and thanks to the new design and manufacturing process, cost much less as well. With higher overhead available from moving to the new process, the Pentium Extreme Edition processors (model number 955 and 965 respectively) are now available at clock speeds of 3.46GHz to 3.73GHz and take advantage of a 1066MHz FSB for greater data throughput.

These top-of-the-line Intel processors had such convincing specifications that we simply had to put them to test and really see if their marketing holds up to reality; and which is what we did in this article. Putting the Pentium Extreme Edition 965 and 955 processors to their paces, we found some rather interesting results. The Pentium XE 955, the first Extreme Edition class processor to debut with the Presler core, while ousting its Pentium XE 840 predecessor by a good margin, it wasn't good enough to best the Athlon 64 X2 4800+, let alone the top FX models from AMD. Yet it retails at almost the same price as the Athlon 64 FX-60 for about US$1,000.

Lucky for Intel, their latest addition to the Pentium XE lineup, the 965 processor fared much better thanks to its extreme clock speeds. Reported to be available at certain e-tailors for close to US$1,100, the results we've obtained indicate that it might be more worthwhile to invest the extra hundred bucks for the Pentium XE 965 than the 955 model. While there were a couple of benchmarks and scenarios where the Pentium XE 965 excelled, AMD's processors still had the better overall balanced performance especially because they fared strongly in the gaming domain. With gaming being one of the foremost criteria for enthusiasts choosing an ultra high-end processor, we too placed emphasis in this performance category.

Fortunately, the Presler cores are rather good at overclocking and it is rather easy to hit 4GHz with good air-cooling. However, you would never achieve this out of the stock cooler provided by Intel, if you are living in the tropics. Heat output is still a concern despite the die shrink (not that AMD's FX-60 is running any cooler) because the transistor density and count have increased quite a bit. As such, we strongly advise in obtaining a good third-party cooler for your overclocking needs. To further facilitate overclockers, Intel has left the multiplier of their Pentium Extreme Editions unlocked, which is sort of a convenience offered after coughing up a huge sum for the processor. Having said this, you can opt for the Pentium XE 955 and easily overclock it to match a 965 processor. While that's not going to save you much, it is sufficient to cover the cost of obtaining a better CPU cooler, which would allow you to hit even higher speeds. Whichever the path, the choice is yours, but do ensure that you pair it only with an Intel 975X platform.

The Pentium XE 965 looks to be Intel's final attempt at peddling the current Netburst CPU architecture that debuted in 2001 with the original Pentium 4 as they are in the process of migrating production to the Core architecture unveiled at IDF earlier this year. Conroe is a Core architecture based desktop processor slated for retail from July 2006 or later and while roadmaps indicate no newer Presler processors till then, it remains to be seen if AMD's upcoming AM2 processor launch could spur Intel for another last-minute retaliation as we've seen in the past. 2006 is a very interesting year for processor and platform developments so stay tuned for more articles in the pipeline to unveil all that's taking shape.

Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.