Kentsfield quad-core performance previewed


Four cores, no waiting
— 2:00 PM on September 28, 2006

INTEL'S FIRST quad-core processor, code-named "Kentsfield," is now set to arrive in November, a few months earlier than originally anticipated. We already discussed the basic technology behind Kentsfield in our first dispatch from this year's Fall Intel Developer Forum, so I won't delve into it again here. The underlying premise is fairly simple, though: take a pair of Core 2 Duo chips, cram them together into a single package, and you have a "quad-core" processor that fits into one socket. This beast can then be dropped into a standard Core 2 Duo motherboard for mega-multiprocessing bliss under a single heat sink.

But how does such a thing perform? We have our first glimpse at an answer to that question today with a quick set of Kentsfield benchmark results compared directly to today's fastest desktop processor, the Core 2 Extreme X6800.

The numbers
These benchmark results come courtesy of an invite-only benchmarking session conducted on the site of Fall IDF, much like the session in which we obtained early Conroe scores at this past Spring IDF. As a result, we should preface these scores by saying that they are not the result of the sort of independent testing we prefer to conduct. I sat at the keyboard of one of these two machines as the benchmarks were run, but we were not involved in the setup of the systems, didn't have sufficient time to inspect their configurations extensively, and were not allowed to install and run our own tests on these machines. The tests and the settings, source media files, and circumstances under which they were run were all chosen by Intel. We will have our own independent review of a Kentsfield-based processor soon enough, and we can verify this performance for ourselves then. For now, these results should serve as a quick preview of what to expect from this quad-core processor in certain types of applications.

The two test systems used were essentially identical with the exception of the processor. The parts list included:

  • Intel DX975XBX2 "BadAxe 2" motherboard (pre-production) rev. 303 with 1066MHz FSB
  • eVGA GeForce 7950 GX2 1GB PCI-E graphics card with ForceWare 91.47 drivers
  • 2 x 1GB Corsair CM2X1024 DDR2 memory at 800MHz with 4-4-4-12 timings
  • Seagata Barracuda 320GB 7200.10 hard disk drive
  • Intel chipset INF update 7.2.2.1007
  • Intel Matrix Storage Console 5.5
  • Integrated audio driver 5.1.5143
  • Sony DVD-ROM DDU1615
  • Windows XP SP2 with DirectX 9.0c (August 2006 update)
In one of the two systems was a Core 2 Extreme X6800 processor with 4MB of L2 cache and dual cores running at 2.93GHz. In the other was a pre-production Kentsfield quad-core CPU with 8MB of total L2 cache and four cores running at a slightly slower 2.66GHz.

As you might expect, the relative performance of these two systems is likely to vary depending on the amount of threading in the application or applications being used. In apps with one or two threads, running by themselves, the dual-core Extreme X6800 is likely to be faster simply by virtue of its higher clock speed—not to mention arcane drags on performance in a Kenstfield system like increased bus contention and cache coherency management overhead. With that in mind, Intel chose benchmarks to show off Kentsfield that are at least partially multithreaded in a way that scales beyond two threads. Here's how the Core 2 Extreme X6800 and Kentsfield stack up in those tests:

Application

Core 2 Extreme X6800Kentsfield 2.66GHz
3DMark06 build 1.0.2 - Overall83048943
3DMark06 build 1.0.2 - CPU25424044
PCMark05 Pro build 1.1.0 - Overall77517576
PCMark05 Pro build 1.1.0 - CPU74798486
3ds max 8 SP280 s49 s
POV-Ray beta 151430.65 PPS2592.69 PPS
DivX 6.2.5 with XMPEG 5.03107 s76 s
Sony Vegas 7.0a build 1115382 s256 s

In tasks like media encoding and 3D rendering, Kentsfield's four cores really shine. The quad-core CPU isn't twice as fast, but it does scale up reasonably well in those applications.

3DMark06's CPU test is also widely multithreaded, and the quad-core processor crunches through it with ease, as well. Since the results of that CPU test are a component of the overall 3DMark score, Kentsfield comes out ahead of the Core 2 Extreme X6800 in 3DMark06. However, have a look at the detailed results, and you'll see a bit different story.

 
Core 2 Extreme X6800 (left) and Kentsfield (right)

The higher frequency dual-core X6800 processor winds up being faster in three of the four 3DMark graphics tests. The fourth one is simply a wash, with performance likely limited by the graphics card.

So what does it all mean?
This small handful of performance results nicely illustrates both the potential of quad-core systems and the challenges they face. In specific, highly parallel tasks where mulithreading is already well established like video encoding and 3D rendering, going from two cores to four produces some nice speedups. That won't be news to those of you who have followed our coverage of dual-socket, dual-core Opteron and Woodcrest systems.

However, the vast majority of desktop applications today aren't even dual-threaded, let alone widely multithreaded. Games, in particular, just aren't there yet. Nvidia and ATI have managed to make their graphics drivers multithreaded enough to extract some advantage from a dual-core system, but based on what we've seen in 3DMark's graphics test, going to four cores doesn't seem to confer any additional performance gains. Intel, AMD, and much of the rest of the industry is working to make the transition to truly multithreaded programming practices happen as easily and quickly as possible. Still, this transition won't happen overnight, and I wouldn't be shocked if we were still waiting for the first wave of really good multithreaded games at this time next year.

Nevertheless, this quick demo of Kenstfield does remind us that Intel's dual-chip package is a pretty neat trick. AMD has promised to deliver its "4x4" platform that will achieve four cores via two sockets before the end of the year, but it looks like Kentsfield may beat them to the punch—and quite possibly with superior performance—using the existing single-socket LGA775 motherboard infrastructure. Part of the reason Intel can manage this trick while AMD can't, we should note, is the fact that Intel is making its chips with a more advanced 65nm fab process, while AMD is still making 90nm chips that are larger and tend to run hotter. There are no doubt some lingering questions about whether four cores can share a single 1066MHz front-side bus and memory subsystem without running into major performance bottlenecks. Intel contends it isn't a big problem, though, and I'm largely persuaded. We will, of course, have to test a production Kentsfield versus AMD's dual-socket offering ourselves in order to know for sure. We should have the chance to do so in the coming months. 

Like what we're doing? Pay what you want to support TR and get nifty extra features.
Top contributors
1. GKey13 - $650 2. JohnC - $600 3. davidbowser - $501
4. cmpxchg - $500 5. DeadOfKnight - $400 6. danny e. - $375
7. the - $360 8. Ryszard - $351 9. rbattle - $350
10. Ryu Connor - $350
Intel's Xeon E5-2687W v3 processor reviewedHaswell-EP brings the hammer down 114
AMD's FX-8370E processor reviewedEight threads at 95W 147
Intel's Core i7-5960X processor reviewedHaswell Extreme cranks up the core count 198
AMD spills beans on Seattle's architecture, reference serverCache networks and coprocessors 46
Intel's Broadwell processor revealedThe 14-nm Core M aims to upend the tablet market 86
AMD's A10-7800 processor reviewed..and the A6-7400K, too 115
Android on x86: A quick look at Asus' Memo Pad ME176C tabletA 7" Bay Trail quad for $149 50
Core i7-4790K 'Devil's Canyon' overclocking revisitedCan a retail chip and a fancy MSI MPower mobo go further? 51

Tags: CPUs