By the end of this year, AMD will introduce Phenom and Intel will introduce Penryn - its first 45nm desktop CPU.

For a while there we wondered how Intel would introduce Penryn into its lineup, but as of Intel's last roadmap update we now know. On top of that, we managed to get our hands on a dual-core Penryn so now we know exactly how fast the new core is going to be. Combine the two datapoints and there's a wealth of knowledge to share. So gather 'round the fire, it's time to benchmark CPUs.

Recapping Penryn

Penryn is more than just a die-shrunk Conroe, but it isn't anywhere near as revolutionary as what Conroe was to its predecessor. The Conroe-Penryn evolution is one cycle of Intel's tick-tock model, Penryn's successor (Nehalem) will be the next revolutionary core you see out of Intel (at least that's what we're expecting).

Penryn is built on Intel's 45nm process, thus dropping power consumption increasing switching speed and allowing Intel to cram more transistors on a die (107mm^2 for dual-core Penryn vs. 143mm^2 for Conroe).

A dual-core Penryn die  

A dual-core Conroe die

Intel lists the following 9 enhancements as what's new in the Penryn family:

The faster divider and super shuffle engine will increase performance in very specific applications and won't be broad increasers of performance. The larger caches obviously will increase performance, but prior to today we weren't sure by how much.

SSE4 support can bring about tremendous performance improvements, but it requires software optimization. In our initial Penryn Preview we reported an over 100% increase in performance in SSE4 optimized DivX encoding for a quad-core Penryn at 3.33GHz vs. a quad-core Kentsfield at 2.93GHz.

We detailed all of the Penryn enhancements and the markets they impact in our original Penryn article. Keep in mind that things like Penryn's Deep Power Down Technology and higher bus speeds may not apply to all Penryn cores (the former is only in Mobile Penryn while the latter will appear on servers first).

Never Take Sides Against the Family

Penryn is the overall family name but there are two cores in particular that we'll be talking about today: Wolfdale and Yorkfield.

Wolfdale is a dual-core desktop Penryn core with a 6MB L2 cache shared between the two cores. Yorkfield is two Wolfdales on the same package, giving the quad-core CPU a total of 12MB of L2 cache.

Wolfdale and Yorkfield are architecturally identical, so performance per core is no different between the processors - Yorkfield simply has more cores.

Penryn's Launch Schedule

The launch will work like this: late this year (Q4), Intel will introduce a a new Extreme Edition processor based on Yorkfield (quad-core Penryn). Given that most Extreme Edition chips start out with a $999 price tag, that's where we'd expect the first quad-core Penryn to fall as well.

More affordable Penryn won't arrive until Q1 2008, which is great news for AMD. In the beginning of next year, Intel will introduce both mainstream Yorkfield and Wolfdale (dual-core Penryn) parts. We'd expect Wolfdale based chips to run at between 2.33GHz and 3.00GHz, while Yorkfield will probably top out at 3.33GHz.

In Q2 2008, Intel will push Penryn even further down the line and introduce a Wolfdale based E4xxx successor. This version of Wolfdale will have a 1066MHz FSB and drops support for Intel Virtualization Technology and TXT.

The Chip

Intel is keeping its few remaining cards close at hand, and thus it is only sampling lowly 2.33GHz Wolfdale/Yorkfield chips. It's not an issue of getting the 45nm process to clock high enough (as you'll see from our overclocking tests), so we'd guess that Intel doesn't want to have the performance of its top of the line 3.33GHz Yorkfield leaked until after Phenom hits.

Wolfdale's underside - we have to hide the other side so Intel doesn't murder our mole in its sleep

Conroe's belly for comparison

The 2.33GHz Wolfdale we're looking at today is best compared to the Core 2 Duo E6550 (2.33GHz Conroe), and that's exactly what we're doing.

Our Wolfdale is also not the final stepping that will be shipping next year; we've got A0 silicon and A1 should be ready in the coming weeks. Despite the not-final state of the silicon, we had absolutely no problems with the chip in our test bed. We tossed it in our Gigabyte GA-P35C-DS3R test bed and ran through a subset of our test suite without so much as a BIOS update. We are waiting on the first Penryn-tuned motherboards and BIOSes to arrive, so performance may increase from where we're at today.

The Test

Wolfdale vs. Conroe Performance

We had heard rumors of Intel introducing a faster, lower latency L2 cache in Wolfdale and it appears to be true:

Not only is Wolfdale's L2 cache larger, but it also happens to be slightly faster than its predecessor. Intel has shaved off a single clock cycle from Wolfdale's L2 access time; we're already off to a good start.

If you want a quick glance at what Wolfdale will offer, the chart below will give you just that. We've taken some of our normal CPU benchmarks and ran them on a 2.33GHz Conroe as well as our 2.33GHz Wolfdale, the chart below illustrates percent performance improvement of Wolfdale over Conroe at the same clock speed:

Let's point out the zeros first: SYSMark, iTunes and Oblivion all showed no performance increase from Conroe to Wolfdale. Not all applications will benefit dramatically from the improved cache or architectural improvements and these are examples of some.

The DivX 6.6 test shows a particularly impressive 10.5% increase in performance, especially when you keep in mind that we are running the same DivX test we always run and not an SSE4 optimized benchmark. If you'll remember back to our Intel-sanctioned Penryn preview, with SSE4 enabled Penryn's DivX performance skyrocketed. But this test here shows us that even without SSE4 optimizations, Wolfdale is a healthy 10% faster than Conroe. Windows Media Encoder 9 saw a 5.4% increase in performance, which is still tangible.

Wolfdale also seems to do quite well in 3D rendering apps, giving us 6.7% better performance in 3dsmax 9 and a similar boost in Lightwave. Cinebench performance improved even further at 9.1%.

Gaming performance is a bit of a mixed bag; we saw everything from Oblivion's 0.4% performance improvement to 8.5% under Lost Planet. Wolfdale is good for gaming, but the degree is very title dependent.

On average, Wolfdale ends up being just under 5% faster clock-for-clock than Conroe. Definitely not an earth shattering improvement, but an improvement nonetheless. Focusing in on specific benchmarks, Wolfdale can look even more impressive. Without taking SSE4 performance into account as we don't know how widespread SSE4 applications will be upon its arrival, Wolfdale will simply make competing more difficult for AMD's Phenom, but not impossible.

Power Consumption

Obviously power consumption should go down thanks to Intel's cooler 45nm process. At idle (sitting at the desktop with no disk queue), the difference between our Conroe and Wolfdale systems is relatively small - a savings of only 3%:

Under load however (running our WME9 test) the total system power consumption gap increases to 12.6W:

Wolfdale decreases total system power consumption by around 10% compared to Conroe.

Overclocking

In order to make sure that Intel wasn't having yield issues at 45nm, we wanted to see how far we could push our test chip. With the only multipliers available being 6.0x and 7.0x, we had to rely on FSB overclocking to get us to where we wanted to go. But we made things a little more difficult: we didn't increase CPU core voltage at all.

According to CPU-Z, our Wolfdale sample ran at 1.168V at 2.33GHz. We left the voltage unchanged and went in pursuit of how high of a stable overclock we could get otherwise.

400MHz x 7.0 wasn't a problem at stock voltage:

Neither was 450MHz x 7.0:

The best we could manage without adjusting the core voltage was 460MHz x 7.0, or 3.22GHz:

While this test tells us nothing of overall clock speed headroom of Wolfdale and Penryn in general, it does give us something to hope for.

The Rest of the Tests

If you want the exact test results that we used to generate the Conroe vs. Penryn chart a few pages back, those graphs follow without interruption:

Final Words

After Intel's price cuts we mentioned that despite Penryn's imminent launch, CPUs had gotten too affordable to pass up building a system now if you needed. Now that we have a general idea of clock-for-clock performance differences between Conroe and Wolfdale, we're not nearly as worried about recommending that you build systems today as we once were. There's no doubt that Wolfdale is faster clock-for-clock, but keep in mind that you won't see Wolfdale until Q1 of next year and the performance advantage simply isn't great enough to justify delaying a purchase by 6+ months if you need a system now.

AMD seems quite confident that Phenom will be able to compete with Conroe/Kentsfield; if the race is really close between those cores, Penryn could be exactly what Intel needs to remain technically ahead in performance. If Phenom is significantly faster than Conroe/Kentsfield, then it doesn't look like Penryn will be able to save Intel. We should know for sure which scenario will pan out in the not too distant future.

We've got more Penryn coverage coming, including a look at Yorkfield vs. Kentsfield so stay tuned.