Six cores. That's how many are in Intel's ridiculous new Core i7-980x. MaximumPC takes us deep inside the world's fastest CPU, with full, mind-searing benchmarks.
Meet the world's fastest CPU. OK, so we just gave away the big reveal to our report before you even flipped one page, and without so much as the common courtesy of a spoiler alert. For that, we do not apologize, because it's not like you couldn't have guessed how this one would end up. After all, Intel's new 3.33GHz Core i7-980X builds on all the goodness of the ass-kicking quad-core 3.33GHz Core i7-975 Extreme Edition, but is smaller, cooler, and has an additional two cores under its heat spreader. With Hyper-Threading enabled, that's a cool 12 threads at the ready. How could anyone screw that one up?
With Hyper-Threading enabled, that's a cool 12 threads at the ready. How could anyone screw that one up?
In fact, Intel's Core i7-980X seems to be one of the most flawless launches we've seen from the company in some time. By flawless, we mean there are no contortionist acts, such as explaining to consumers that a new socket (LGA1156) will have the same CPU branding as an incompatible existing socket. Nor is there the head-scratcher of a very novel, yet very limp, integrated graphics chip in a CPU (Clarkdale), which, by the way, won't work in boards that lack graphics output ports.
With Core i7-980X, you update your BIOS, drop the chip in, and-voilà-you spend hours rocking a six-core high. Put simply, Core i7-980X is 24-ounces of prime-rib red meat for performance enthusiasts who really haven't had much to gnaw on since the original 3.2GHz Core i7-965 Extreme Edition came out two years ago.
So we're done, right? You don't need to read on? Sorry, there's still more to learn. If you want to know if your motherboard works with the new chip, what applications can really exploit the six cores, and how this bad boy performs, you'll have to keep reading.
Click to enlarge
What's in a Name?
We know that, by now, enthusiasts should be immune to Intel's confusing model numbers, but there's one thing that sticks in our craw about the Core i7-980X: Despite it being the world's first consumer x86 hexa-core, and despite it using the latest 32nm process, it's label is a mere five notches greater than the quad-core Core i7-975 Extreme Edition part it ostensibly replaces.
Surely, all the goodness of two more cores and a total 12 threads of computing would warrant a Core i9 designation, or at the very least, a much higher model number, right? No, Intel officials told us. The company said that, despite previous reports that it would call its hexa-core Core i9, Intel backed off when retailers and vendors complained of too many blasted brands. And as to why it isn't a 999X or 9900X, Intel said such gestures are unnecessary. The part is designed for enthusiasts and the folks who buy it will know that it's not a mere five clicks more than a Core i7-975.
Beneath the Surface
Fortunately, the chip is fairly simple to understand. It uses the new 32nm process that was introduced with the dual-core Core i5/Core i3 Clarkdale CPUs. For code-name junkies, that makes it part of the Westmere family-not part of the original 45nm Nehalem family. All six-cores reside on a single contiguous piece of silicon. Like the original Nehalem CPUs, each core has 2MB of L3 available to it, giving the CPU a total of 12MB of L3 cache. The cache is shared across all the individual cores, which means a single core can have up to 12MB of L3 cache if the other five cores are sleeping.
As is the case with all Extreme processors, the chip is fully unlocked letting you change multipliers as well as Turbo Mode ratios. Turbo Boost is present but not as pedal-to-the-firewall as the LGA1156 parts. The Core i7-980X will give you a Turbo Boost up to 133MHz if more than one core is active. With single-threaded apps, the CPU will Turbo Boost up to 266MHz. That's boring compared to the Core i7-870, which will boost from 2.93GHz to 3.6GHz, or about 733MHz. We'd be remiss, though, if we didn't point out that the Core i7-870 starts out at a much lower clock speed.
Keeping with Intel's tick-tock model, with ticks being little jumps and tocks being huge jumps, 980X is a tick. For the most part, besides the process shrink, there's very little that's changed from Nehalem to Westmere. The most notable new feature is the inclusion of advanced encryption instructions, which accelerate encryption.
Overall, Westmere is just a smaller, denser 45nm Nehalem. How much smaller? The Core i7-975 Extreme Edition weighed in at 731 million transistors and occupied 263mm2 of die space. Core i7-980X has 1.17 billion transistors but occupies just 248mm2 of die space.
Westmere will run its course until 2011 or 2012 when Intel introduces its Sandy Bridge CPUs. Where Westmere is a tick, Sandry Bridge will be a tock, introducing a new microarchitecture that will include advanced vector extensions as well as other enhancements. For entry-level CPUs, Sandy Bridge will also move the GPU core onto the die. Initial Sandy Bridge chips will be 32nm with a shrink to 22nm due soon after.
Pushing the Boundaries
Normally, a smaller process leads to enhanced overclocking and the same holds true for the Core i7-980X. With the original Core i7-965, we've never exceeded 4GHz on air. The D0-step Core i7-975 improved overclocking, but even there, we've never seen production machines exceed 4.2GHz reliably-and that's with water cooling. With the Core i7-980X, we went into the BIOS and dialed the base clock up until the processor was at 4GHz. From there, we had no stability issues and ran multiple benchmark runs without incident. Mind you, this was without tweaking core voltage for the CPU, the QPI, RAM, or other various knobs we could have turned to get more reliability. We even got the machine to POST at 4.5GHz on air cooling, but then it crashed. The verdict is that the Core i7-980X looks to be a wonderful overclocker.
Early Adopters Get the Respect
Let's face it: When Intel introduced its LGA1156 Lynnfield CPUs last year, every single person who bought into the Core i7 CPUs and LGA1366 motherboards had a panic attack. Would Intel, as some feared, abandon the LGA1366 platform altogether in favor of the more cost-conscious new socket? It's happened before. Think of Intel's short-lived Socket 423 and AMD's original Socket 940. With those, early adopters got one or two upgrades and then were left waving their DIMMs in the wind.
Fortunately, users who chose the early adopter route will be rewarded for once. The Core i7-980X is an LGA1366 CPU that should be drop-in compatible with nearly every LGA1366 motherboard. To keep things compatible, Intel even kept the official spec for the Core i7-980X to DDR3/1066 only. Even though the CPU is quite capable of supporting memory at far higher speeds, Intel said it didn't want to require motherboards makers to recertify boards for higher speeds of RAM. For what it's worth, we tested both the Bloomfield and Gulftown LGA1366 Core i7s at DDR3/1333.
You'll still have to update the BIOS before dropping in a Core i7-980X, but we haven't heard of any LGA1366 motherboards being incompatible with the new chip. That's quite an accomplishment for Intel, which has a history of burning people when new CPUs are launched. We don't want to rehash ancient history, but let's just say we're happy it worked out for early adopters for once.
Intel has long had a dilemma with its Extreme series of CPUs. Only folks with deep pockets actually purchased the Core i7-975-most consumers just bought the poor-boy Core i7-920 and overclocked that puppy up to the 3.7GHz+ range. There was simply very little incentive to buy the top-end part when the low-end part overclocked so well. That little cheat no longer works, though. To get a hexa-core chip today, you'll have to pay for an Extreme series. That's why we were actually surprised when Intel priced the Core i7-980X at $999. Sure, it's still too rich for most, but as the only game in town, we expected Intel to charge $1,500 for the CPU. At $999, the Core i7-980X is actually the same price as the Core i7-975 part that it will slowly replace.
When will Intel offer a friendlier-priced hexa-core? The company won't talk about unannounced product but several Internet rumor sites have reported that Intel has a hexa-core Core i7-970 in the $500 range on tap for the end of the year.
If You Build It, Will They Come?
If you think it's all sunshine and lollipops for hexa-core computing, it's not. As always, the problem is finding applications that will actually use the available threads. That was a problem with the original dual-cores and quad-cores; now with a hexa-core and Hyper-Threading, the situation hasn't improved much. The apps aren't nonexistent, but they're certainly not as prevalent as you would hope. That makes upgrading to the Core i7-980X something you'll want to think about first. Certainly, if you are a mega-multitasker, more cores don't hurt. But if your primary applications are single- or dual-threaded, the extra cores will just sit idle, so you'll need to seriously consider whether paying for a hexa-core makes sense.
A Close-Up Look at the Core i7-980X
All six cores of Intel's Core i7-980X share 12MB of L3 cache on the die. The 1.17 billion–transistor CPU also has two QPI connections but only one is enabled on consumer CPUs. The second QPI is use for multi-CPU Xeon configurations.
Even 100 Cores Won't Help Lazy Code
Multiple cores are only useful if there's software that takes advantage of them. Thus, we queried a couple leading software developers on where they saw the multicore sweet spot to be. Their answers shed interesting light on the quest for more threads.
Jeff Stephens, president of Bibble Labs: "Bibble 5 ‘supports' unlimited cores, and with fast enough disks and efficient OS-level scheduling, we can scale up to about 30 CPUs (performance benefits stop there, so 32 CPUs runs as fast as 30-right now). Without sounding glib, the reason no one else is doing this is because it's hard.… To scale beyond four or so threads, all aspects of a program must be built around parallel processing of huge amounts of data, efficient scheduling of processing tasks, and disk reads/writes to prevent starving CPUs of work to do by waiting for data, etc."
Paul Schmidt, president of Photodex: "In my opinion, more cores don't solve the biggest problem. The biggest problem is how the code is written-most code just isn't written to take advantage of more cores. I don't see that changing soon because writing code for multiple cores is hard and the development world is moving away from hard and toward easy. I think the trend is due to the same old brute-force single-core speed improvements that have been happening combined with how cheap computers are now. Why rewrite for more cores when you can wait a year and get a CPU that is another 20 percent faster?
AMD Responds with Phenom II X6
By now, we've pretty much become accustomed to AMD taking a back seat to Intel, particularly in matters of core count and performance. This year, however, it doesn't look like AMD fans will to wait as long for a six-core proc.
AMD expects to release its own hexa-core processor, the Phenom II X6, hot on the heels of Core i7-980X this spring. The chip will be a derivative of the company's Istanbul CPU that's been available for some time in Opteron-based servers. The chip is likely to have 6MB of L3 and will be compatible with AM3 sockets. It's not clear if the new chip will work in AM2+ boards, as we've been told that DDR3 will be mandatory for the new chip.
One other trick AMD may have up its sleeve-if a news report from Xbit Labs is correct -is its own spin on Turbo Boost. Using so-called Dynamic Speed Boost, Phenom II X6 processors may overclock individual cores when the full complement of cores is not in use.
AMD is also continuing to forge ahead with its Bulldozer core, which the company hopes will put it back on a competitive edge with Intel. Bulldozer's new microarchitecture will support advanced vector extensions and will be built on a 32nm process. Bulldozer is expected to be available in early 2011.
Let the Benchmarks Begin!
For our showdown, we decided that the new hexa-core has two primary competitors: the Core i7-975 Extreme Edition and the LGA1156-bound Core i7-870. We considered adding AMD's Phenom II X4 965 to the mix but the pricing ($185) and performance of that CPU puts it in a different class than the three Intel chips. When AMD's Phenom II X6 hexa-core hits in the near future, we'll certainly put it into the mix.
For our benchmarks, we used both older and newer benchmarks to stretch the Core i7-980X. We used both synthetic and real-world applications for video editing, encoding, 3D rendering, and memory tests, along with a handful of gaming benchmarks. Be advised, when we review a CPU, we set resolutions fairly low in order to remove the GPU from the equation.
The verdict: We have no problem proclaiming the Core i7-980X as the world's fastest. Obviously, it shined the brightest in our multithreaded 3D-rendering benchmarks, where its performance surmounted the already ludicrously fast Core i7-975 by 37 to 55 percent. Encoding also gave us a healthy 25 percent performance boost. Likewise, video editing saw the hexa-core achieve anywhere from 10 to 25 percent performance boosts. In applications where multithreading is minimal, the Core i7-980X was usually tied with the similarly clocked Core i7-975. We do suspect that the larger L3 cache of the Core i-980X paid off dividends in several of our gaming benchmarks.
One figure we couldn't quite square was the memory performance of the Core i7-980X. We expected its memory bandwidth in the synthetic tests to be equal to the Core i7-975's, but the hexa-core was at a disadvantage. The lower memory bandwidth didn't seem to hurt in the other benchmarks, though.
In the final analysis, this is a CPU that turns in performance that is, at its worst, equivalent to the Core i7-975 it replaces. At its best, the i7-980X offers up to 50 percent more performance than its closest competitor. That's pretty much unprecedented and certainly helps the Core i7-980X earn its crown as the new performance king.
Best scores are bolded. We tested both LGA1366 CPUs using an Asus P6X58D Premium motherboard with 6GB of Corsair DDR3/1333, an EVGA GeForce GTX 280, and 64-bit Windows 7 Professional. The LGA1156 CPU was tested with a Gigabyte P55A-UD6 motherboard, 8GB of Corsair DDR3/1333, an EVGA GeForce GTX 280, and 64-bit Windows 7 Professional. Both configurations used a 150GB Western Digital Raptor hard drive.
Maximum PC brings you the latest in PC news, reviews, and how-tos.