It’s the worst kept secret in the industry: Intel’s next-generation Penryn killer, codenamed Nehalem is just around the corner. We’ve been seeing leaked benchmarks based on early silicon for months, and Nehalem’s Wikipedia page is already packed with unconfirmed specifications. All indications – and this is with more optimizations to come, mind you – is that Nehalem may be a bad mother worthy of having Isaac Hayes pound out a theme song for it.
A Bloomfield chip compared to a Core 2 Quad 6700
OK, we get it. It’s going to be fast, but just how difficult is it to build a Nehalem rig? What are the catches? Will the new motherboard and socket require some silly new BTX form factor?
To find out, we convinced one of our hardware contacts (who’ll remain unnamed) to let us into its lab so we could finally get our hands on the new chip. There, we were provided with the desktop version of Nehalem – called Bloomfield – and an Intel D58XSO “Smackover” board.
This particular sample was the 2.93GHz Bloomfield, which will put it in the mid-range of the three desktop Bloomfield chips expected to be due by the end of this year. One thing we immediately noticed when holding a Bloomfield next to a Core 2 Quad is how much larger the actual chip is. The new LGA1366 socket is also a bit larger, of course. As such, previous LGA775 heatsinks will not be compatible with the new CPU. Some coolers may possibly be adapted to work with Bloomfield with new mounting brackets but most will need new heat sinks.
The larger socket and “keep out zone” also means motherboard configurations will be different as well. On the Smackover board, for example, Intel chose to move the north bridge from the usual spot just under the CPU about three inches to the right. This shouldn’t be a problem since the memory controller now gets relocated from the chipset to the CPU, so there are far fewer wires to run from the north bridge.
The X58 motherboard (on the left) next to a LGA 775 mobo. Note that the socket orientation is rotated on the new board.
RAM configurations with Bloomfield will certainly confuse system builders who’ve been trained to think in terms of dual-channel memory pairings. Bloomfield’s onboard DDR3 controller supports three individual DDR3 channels. To get all three channels up and running you need three separate DIMMs. Board vendors will likely take two different routes.
The first is what Intel did on the Smackover, which has four DIMM slots. Two of the four slots are individual channels. The third and fourth slot share a channel. Normally you would run three slots filled for optimal performance. You can an optional fourth for expanded capacity but it may impact performance since the capacities in the third channel will likely exceed the first two. That won’t always be the case though. You could run 2GB DIMMs in slots 1 and 2, and 1GB DIMMS in 3 and 4 and all three would operate at full speed.
This board has 4 DIMM slots on three channels. Slots 3 and 4 share a channel. We tested memory benchmarks on single, dual, and triple channel configurations with Corsair DDR3 1333 memory.
With the larger chip, comes a larger heatsink and fan cooler.
The second option is what other more performance-oriented makers will take: populating the board with six DIMM slots. One engineering X58 sample board we saw on a nearby workbench did this and while it looked tight, it’s possible.
We agreed not to report benchmarks numbers since the BIOS, drivers and early chip silicon could bias people away from Bloomfield but we did want to see the impact of tri-channel DDR3 so we ran memory benchmarks against several different memory configurations.
The best performance bump was going from single to dual-channel but going from dual to triple didn’t seem to pay the same dividends. Remember, the caveat here is that more performance is likely to come as BIOS and board makers tweak for the new chip and RAM vendors tweak their SPDs. Our test, in fact, was with the DDR3 at 1333 speeds. At higher speeds of 1600, 1800 or higher, the tri-channel may pay off.
The good is news is that the memory controller is flexible. If you think that you’ll have to buy three sticks of DDR3 just to get the system to work, you don’t. We ran with single, dual, and tri-channel modes with no issues.
There are also some interesting overclocking features that will introduce technologies that Intel has talked about previously but we agreed not to reveal yet. Let’s just say it’s pretty cool stuff.
But what about SLI? If you don’t know by now, SLI capability for Bloomfield will only come through motherboard vendors who buy and integrate Nvidia’s nForce 200 chips onto the boards. Not all X58 board vendors will do this and none of the X58 boards we’ve seen have had the SLI chips. Even more troubling for Nvidia is that a recent Digitimes.com story quoted unnamed vendor sources as saying that few were event interested in even adopting the bridge chip for SLI capability. Board vendors we’ve talked to, however, say they’re taking a hard look at adopting it.