Introduction — continued
The controller will perform best when dealing with symmetrical memory configurations between the two memory channels, using what NVIDIA calls fine-grained interleaving. However, the nForce4 SLI also supports a coarse-grained interleaving mode that will operate in 128 bits with asymmetrical memory configs (where, say, one has a 256MB DIMM in one slot in channel A and a 512MB DIMM in another slot on channel B.) Performance isn't quite as good, but the company says it's better than dropping into 64-bit mode.
NVIDIA has coined a pair of marketing names and attached them to the nForce4 Intel Edition's memory controller. The first of those is QuickSync, a design feature intended to provide low-latency memory access even when the CPU's front-side bus is running at a different clock speed than system memory, as is often the case in today's Pentium-based PCs. Intel has demonstrated that it can achieve relatively low memory access latencies with its 915/925 memory controller when the front-side bus is running at 800MHz and talking to dual channels of DDR2 533 memory, so naturally NVIDIA would want to match this capability. QuickSync is NVIDIA's attempt to do so.
The second marketing name attached to the nF4 Intel Edition's memory controller will bring back memories for those familiar with the original nForce and nForce2 chipsets: DASP, for Dynamic Adaptive Speculative Preprocessor. DASP is a mechanism that watches memory accesses and speculatively pulls data into a buffer on the memory controller, and the nF4 Intel Edition is outfitted with version 3.0 of DASP. The potential trouble with that is simple: Pentium 4 processors (and derivatives) already have a large L2 cache into which they prefetch data themselves. NVIDIA says it took the P4's prefetcher into account in designing DASP 3.0 and built in mechanisms to give the CPU priority when memory bandwidth is at a premium. Even with the P4's own aggressive prefetcher, NVIDIA claims a performance increase between 2% and 4% from DASP 3.0. I tried to shake loose the exact size of the on-chip DASP buffer from NVIDIA's reps, but they wouldn't budge, other than to point out that the chip is pretty big.
The remaining PCI-E lanes, of course, can be used as needed for additional PCI-E slots and devices.
The big news with the Intel Edition is the addition of a tantalizing new option: RAID 5 arrays. Many folks consider RAID 5 the best possible combination of economy, performance, and data integrity, because RAID 5 can (for example) offer 240GB of capacity, with striping and redundancy, from only three 120GB drives. This magical combination is achieved by storing parity data on the drives, from which the contents of dead hard drive can be reconstructed. The nF4 Intel Edition can handle RAID 5 arrays as large as eight drives, but unlike high-end RAID controllers, the parity calculations must be performed in real time by the host CPU. That means RAID 5 will involve some CPU overhead on the nF4 Intel Edition, although today's fast processors may not be terribly burdened by this task.
I'm still a big proponent of RAID 0+1 arrays as the ultimate in performance and redundancy (or better yet, RAID 10), but those configs require four drives, so RAID 5 may become a popular option for enthusiasts. At least it's not borderline suicidal, like RAID 0.
NVIDIA says MediaShield should get a new, consumer-friendly interface, complete with simple wizards for array setup, in the June-ish timeframe. This interface will come for all nForce4 products in the form of a software update. Also, the new RAID capabilities in the Intel Edition should find their way to the AMD side of the house in a future chipset update.
These features should put NVIDIA more or less on par with Intel, who is a moving target right now; the introduction of its 945/955 chipsets is imminent. At present, NVIDIA has a leg up on Intel's 915/925 chipsets, because those chipsets officially only support DDR2 speeds up to 533MHz.
NVIDIA says the nForce4 SLI Intel Edition is the first of a family of chipsets for Intel processors, just as we've seen on the AMD side.
The joy of chipset launches
As with most chipset launches, we are dealing with pre-release hardware, and not everything is as stable and polished as we'd like. The board we're testing is an NVIDIA reference design, and it wasn't 100% crash-free as we used it. Since this isn't a production board, we'll chalk it up to teething problems, but the fact that it's not 100% stable is worth noting. NVIDIA says to expect production mobos based on this chipset to be available later in April, and the big three motherboard makers are apparently already sampling products.
One quirk of NVIDIA's reference board is a lack of support for the Pentium 4 600 series' new C1E halt state, SpeedStep, and TM2 thermal throttlingall of which use the same transistors to do the same basic thing in response to different inputs. I sorted through these various functions on this page of my Pentium 4 600-series review, in case you missed it. Without the C1E halt state or SpeedStep, a Pentium 4-based system tends to consume about 35W more power at idle, with the corresponding additional heat and noise output. I mention this problem because if the reference design doesn't support these power management features, it's possible that the first wave of production boards based on it won't, either. NVIDIA claims the nForce4 SLI Intel Edition can support these things with a BIOS update, though.
A more interesting question is whether the nForce4 SLI Intel Edition will work properly with Intel's new dual-core processors. Intel itself has prepared new chipsets for its dual-core products, after all. The folks at NVIDIA say dual-core support is built into the nForce4 SLI, but that actual support for dual-core CPUs will depend on the motherboard. I hope that's the case, but as you probably know, Intel has really moved quickly to bring its dual-core products to market. We will be watching the question of dual-core support in production nForce4 Intel Edition motherboards closely.
|G.Skill KM560 MX keyboard drops the numpad||3|
|Rumor: Acer Triton 700 may use an unreleased Pascal GPU||1|
|Silverstone Vital VT02 could hold a Core i7 in under two liters||0|
|Galax and KFA2 induct the GTX 1080 Ti into the Hall of Fame||17|
|Acer's Aspire GX-281 lineup brings Ryzen to the masses||10|
|Deals of the week: discounts on CPUs, mobos, and more||7|
|Asetek gets $600,000 from Cooler Master in AIO cooler patent spat||10|
|Acer Predator Triton and Helios laptops are ready for serious play||8|
|Intel enjoys healthy revenue and profits for Q1 2017||27|