Intel has revealed Rocket Lake’s full pricing, product mix, and new overclocking features ahead of the CPU’s retail launch on March 30. Rocket Lake CPUs are based on the Cypress Cove microarchitecture and they represent a breath of fresh air after nearly six years of Skylake-era refreshes.
The slide above covers the high-level improvements. To build Cypress Cove, Intel backported the 10nm Sunny Cove microarchitecture it first debuted with Ice Lake back in 2019. On mobile, Sunny Cove had trouble hitting high clock speed, but this was attributed to its process node more than its underlying architecture. The speeds, feeds, and prices on Intel’s new desktop chips broadly confirm this, though there are a few interesting differences between 10th and 11th Gen CPUs that we’ll discuss later in this article.
Rocket Lake fields fewer CPU cores in its highest-end SKU than 10th Gen Comet Lake did. We’re back to a maximum of eight cores and 16 threads in a top-end desktop system, down from 10C/20T. According to Intel, Rocket Lake should benefit from a 1.19x IPC uplift compared with Comet Lake. This should reduce or eliminate any performance differences between the Core i9-11900K and the Core i9-1900K.
The performance uplift from Rocket Lake will be more noticeable below the Core i9 because core counts in the Core i7 and Core i5 families aren’t changing. Intel isn’t launching new Pentium, Celeron, or Core i3 CPUs on Rocket Lake; the new iteration of these CPUs are referred to as “Comet Lake Refresh” and will be clock-bumped variants of already in-market silicon with slightly higher model numbers.
New CPU Means a New Chipset
Right along with a new CPU, we’re getting a new motherboard family. Those of you with Z490, Q470, and H470 boards with 10th Gen chips have an upgrade path to Rocket Lake regardless, while everyone else will need a new LGA1200 CPU and motherboard. Alder Lake on LGA1700 is expected to replace Rocket Lake so the new family of 500-series boards will likely have no upgrade path.
There are a few caveats to this slide. Wi-Fi 6E, 2.5Gbit Ethernet, and TB4 are supported,* not “supported.” Specifically, they’re supported if and only if your motherboard manufacturer bothered to pay for the additional hardware required to turn the feature on. Wi-Fi support requires an RF module while TB4 and 2.5Gbit ethernet require their own respective PHYs.
Benefits to the Z590 over Z490, specifically, include:
- Formal PCIe 4.0 support
- Doubled chipset bandwidth in some cases. Z590 + 11th Gen CPUs utilize an x8 PCIe 3.0 connection, while other CPUs use x4
- Native support for USB 3.2 Gen 2×2
Other benefits are conditional and depend on your motherboard vendor.
Memory Clock and Controller Changes
One major change to the memory subsystem is the introduction of DRAM controller ratios. AMD introduced this feature with Zen 2, when it allowed for the memory controller and Infinity Fabric to be clocked at 1/2 of DRAM clock. This allowed for theoretically higher memory OCs, though in practice running the fabric and memory controller at half clock often hurts performance more than faster RAM helps it.
Intel is only certifying DDR4-3200 clock speeds for 1:1 operation on its Core i9 family. Other CPUs are supposed to run at 1:1 ratios only at DDR4-2933 and above. Presumably, enthusiasts have the ability to override these presets and specify 1:1 gearing for DDR4-3200 and above.
One old feature making a return this generation is memory overclocking support on the H570 and B560 motherboard platforms, in addition to the enthusiast-oriented Z590. H510 will not support this functionality.
There are a variety of overclocking-friendly changes going in, including AVX-512 offsets and the option to disable it altogether. Extreme overclockers with subzero cooling will have the option to switch between two different memory multipliers without rebooting, allowing the OS to switch between them. This is being touted as a way to break overclocking records, but honestly, I’m a lot more interested in the possibility of using it to check DRAM stability in more mundane circumstances. Intel also now allows for per-core Hyper-Threading and individual core frequency adjustments on Z590 motherboards.
None of these improvements changes the fact that overclocking is in pretty sorry shape. Any headroom is going to come from CPUs like the Core i5-11600K, which has a relatively low all-core of 4.6GHz, compared with 4.8GHz for the Core i5-10600K.
TDP and Pricing
Here are Intel’s price sheets for RKL, covering the Core i9/i7/i5 families:
Some of the upper-end CPUs are getting price increases. The Core i9-11900K’s $539 official price is higher than the $500 on the Core i9-10900K, while the Core i7-11700K picks up a $25 price increase over official MSRP in 1K units for the Core i7-10700K. The Core i5-11600K, however, is about $10 cheaper than its predecessor. Intel will retain the higher prices that distinguish it from AMD at the top of the product stack, but there’s a wide range of Core i7 and Core i5 CPUs covering a number of price points.
If you do not care about onboard graphics, Intel’s “F” and “KF” CPUs offer a small discount in exchange for no integrated video. Readers eying lower-power TDP CPUs should be aware that making those chips work at lower power often requires forcing the motherboard to use Intel’s programmed TDP and Turbo values rather than the motherboard defaults. By default, a lot of 65W CPUs will use just as much power as their 125W counterparts.
Intel has reduced the base frequency of some RKL CPUs by 100-200MHz compared with their counterparts in 10th Gen, even as some boost frequencies are 100-200MHz higher. Intel derives its TDP values from its base clocks, so the reductions indicate that Rocket Lake consumes more power than Comet Lake at the same frequency. The Core i9-11900T has a base clock of 1.5GHz and a TDP of 35W, while the Core i9-10900T has a base clock of 1.9GHz and the same 35W TDP.
We’ll see how the Core i9-11900K compares against AMD, and how much additional performance Intel was able to wring out of 14nm, come March 30.
Now Read:
- Silicon Is the New Toilet Paper as Chip Shortage Hits Industrial Manufacturing
- DARPA Chooses Intel, Microsoft to Quest for Cryptography’s Holy Grail
- Intel Discontinues Overclocking Warranties as Hobby Continues to Die
No comments:
Post a Comment