Where our samples are not retail samples, we cannot comment on the shipping container retail buyers will get. What we can say is the limited-edition reviewer’s box is pretty. Big and a little less than optimal from a protection point of view… but pretty.
The same is true of accessories…as we doubt Intel will be including a free ASUS Republic of Gamers Maximus XIII Hero motherboard with the purchase of an i9. That certainly would be a way to gain sales, but highly impractical to say the least.
As with the last generation of Intel consumer orientated processors the best way to begin to describe the 11th gen Core series is to hit that angry elephant in the room head on. The 11th generation of Intel Core I processors is 14nm based. Not 10nm. This is disappointing. Not because the 14nm node process is bad, and only new and shiny tech is good. No. The reason this is disappointing is that users still are not going to get Sunny Cove based processors for another generation. Instead, Intel had to backport Sunny Cove to a 14nm node design and dubbed this stop-gap architecture “Cypress Cove”.
While we could care less about the node size a given CPU is built upon, it does matter when the architecture was designed around a smaller node process. A CPU, and any computer component, design team know precisely how much ‘space’ they have to work with. Be it transistors per square millimeter, or thermal dissipation per square millimeter, or how many square millimeters in total they have to work with… they are given the parameters and then they go about maximizing the transistor/NAND/widgets that they can fit inside that envelope.
That is how Sunny Cove was originally planned, and designed. It was meant to be based on Intel’ 10nm fab process and fit the Socket 1200 specifications as laid down in the 400-series chipset release. The last bit is the real reason we are not seeing the Socket 1700 make an early debut with the 11th generation’s Core I series (instead of its planned 12th gen release date)… because Intel actually does try and keep its promises. So, on top of everything else Intel also needed their 11th generation processor to be ‘backwards’ compatible with existing 400-series motherboards and not require a 500-series based motherboard. Yes, this further limited the options the Cypress Cove team had to work with… as this ‘future proofing’ was a promised feature of the Z490 and 10th generation Core i releases. One Intel intended to keep.
What all this means is that the actual die size could not be changed. Maximum power draw limitations of the S1200 design could not be changed. Nothing could be changed except reworking the architecture to ‘work’ with the 14nm node process. To grasp how big a deal this undertaking was… Intel’s 10nm was/is going to be an estimated 100.8 MTr / mm2, whereas Intel’s 14nm is 37.5 MTr / mm2. Huge difference. Such a huge difference it would be like trying to stuff 12 clowns into a sub-compact car.
In other words, the design team of “Cypress Cove” had to take out ye olde scissors and start cutting features. Snip. Due to thermal limitations the planned frequency increases were cut back (or eliminated altogether). Snip. Due to shear size of the backported monolithic architecture…. gone was the idea of anything more than an 8-core design physically fitting inside the Socket 1200 specifications. Snip. There goes the idea of a massively upgraded Integrated Memory Controller. Snip. There goes the idea of game changingly massive IPC improvements with even lower latency.
Thankfully, that is basically ‘all’ that had to be cut, and the design team was indeed able to leave a lot of meat on the Sunny Cove bones. For example, the IMC may not be a fire-breathing beast compared to Comet Lake’s but it still has been improved. DDR4-3200, and thus parity with AMD Ryzen 5000-series’ IMC, is now natively supported instead of DDR4-2933 (i.e. its highest ‘non-overclocked’ memory frequency). More impressive still, real-time memory tuning is now possible. Yes, no more tediously long “change a memory setting, reboot, hope it POSTS, test, change another setting, reboot, hope it POSTS, test” memory overclocking process. That is a feature that was so long overdue it is not even funny.
Moving on. L1 cache has been increased from 64KB per core to 80KB. For the Core i5 this means more L1 cache… but less total for the Core i9 when compared with its predecessor. L2 cache is now doubled from the 10th generations 256KB per core to 512KB. So even with two fewer cores the Core i9 has more (4MB vs 2.5MB) and so too does the Core i5 (480KB vs 384KB). Cypress Cove design team even found room to make the L3 stay the same at 2MB per core. These are Good Things ™. Good Things that do improve overall performance.
Further helping to keep too many Intel engineers and managers from being tarred and feathered is the fact that Rocket Lake was able to keep, and even improve, the high clock frequencies of the 10th series. For example, while it is indeed true the Core i9-11900K comes with a dual core boost clock of upward of 5.3Ghz (5.2 if the CPU is above 70c and Thermal Velocity Boost cannot kick in) Intel has added in a new Adaptive Boost Technology (ABT for short). ABT allows all the Core i9’s eight cores to run at (up to) a whopping 5.1GHz… as long as TDP and thermals are within limits. Of course, it will not just go ‘bam here is an all core 5.1GHz frequency setting’. It will start by going 100Mhz above Turbo Boost 2’s max setting (for example with all 8 cores that means 4.7Ghz) and check to make sure the power draw is not above power or thermal budgets. Then it will add another 100Mhz. Check the results, and repeat if things are still looking good. Needless to say, the more cooling you throw at the Core i9, the faster it will be.
Sadly, Adaptive Boost tech is only included with the Core i9 processor(s) line. Core i7’s top out at Turbo Boost Max 3.0 and the Core i5’s don’t even get Turbo Boost Max 3. Put another way, beyond the Core i9 the rest of the lower numbered chip’s ‘auto overclocking’ features are the same as the last gen. With that said, the Core i5-11600K can boost to 4.9Ghz and has had its PL2 set to a whopping 251 watts (and its Tau timer further relaxed to an ‘optional’ 56 seconds from previous gen’s 28 seconds)… just like the Core i9-11900K comes with. Now that is impressive. A lot of power, and thus heat production, from a Core i5… but impressive never the less. Also in the positive column, is even with all the automatic overclocking features included there is still a bit of room left over for manual overclocking. With not much difficulty (at all) we hit the same 5.3Ghz on the Core i9-11900K as we did the Core i9-10900K, and the Core i5-11600K was a breeze to get to 5.0GHz all core frequencies.
Needless to say, these are all Good Things that Intel’s miracle workers managed to add in to a backported processor design. Especially when you consider that, backported or not, the 14nm variant of Sunny Cove is more efficient and does come with better IPC than the, Cascade Lake architecture, Comet Lake processors used. Intel claims upwards of 19 percent IPC gains. In testing, it is not quite that good… but it is better. In testing the IPC gain ranges from low single to low double-digit improvement depending on scenario encountered (and the 19 percent average obviously included AVX decoding scenarios which are niche scenarios to say the least). This real-world improvement is more than “good enough” to easily regain the IPC crown from AMD and their Zen 3 architecture. We can just imagine what Sunny Cove will do when it is paired with a 10nm (or 7nm) node process.
Also firmly in the Good Thing™ category, is the integrated Graphics Processing Unit has gotten a much needed upgrade. Thanks to Intel Xe graphics actually being ready for prime time(ish) the UHD 750 series iGPU gets a 33% percent bump in Execution units (32 vs 24 of the UHD 630 series used in the 10th generation Core I processors), a frequency boost of 100MHz (1.3Ghz vs 1.2GHz), and each Xe Execution Unit is simply more efficient than the older EU’s Intel used. The end result? Upwards of 50 percent more graphic processing power. Which is enough of a boost for it actually to maybe good enough for some light gaming at lower resolutions. PC gaming on iGPU’s jokes aside, as a QuickSync decoding platform its native support for AV1 codec, as well as integrated HDMI 2.0 and DisplayPort High Bit Rate 3 (HBR3) support will all be impressive for certain scenarios. Generally speaking, this is a much-needed upgrade over an iGPU that was first released in 2017.
On top of all these tangible improvements the Cypress Cove also kept the largest promise of them all: PCIe 4.0. Yes, Intel has finally achieved parity with AMD’s processors. Up until the release of Nvidia’s RTX 3000-series the lack of PCIe 4.0 abilities really was ‘meh’ for anyone not rocking the latest gen NVMe solid state drives. However, with NVMe PCIe 4.0 x4 drives and PCIe 4.0 x16 videos cards now both widely available this is indeed a Big Thing. A Big Good Thing™. Furthermore the 11th gen Intel Core I series not only has 16 of these PCIe 4.0 lanes set aside for Add In Cards, it also has another 4 PCIe lanes set aside for PCIe storage. Put another way, Cypress Cove not only doubles the bandwidth of the PCIe bus from the last gen’s 16 lanes… it then adds in another four lanes of PCIe 4.0 goodness.
On paper this still appears to be four PICe lanes short of parity with AMD’s Ryzen 5000-series… but the reality is a bit more complicated than that. Yes, AMD technically has 24 PCIe 4.0 lane abilities baked into their consumer orientated processors. In reality it is more like a 16+4+4 configuration, with four set aside for the Infinity Fabric (and thus not ‘useable’ by lowly Add In Cards or Storage). So, for all intents and purposes Intel’s 16+4 is the same as AMD’s 16+4(+4). Kinda-sorta.
We say kind of the same as each CPU architecture’s PCIe lane configuration has strengths and weaknesses. In AMD the ‘southbridge’ (and memory channels) are connected to the CPU via a total four PCIe 4.0 lanes (via the Scalable Data Fabric portion of the umbrella ‘Infinity Fabric’ term). In Intel the ‘PCH’ (aka southbridge in old school / AMD speak) and memory channels are connected to the CPU via eight PCIe 3.0 lanes. Half the speed per lane… but twice the number. Pick your poison based on what you need the most.
Either way, this is another improvement Sunny Cove 0.5…. err… Cypress Cove brings to the table. Yes, this means Intel has finally doubled its Direct Media Interface bus to eight PCIe 3.0 lanes. This is a much needed, and arguably long overdue upgrade. Much needed when you consider that Intel has kept all its 500-series chipset promises.
Put simply, the 500 series chipset (especially the Z590 variant) is now choke full of goodness. Forget terms like ‘parity’ when comparing to AMD. It out and out smokes AMD’s X570 chipset. It smokes it because in addition to all the integrated features of the 400-series Intel has upped the game with the addition of very, very cool (and useful) goodies. For example, with a Z590 motherboard, power-users need not worry over ASUS/ASRock/Gigabyte/MSI/etc. using (ugh) Aquantia’s 2.5GbE network interface controller… as 2.5GbE is baked right into the PCH. WiFi users also are shown the love and the 500-series has WiFi 6 wireless networking abilities baked right in. Arguably even more important, and pouring salt directly into the wound, is Intel also has baked ThunderBolt 4 and USB 3.2 gen 2×2 into the 500 series chipset. Yes. three ports with 20Gbps goodness is now on tap. No Add In Cards needed. Counteracting these improvements is the fact that the 500-series chipset is still PCIe 3.0 based. Yes, Intel has upped it to 24 lanes of PCIe 3.0 goodness… but AMD’s latest 500-series chipset offers 20 lanes of PCIe 4.0. Once again more but slower vs. faster but fewer. That really is the only weakness left in Team Blue’s chipset.
Overall, we are actually impressed with what Intel’s design team have accomplished. So while yes, we sincerely wish the 11th generation was built on a 10nm node process… the end results should be rather impressive for the average buyer. Impressive enough that it makes a rather good argument in its favor over opting for either an older Z490+10th gen or Team Red based system. Now if only Intel would get either their 10nm or 7nm node up and running… and give us a true HEDT offering (and thus competition to ThreadRipper) we would be truly happy.