Houston, we have BIOS Problems: The Rocket Lake [P]review
Article updated 4/3/21 : Miscellaneous grammar and formatting edits, corrected a few sentences for accuracy. Added 2nd video for demonstration of UHD 750 Gameplay + Recording performance.
Today Intel launches their 11th generation CPUs, codenamed Rocket Lake. It is a bit of a “Frankenstein” architecture – being comprised of Ice Lake-based CPU cores and Tiger Lake gen GPU cores, both backported from a 10nm manufacturing process to 14nm. As a result of backporting, RKL’s CPU cores are larger than Comet Lake’s – which is the primary reason that Rocket Lake is limited to 8 CPU cores (vs Comet Lake’s 10 CPU cores). With the promise of up to 19% increased IPC, many Intel users – including myself – were excited to see what it might bring. Did Intel deliver? Keep reading.
We’d like to take a second and thank Intel for sending us the CPUs and NVMe drive for this review.
First we’ll take a look at Cinebench. In its single core benchmark, Intel’s i9-11900K scores 1633 points, 19.4% higher than the 1367 points scored by the “old” i9-10900K. This lines up almost exactly with Intel’s IPC improvement claims.
In Cinebench’s multi-core benchmark, the last generation i9-10900K had a slight edge – beating the i9-11900K by 1.6% with a score of 15677 vs 15428 for Rocket Lake.
For CPU-Z, I only tested it on the 11900K, as it already had comparative results for the i9-10850K which is virtually the same level of performance as the 10900K.
Single-core performance scored 711.1 for the i9-11900K, 27.6% higher than the 557 points listed for the i9-10850K. In multi-core performance, however, the i9-10850K beat the Rocket Lake CPU by 8.2% – 7200 vs 6652.1
Testing in PassMark’s CPUMark shows an overall lead of 11.8% for Rocket Lake – 26337 for the 11900K vs 23559 for the 10900K. Encryption was an especially strong win for Rocket Lake, with it nearly doubling performance vs Comet Lake with a score of 18046 vs 9371. SSE was nominally behind Comet Lake’s performance, and in sorting Comet Lake lead by 22% with 49503 vs 40439!
MemoryMark was not as kind to the 11900K, with Comet Lake beating it by 9.5% overall (4063 vs 3711). Memory Latency was especially brutal to this score, being 35% slower than Comet Lake (23 vs 31 – lower is better)
With image editing, Rocket Lake showed a huge performance increase – scoring 268088 vs 163212. It completed this test in 19.8 seconds compared to 32.6 seconds for the 10th generation CPU.
With encoding, the i9-11900K fell slightly behind the 10th generation CPU – completing the task in 21.9 seconds vs 19.8 seconds for the i9-10900K.
OpenCL performance was on par between CPUs, with 29801 KSamples/sec for the 11th generation CPU and 29835 KSamples/sec on the 10th generation CPU.
For heavy multitasking scenarios, Rocket Lake performed ~6% better than it’s predecessor. This surprised me, as I had (incorrectly) thought that in these scenarios Comet Lake’s 10 cores would perform better.
Gaming: Welcome to Buggy BIOS Hell
You may notice that I have declined to state the make and model of the motherboards I tested for this preview – and that I’m calling it a preview instead of a review. This is because all BIOSes which I have tested are currently plagued with problems.
Some examples of said issues: Using the Z590 board, every time I installed the i9-11900K the Z590 would spit out error codes – 4c being the most common. Each time I would have to unplug my motherboard and remove the bios battery and then try again after a few minutes – I’d usually have to do that twice.
Some games would crash, usually during loading, if RivaTuner (used to display overlays for MSI Afterburner and CapframeX) was running at the same time using both Z490 and Z590 motherboards. These crashes occurred most frequently in Hitman 2 and Death Stranding. Other titles had extreme performance deficits running the oldest BIOS versions. For example, take this benchmark of CS:GO – this was performance from Rocket Lake running on the same Z590 motherboard. We’re literally looking at a 42.3% performance increase from one BIOS version to another.
But even with this newer BIOS Update, in my testing the i9-11900K lost to the 10900K by 4%. Another reviewer whose BIOS supports a newer microcode version showed me his results – which show the 11900K winning in CS:GO by 5%. This is an extreme amount of variance between BIOS/Microcode updates.
It appears that these microcode updates significantly improve memory latency. The results below on the left are from a board using the latest 0x3c microcode, the results from the right are from a board using the older 0x34 microcode. Between these two boards, there is a 12.5ms variance in memory latency – that’s roughly 20% slower!
Unfortunately, I wasn’t able to finish my testing with the latest BIOS update available. After I finished benchmarking CS:GO, I turned the computer off to make dinner. When I attempted to restart the computer, it gave 5D error codes. I reseated the RAM, Cooler, and CPU – and cleared the CMOS. This time, it gave 4C error codes which persisted after further troubleshooting – these errors persist whenever I use DIMM slots A1 or A2.
As a result, I can only run the motherboard in single channel memory mode (making it unfit for testing). Gavin Bonshor, from Anandtech, mentioned to me that he initially experienced the same issue (on all DIMMs) after updating a one of the boards he tested to it’s latest BIOS.
When a BIOS Update is available, with the 0x39 microcode or newer, I will follow up with additional testing.
Gaming: Power Consumption & Cooling
In every title tested, the i9-11900K had higher power consumption versus it’s predecessor – likely due to being backported from 10nm. However, I noticed two peculiar trends while recording gaming data. First, both CPUs consumed significantly less power on the Z590 motherboard I used – despite all settings being otherwise equal. The Comet Lake used 31% more watts with the Z490 board (72.6W vs 55.4W) and the Rocket Lake CPU used 26% more watts (83.5W vs 66.2W)
The other finding of interest was that despite using more power, the Rocket Lake CPU was actually easier to cool – averaging 50.8C vs 53.6C across ten games when cooled by a Noctua NH-U12.
Gaming: UHD 750 iGPU tested at 1920×1080
Instead of the vastly under-powered Gen9 iGPUs that have come with previous Intel CPUs, Rocket Lake has a 32EU Gen12 iGPU based on Intel’s Xe architecture. I wasn’t expecting much out of this iGPU, considering it has only 1/3 of the EUs as Tiger lake’s G7 iGPUs do.
Because the UHD 750 supports VESA Adaptive Sync (aka “FreeSync”) I didn’t have to deal with any screen tearing and the experience was generally smooth even when frame rates were low. There was one annoying glitch during my testing: the text for speech dialogue menus in Dragon Age: Origins did not display. Other than this issue, I was pleasantly surprised to find that many older titles, and some newer titles, achieved “playable” frame rates at 1920×1080 As my motherboard is on a BIOS version which has an older microcode version, memory latency is not currently ideal. In theory, iGPU performance could perform better with the 0x39 or newer microcode due to it’s lower memory latency. I will follow up with additional testing when an update is available for my motherboard.
As Intel’s Graphics Command Center includes a streaming feature, I have recorded gameplay performance and uploaded it to YouTube for demonstration. If you check it out, keep in mind streaming + gaming impacts performance by 10-20% – “Real Life” performance is slightly better.
CPU Gaming Tests – the good and the ugly
Let’s start with the good. In Ashes of the Singularity: Escalation, Rocket Lake shows a 14% improvement vs its predecessor: 38.3fps vs 33.5fps
Unfortunately, the rest of the results…. are quite ugly. The memory latency issues which occur as a result of my motherboard’s older microcode plague Rocket Lake, causing it to lose in literally every other title I had tested. Excluding Ashes of the Singularity, the 10th generation i9-10900K lead by 11.5% in gaming performance.
I had initially planned to show testing with multiple resolutions with 30 different games, but because of these issues Rocket Lake currently lags behind at even “GPU Bottlenecked” Settings. I will follow up with additional testing when BIOS Updates are available.
Where it stands now, the results are a bit mixed. On the positive side – In purely single threaded loads, there are definite improvements. 27% in CPU-Z, 19% in Cinebench. Encryption and Image Editing loads have impressive gains of 92% and 64% respectively. iGPU performance is nice, finally bringing acceptable performance for basic gaming to Intel Desktops.
On the other hand, with the older 0x34 microcode I’m seeing the i9-10900K beat the i9-11900K by an average of 11.5% in nearly every game. Multithreaded results are also mixed. I had some results where the Rocket Lake CPU was 5% ahead of Comet Lake, others where Comet Lake lead by 8%
At this moment I will avoid a final conclusion. As it is right now, I couldn’t recommend Rocket Lake to any gamer – as it is a downgrade with the BIOS I am running. Even if the issues with memory latency are fixed, I’m not sure it would be worth upgrading from 10th gen for the purpose of gaming alone – that would likely be better invested into a stronger GPU.
Once my motherboards have been updated to BIOSes with the newest microcode, I will retest and expand to add a larger amount of games. In theory, this should be within a few weeks.
Liked it? Take a second to support Albert Thomas on Patreon!