It can be seen that compared to the Core i9-9900KS running multi-core at 5 GHz, the Core i9-11900K is 13% stronger with the CB R20 test and 10% with the CB R15 test despite running multi-core at 4.8 GHz. If running at 5 GHz, then surely this distance is extended and it shows the improvement in IPC more clearly. As for Core i9-10900K, the Core i9-11900K has less than 2 cores, so it loses about 7.5% with the CB R20 test and 9.5% with the multi-core CB R15. As for single-core scores, the Core i9-11900K still has no competitor.
When running CB 15 all-core, the Vcore voltage is only about 1.3 V, the Core i9-11900K eats up to 185 W and with this ROG Ryujin heatsink, the CPU temperature is not more than 65 degrees Celsius. 5.3 GHz single core then the voltage for the Vcore will need 1.43 V.
So what if you turn on ABT? When turned on, the Core i9-11900K will easily reach 5.1 GHz all-core clock and run at this clock speed with normal loads but will cut it down to 5 GHz when running heavy tasks like Cinebench. So I tried one more setting and OC it to 5.1 GHz full core, forcing it to always run at this clock speed in all situations. As for the single-core pulse, I found it always reached 5.3 GHz, the 2 best cores will automatically jump to this pulse level, so the results do not have much difference.
Compared with the Cinebench scores of the Core i9-10900K and the Core i9-9900KS both 5.2 GHz all-core OC, you can see that the Core i9-11900K gives very good performance right at 5 GHz (with Intel ABT enabled) . It delivers nearly 13% higher multi-core performance than the 5.2GHz all-core Core i9-9900KS by nearly 13% with the CB R20 and 8.5% with the CB R15. Compared to the 5.2 GHz all-core Core i9-10900K, the Core i9-11900K lost 8.7% in the CB R20 and 11.6% in the CB R15 because it had less than 2 cores.
However, the Core i7-11900K consumes power very clearly because when running 8 cores at 5 GHz with Cinebench R15, the Vcore voltage will be brought to a very high level of 1.42 – 1.43 V, which was previously with Comet Lake or Coffee Lake line just 1.36 – 1.38 V is enough. The Core i9-11900K ate up to 250 W and the CPU temperature when I ran Cinebench R15 was 83 degrees C. When OC Core i9-11900K went up to 5.1 GHz full core, it required a voltage of 1.46 V and the motherboard should also have a good enough VRM system.
Trying with other rendering engines like Corona, V-RAY 4 and Blender, the performance difference between multi-core clock levels is quite significant. With Corona, when the ABT is turned on, the multi-core clock at 5 Ghz, it completes the test only 1 second faster than the default clock, when overclocked to 5.1 GHz with full core, the time is shortened to 3 seconds. With V-RAY 4, the 5 GHz all-core clock delivers 2.5% higher performance and at 5.1 GHz OC, 4.2% higher performance. If you use Blender to render, you should best turn on ABT, it significantly shortens the rendering time, I do not recommend OC 5.1 GHz because of the large power consumption and the CPU is very hot, up to 91 degrees C when running. this test.
Compared with previous generations, the performance of Core i9-11900K before other 8-core “predecessors” like Core i9-9900K or Core i9-9900KS is really impressive. 4.8 GHz full-core is enough for 17 – 17.5% higher V-RAY performance compared to the 5 GHz full-core of the Core i9-9900K or KS and at 5 GHz full-core, the Core i9-11900K is 20 more powerful % – a very significant figure and it shows a 2-digit improvement as Intel said on Cypress Cove’s IPC. Compared with the Core i9-10900K, although it has more than 2 cores, it is only about 4.5% – 7% stronger than the Core i9-11900K.
Use Handbrake to check the CPU encode performance and it’s surprising that the Core i9-11900K can give a much better performance than the Core i9-10900K despite losing in terms of multiplier. To test Handbrake, I used Big Bug Bunny H.264 4K @ 60fps video to convert to 1080p @ 30fps with Very Fast preset, still let Core i9-11900K run with 2 settings as default 4.8 GHz and Intel ABT enabled. The difference between these two modes is not much because it differs from rendering applications because the full core clock when running Handbrake is set to 4.8 GHz. However, if compared with the Core i9-10900K, you can see a very clear improvement, Core i9-11900K completes this test 45 seconds faster than the Core i9-10900K running 5.2 GHz with full core and has more than 2 multipliers. It is worth noting that Handbrake again exploits QuickSync if you use Intel and the GPU on the Core i9-11900K is more powerful, supports decoding advanced codecs and gives higher performance than the GPU on the Core i9-10900K.
The Core i9-11900K has 16 MB of L3 cache, 2 MB per core, 4 MB less than the Core i9-10900K but in exchange the L2 and L1 cache are both twice larger, 512 KB L2 and 48 KB L1. The Core i9-11900K has a significantly larger low-end cache bandwidth like L1 and L2 compared to the Core i9-10900K, although the bandwidth of the L3 is lower while the L3 latency is not much difference between the Core i9-11900K and Core i9-10900K. Particularly, there is a difference that I see when testing is the latency of RAM.
With Rocket Lake, Intel has designed a new memory microcontroller (IMC) to better support high-clocked RAM types. I use a fixed RAM kit to test all Intel CPUs, which is the G.Skills TridentZ DDR4-3200 CL14 kit (14-14-14-34) but on the Core i9-10900K, this RAM kit has significantly lower latency. with 46 – 47 ns (nanoseconds) while on the Core i9-11900K it is 53 – 56 ns. I think this delay is due to Rocket Lake’s new memory microcontroller, it has 2 modes including Gear 1, a 1: 1 ratio, synchronizing between IMC and ram (MCLK) pulses with IMC clock of 1800MHz and Gear. 2 ratio 1/2: 1 asynchronous with IMC pulse of 900 MHz, this solution is similar to pulse of Infinity Fabric bridge with MCLK on Ryzen series.
According to the formula for calculating the pulse of Dual Data Rate (DDR), then MCLK = (BCLK x Gear x QCLK ratio) / 2. BCLK is the reference pulse of components such as buffers and, the gear ratio will be 1: 1 (1.00) or 1/2: 1 (1.33) while the QCLK ratio is an odd ratio depending on IMC and RAM, this ratio is from 6 to 31. So to a RAM stick like G.Skill TridentZ DDR4-3200 running on Gear 1 ie 1: 1 ratio with microcontroller pulse = 1800 MHz. DDR RAM means we have 3200/2 = 100 x 1.00 x 32) / 2. Thus, the ratio of 32 of the QCLK ratio exceeds the limit of 31 but can still be achieved on the Core i9-11900K, but this largely depends on the quality of the CPU with good IMC and RAM kit. As a result, when I set it to auto, my RAM kit runs on Gear 2 and the latency cannot be less than 50 ns.
With 7-zip to test compress / decompress, Core i9-11900K gives pretty good compression performance, but decompression is still not comparable to Core i9-10900K with more than 2 cores.
The Core i9-11900K is optimized for gaming by Intel and has the advantage of being clocked up to 5.3 GHz single and double. This is true because the difference in fps of the games I test below with the same RTX 2060 Super graphics card has proven. Especially for CPU-oriented games that need high CPU speeds such as CS: GO, the fps you have when playing with the Core i9-11900K and RTX 2060 Super combo is up to 445 fps, the advantage of screens that have scanning speeds above 360 Hz will maximize.
3DMark scores with Fire Strike (DirectX 11) tests with 3 FHD, 2K and 4K resolutions plus Time Spy (DirectX 12) with 2K and 4K resolutions show a huge improvement in CPU performance when playing. game. I am not surprised by this because Core i9-10900K when playing games, the cores run at 4.9 GHz, sometimes up to 5.1 GHz while the Core i9-11900K easily reaches 5.1 GHz multi-core clock and arrives. 5.3 GHz.
Power and temperature
Regarding power consumption and temperature, as you have heard about the Core i9-11900K can eat over 300 W of electricity. When does it eat? That’s when you turn on the AVX-512 and stress it with Prime95. I tried stress with AIDA64 but it could not eat that much electricity, for all the applications I tested, the maximum power I observed was 271.66 W when running Blender. The table below shows Rocket Lake PL1 and PL2 levels at the same time as Tau. The reason it can be eaten on 251 W according to the PL2 design is due to Intel Adaptive Frequency, the CPU is cool, the Vcore voltage can also be increased, it will let the CPU run on PL2, ignore the TB 3.0 or TVB pulse level. and even the Core i9-11900K can fully load at 100 degrees C without crashing.
The temperature issue is something to keep in mind because I am using the heavy-duty AIO water cooler, the ASUS ROG Ryujin 360, which uses Asetek’s Gen7 pump and it is equipped with 3 Noctua’s NF fans. -F12 industrialPPC 2000 PWM with static pressure up to 3.94 mmH2O. So its performance is very high and enough to keep the Core i9-11900K at a safe temperature when stress test with Intel ABT turned on. So you need to invest in good heat dissipation for Core i9-11900K if you want to exploit its maximum performance. As for OC, the AIO dissipation like Ryujin is still not strong enough because with very high Vcore voltage right from the default pulse levels, I can only OC it at 5.1 GHz with a Vcore 1.46 V. tried pushing it to 5.2 GHz full core but it was too hot to run properly.
In short, the Core i9-11900K is still a good CPU, but it’s already the end of the 14nm process. I clearly see Intel’s intent to experiment with this generation, testing the performance of a new architecture on an old process – something unprecedented. Personally, I really “respect” Intel for their optimization performance on the old process, the CPU clock is naturally pushed to 5.3 GHz, without OC, the Cypress Cove architecture really brings about improvement. advancing IPC and Xe-LP graphics core has taken the encode performance of CPU to a new level. In terms of games, the Core i9-11900K with its high pulse advantage continues to prove it is the best CPU for current gaming. But unfortunately it still uses 14nm process so it is very power hungry, thereby making the CPU hotter and also harder to OC than previous generations. It was not until the end of this year that with Alder Lake, things really changed because Intel will finally officially bring the desktop CPU down to the 10nm SuperFIN process – a process with a semiconductor density equivalent to the 7nm TSMC that AMD is using. use with Ryzen. Then Intel’s CPUs really changed, in terms of the performance of the new architecture and in terms of power and temperature.