WHY IS NO ONE TESTING THIS?? Intel i9-12900K & i5-12600K P Core & E Core Benchmarks
Something I was incredibly surprised to find missing from the vast majority of launch day reviews – including from the big folk like Linus – was test results from each type of core in Intel’s new Alder Lake CPUs. If you didn’t know, these new chips feature a “hybrid” design which you can think of like having two VERY different CPUs under one heatsink.
The P Cores, or performance cores, are the full fat, high power monsters we are used to calling “cores” in Intel CPUs. These are an evolution of the same design from the 11th gen chips and before, they boost like crazy up to 5.2GHz in the i9, feature HyperThreading, and are pretty large on the die.
By contrast, the E cores, or efficiency cores, are tiny. They are grouped in clumps of 4, and that grouping takes up a touch LESS space on the die than a single P core. These don’t support HyperThreading, and only boost up to 3.9GHz at most, but are meant to draw significantly less power than their P core counterparts.
The idea of including both on a single chip is that the low power E cores will handle all the background tasks, while the P cores get to be switched off saving power, then when you fire up a game or a render they’ll kick into action and offer exceptional performance, then switch back off again when you’re done. The trouble with that idea is that it isn’t exactly simple to know what counts as a “background task” or even a “high performance task”. Linux, and by extension Android, has had kernel-level support since 2012 and has evolved it since then, and Apple has been using some version of big.Little since 2016 and again had considerable development since then. By contrast, these Alder Lake chips are the first major launch of this hybrid design that Windows is going to have to seriously optimise for, hence Windows 11 existing.
If these new chips only used the P cores, they’d be pretty standard, it’d be just like every other launch and testing the chip as a whole would make perfect sense. But, seeing as they don’t, I think it’s really important to see how each of them perform, independent of each other. I covered some of this in both CPU reviews – which I highly recommend you also check out if you haven’t already – but for this video I wanted to retest a few things and collect more detailed data.
So, what’s the big deal? How different could they really be? Around 100%. As in, the P cores offer almost 100% more performance on SINGLE THREADED WORK, like Cinebench R23. The E cores in both the i5 and i9 run at about half the performance of their respective P cores. Now, that’s to be expected. They run at much lower clock speeds, they don’t support HyperThreading (not that that is a factor for single threaded work), and they are much, much smaller with significantly reduced computation power, so obviously they are going to be slower. What makes this metric interesting is their power usage.
The P cores draw around 40W package power, and the E cores? They are about 23W, with a peak of 28W. That correlates quite clearly with their performance, with a negative offset for the E cores. I’m quite surprised they draw that much, I was expecting them to draw less than half the power, not more.
When it comes to multi-threaded workloads, also unsurprisingly the P cores offer an order of magnitude more performance. Comparing 8 P cores (and 16 Threads) to 8 E cores (and 8 Threads), the P cores in the i9 offer over 150% more performance, at least in Cinebench R23. Looking at the power consumption though, that’s where the efficiency difference stands out. The P cores chew through an insane 232W, or 238W peak, whereas the E cores practically sip power at 51W stable or 60W peak. That works out to around 4x the power for the P cores, while providing around 2.5x more performance.
What’s also interesting is when you add the individual results from the 12600K’s cores together, you get a result that’s slightly higher than their stock combined result. While there is some room for margin of error, the disabled E cores mean that the entirety of the power budget is made available to the P cores, rather than the E cores skimming a little off the top.
With the E cores disabled, it’s also a good time to look at how a more true representation of the generational improvement compares to its peers. The i9 still definitively punches above its weight, clocking in 22% more performance than the 10 core 10850K, almost 40% more performance than last gen’s 8 core 11900K, and only comes in at 10% slower than the 12 core Ryzen 5900X. Beating an 18 month old 10 core with just 8 by over 20% is no mean feat, and coming in just 10% slower than AMD’s current generation 12 core is nothing to be ashamed about. For me, that’s what makes these chips so interesting and impressive. It’s a significant improvement gen on gen – almost 40% – without the extra E cores.
What about something a bit more real world though? Well, in Blender and the BMW scene, the 8 E cores take a little over 5 and a half minutes to render the scene, whereas the P cores take just 2 minutes and 10 seconds. While it’s no longer a fair 8 v 8 comparison like it is with the i9, the i5’s 6 v 4 is significantly more extreme. The E cores take over 11 and a half minutes, compared to the P core’s 3 minutes and 6 seconds, although interestingly the i5’s P cores are significantly more efficient than the i9’s meaning they draw around 275% more power, but offer around 230% more performance, or much closer to a 1-1 ratio.
That difference gets extrapolated greatly in the Gooseberry render thanks to its increased render time. The i9’s 8 P cores takes 11 minutes and 13 seconds to render the frame, versus a whopping 32 minutes for the 8 E cores – which at the same power draw means the gap is a lot closer at almost 200% more performance for around 300% more power. Interestingly the i5 actually swings in the P cores favour fully, as they offer over 300% more performance, but for the same 230% more power meaning the E cores actually used a little over 20% more power to complete that render as drawing 36W for AN HOUR – actually an hour and 6 minutes – ends up with a higher total than 120W for 16 minutes.
Comparing to their respective peers again, the i9’s P cores again sit with a comfortable lead over the last two generations of i9’s, although now also a sizable chunk behind the 12 core Ryzen 5900X – which is to be expected. It’s still only a 25% margin, which considering the 5900X has 50% more cores, that’s not bad at all! The i5 on the other hand, doesn’t fare quite as well, only ending up with a 4% lead over the 6 core Ryzen 5600X as these chips struggle in longer workloads compared to Ryzen. As for their E cores, the 8 in the 12900K sit somewhere between an i3 10100 and an i5 10400F here, offering around 15% more performance than the i3, but 30% down on the i5. As for the 12600K’s 4 E cores… the only thing I’ve ever tested that ran slower was a severely throttled 1185G7 in a Lenovo Yoga ultrabook – although a better cooled 1165G7 in the Razer Blade Stealth handily outpaced it – while drawing around the same or slightly less power.
One final thing I want to note before we jump into the gaming results is an observation. Since the i9’s P cores draw around 28W per core in all core workloads, and the E cores draw between 25W and 30W per cluster of 4, and the cluster of 4 E cores takes up the same space as one P core, it’s conceivable that Intel could make a 10P + 0E design of this chip and still maintain the same power envelope and die size. That would also alleviate any issues with the scheduler sending tasks to the slower E cores, and with my napkin maths would put it over 10% faster than the 12 core 5900X with more consistent performance too. Of course that’s technically a slower chip overall as the 8 E cores are able to offer more performance per watt than 2 more P cores, but I thought that was interesting.
Ok, let’s talk gaming. Unsurprisingly, the performance difference between the P and E cores, on both chips actually, is pretty substantial. In CSGO, the E cores consistently get less than half the average FPS and it’s worth noting that both chips experienced stuttering and hitching with the i5 struggling a lot more. The playing experience on the i9 was mostly fine with the occasional hiccup, whereas the i5’s E cores alone was almost unplayable. Interestingly when adding some more chips in for comparison, the i9’s P cores actually gain a touch more performance over stock.
In Cyberpunk, the story is much the same. The E cores offer significantly reduced performance, although the i9 sporting 8 of them results in a reasonable experience overall. Again, disabling the E cores actually improves performance over stock here too. That could be a scheduler issue, or power budget, but it’s worth noting if you do have any of these chips.
Watchdogs straight up refused to run on either set of E cores, so I’ll have to skip over that one, although I will mention that Intel says 32 games with Denuvo currently aren’t compatible with these new Alder Lake CPUs so this isn’t exclusive to Watchdogs.
Microsoft flight again sees a performance increase using just it’s P cores, on both chips, although the E cores do suffer some performance loss. Remarkably the average frame rate doesn’t drop to an insane degree, regardless of it running on 4 or 8 cores!
Finally in Fortnite this is the only game that didn’t improve much if at all on the P cores alone, and actually remained a playable experience on the E cores alone as well. Performance does take a significant hit, going from 274 FPS on the i9 to 179 FPS, or from 266 FPS on the i5 P cores to 129 FPS, but it was still remarkably playable.
I think testing on the E cores alone gives a good indication of what we can expect from the lower end chips, especially the mobile and ultra-mobile options, some of which I could imagine only coming with these E cores. Of course I’m testing with an RTX 3080 desktop GPU so it’s unlikely that it’ll translate quite the same to mobile and ultra-mobile performance, but it’s interesting to see nonetheless.
Looking at the results, especially from the i5 without it’s E cores active, I’m personally more excited for the non-hybrid versions of these chips. A 6P + 0E design would potentially offer more gaming performance, cost less, and have no issues with schedulers, and that sounds like a win to me. I’d be very interested to see how a 10 core i9 actually stacks up, especially with the thermal hotspot issues I noticed in my review, with such high power consumption in such a small die it’s just really difficult to evacuate heat fast enough away making the central cores spike in temperature.
It’s also interesting to see the power consumption on the E cores, sitting somewhere between 6 and 9W per core in all core workloads – or coming rather close to what AMD is currently offering in their “full fat” Zen 3 cores, and of course with significantly less performance, and compared to Apple’s M1 Pro and Max chips, well they are drawing more like 3-4W per core while providing 60% more all core performance, or a shocking 280% more single threaded performance at half the power consumption according to Anandtech. That’s absolutely insane. I think Intel could do with a few more generations of both core designs to reduce power consumption significantly.
