The Core Ultra 7 270K was too good, so Intel scrapped the flagship Core Ultra 9 290K Plus — benchmarks of the 290K prototype find slim 2% faster performance in gaming and applications

Alder Lake
(Image credit: Intel)

Intel canned the Core Ultra 9 290K Plus from the Arrow Lake refresh lineup announced a few months ago, despite a swirling of leaks and rumors confirming its existence. The chip ultimately never came out, but a Chinese reviewer just got their hands on an engineering sample and put it through the wringer — the underwhelming results in games and professional apps show why Intel likely chose to keep it in the archives.

As a reminder, the Core Ultra 9 290K Plus would be based on the existing 285K, so it'd share the same 24-core config (8P+8E) but with slightly tuned clock speeds, DDR5-7200 support, and newer features such as Intel's binary optimization tool. That tool is actually one of the ways to confirm this 290K Plus was legit since it only supports Arrow Lake refresh silicon at the moment, and the BIOS recognized the CPU correctly.

Jumping to the tests, multi-core results in synthetic benchmarks were more impressive than the single-core numbers. The biggest win was seen in CPU-Z, where the 290K Plus was 2.84% faster than the 270K Plus. In Cinebench R24, the 290K Plus managed only a 0.69% higher score in the single-core test. On average across all synthetic workloads, the 290K Plus beat its lower-tier counterpart by only 1.5%.

Latest Videos From

Intel Core Ultra 9 290K Plus Benchmarks

Swipe to scroll horizontally

Benchmark Metric

Core Ultra 9 290K Plus

Core Ultra 7 270K Plus

Performance Delta (U9 vs U7)

CPU-Z (Single-Core)

920

905

+1.65%

CPU-Z (Multi-Core)

19,546

19,007

+2.84%

Cinebench R23 (Single-Core)

2,465

2,433

+1.32%

Cinebench R23 (Multi-Core)

44,810

44,230

+1.31%

Cinebench R24 (Single-Core)

146

145

+0.69%

Cinebench R24 (Multi-Core)

2,568

2,540

+1.10%

Geekbench 6 (Single-Core)

3,315

3,286

+0.88%

Geekbench 6 (Multi-Core)

24,273

23,642

+2.67%

In more intensive tasks such as compression, real-time rendering, and compiling, AMD's new Ryzen 9 9950X3D2 won in all but one test: Ansys Fluent Simulation. Here, the Core Ultra 9 290K Plus was 9.3% faster than AMD's offering and about 4.6% faster than the 270K Plus. Averaging out all the results, the 290K Plus was 6.3% faster than the 270K Plus but about 8.3% behind the 9950X3D2.

Intel Core Ultra 9 290K Plus Gaming Benchmarks

At 1080p, the average FPS improvement over the 270K Plus is about 2% across six titles. The biggest difference was in Delta Force, where the 290K Plus achieved 8.3% higher FPS and 3.33% better 1% lows. Both Black Myth: Wukong and Resident Evil 9 actually saw it lose to the 270K Plus by around 1%. The 9950X3D2, as you'd expect, bested either Intel offering with ease thanks to its massive cache pool.

Swipe to scroll horizontally

Game

Core Ultra 9 290K Plus

Core Ultra 7 270K Plus

Performance Delta (U9 vs U7)

Counter Strike 2

Avg: 368 / 1% Low: 214

Avg: 364 / 1% Low: 212

Avg: +1.10% / 1% Low: +0.94%

PUBG

Avg: 193 / 1% Low: 99

Avg: 189 / 1% Low: 96

Avg: +2.12% / 1% Low: +3.12%

Delta Force

Avg: 234 / 1% Low: 93

Avg: 216 / 1% Low: 90

Avg: +8.33% / 1% Low: +3.33%

Black Myth: Wukong

Avg: 98 / 1% Low: 87

Avg: 99 / 1% Low: 88

Avg: -1.01% / 1% Low: -1.14%

Resident Evil 9

Avg: 138 / 1% Low: 103

Avg: 139 / 1% Low: 100

Avg: -0.72% / 1% Low: +3.00%

Cyberpunk 2077

Avg: 206 / 1% Low: 123

Avg: 201 / 1% Low: 123

Avg: +2.49% / 1% Low: 0.00%

Moving to 1440p gaming, the difference shrinks even more since the games become more GPU-reliant as you scale the resolution ladder. Delta Force once again exhibits the largest gap, about 6.8% ahead of the 270K Plus, and a surprising 14% ahead in 1% lows. The 290K Plus still falls 1% behind in Black Myth: Wukong while matching the 270K Plus in Resident Evil 9. On average, the unreleased flagship is 1.5% faster than the actual top-end Arrow Lake refresh CPU.

Swipe to scroll horizontally

Game

Core Ultra 9 290K Plus

Core Ultra 7 270K Plus

Performance Delta (U9 vs U7)

Counter Strike 2

Avg: 352 / 1% Low: 211

Avg: 344 / 1% Low: 209

Avg: +2.33% / 1% Low: +0.96%

PUBG

Avg: 189 / 1% Low: 103

Avg: 188 / 1% Low: 94

Avg: +0.53% / 1% Low: +9.57%

Delta Force

Avg: 218 / 1% Low: 89

Avg: 204 / 1% Low: 78

Avg: +6.86% / 1% Low: +14.10%

Black Myth: Wukong

Avg: 86 / 1% Low: 76

Avg: 87 / 1% Low: 78

Avg: -1.15% / 1% Low: -2.56%

Resident Evil 9

Avg: 95 / 1% Low: 73

Avg: 95 / 1% Low: 73

Avg: 0.00% / 1% Low: 0.00%

Cyberpunk 2077

Avg: 184 / 1% Low: 127

Avg: 183 / 1% Low: 129

Avg: +0.55% / 1% Low: -1.55%

If we put all the numbers together, we get roughly 2% gains in gaming and almost 4% in productivity tasks, compared to the Core Ultra 270K Plus. Those slim margins would make it hard to justify a much higher price tag for a Core Ultra 9 SKU, which explains why Intel likely never released it. The chips that did come out are excellent value, so that a flagship offering might've thrown the whole lineup off-balance, especially in terms of optics.

Google Preferred Source

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

Hassam Nasir
Contributing Writer

Hassam Nasir is a die-hard hardware enthusiast with years of experience as a tech editor and writer, focusing on detailed CPU comparisons and general hardware news. When he’s not working, you’ll find him bending tubes for his ever-evolving custom water-loop gaming rig or benchmarking the latest CPUs and GPUs just for fun.

  • usertests
    It was obvious from the specs.

    This hasn't stopped Intel before, see the KS models. But it was better to get rid of it.
    Reply
  • bit_user
    The article said:
    As a reminder, the Core Ultra 9 290K Plus would be based on the existing 285K, so it'd share the same 24-core config (8P+8E) but with slightly tuned clock speeds
    Slight error: it's a 8P + 16E config, but you got the total core count right.

    Seems like they still could've offered a "KS", though. Those don't adhere to the normal perf/$ value formulation. Not that I care, either way.
    Reply
  • arronox
    No, No, No. False. The 270k is a great cpu sure, for its overall performance. Can game but not the best, can multitasking but not the best, highest chip that can do both, for a great price. Intel could have done better with the 290K, 2 more p cores slightly better/ faster fabrics, and an aggressive more cache move. All to keep up. A reason for asking 500 dollars.
    Reply
  • usertests
    arronox said:
    No, No, No. False. The 270k is a great cpu sure, for its overall performance. Can game but not the best, can multitasking but not the best, highest chip that can do both, for a great price. Intel could have done better with the 290K, 2 more p cores slightly better/ faster fabrics, and an aggressive more cache move. All to keep up. A reason for asking 500 dollars.
    No, No, No. False. This is Arrow Lake Refresh. You aren't getting a new compute die with more P-cores and cache out of nowhere. They had to do what they could with the same die.

    If you want absurd core counts, you'll get them with Nova Lake.
    Reply
  • arronox
    usertests said:
    No, No, No. False. This is Arrow Lake Refresh. You aren't getting a new compute die with more P-cores and cache out of nowhere. They had to do what they could with the same die.

    If you want absurd core counts, you'll get them with Nova Lake.
    Isn't exactly what the ultra 7 got, more cores, faster fabric, some more cache... oh no...
    Reply
  • JRStern
    usertests said:
    It was obvious from the specs.
    How is it obvious from the specs? Where's the bottleneck? More cores, same cache?

    Also is Intel and/or Microsoft not really doing much with the e-core/p-core deal? I still can't believe they allocate those dynamically.
    Reply
  • bit_user
    arronox said:
    Isn't exactly what the ultra 7 got, more cores, faster fabric, some more cache... oh no...
    The way Intel gave the Ultra 7 270K+ "extra" cores is just by disabling fewer than they did in the Ultra 7 265K.

    Arrow Lake's biggest compute die has 8P + 16E. In the 265K, they disabled 4E cores, yielding a config of 8P + 12E. When they made the 270K+, they didn't disable anything, so you just get the full die with all cores enabled.

    Also, each cluster of 4E cores has some cache. So, when they enabled those last 4 E cores, the L3 cache went from 30 -> 36 MB and the total L2 cache went from 36 MB -> 40 MB.

    As @usertests was saying, Intel makes these "Refresh" parts by taking existing silicon and trying to tune up the settings a little bit, possibly supported by better binning and more mature manufacturing nodes. They don't actually design any new silicon, as that would be far too expensive.
    Reply
  • AmirTHforum
    Ok, can someone please help me understand; I have asked this once before (I don't know if it was here or not), but the answer seems not to make sense to me.
    In my opinion, and I could be very wrong about this, the number of E Cores of any CPU, especially a Desktop CPU should not exceed 4 Cores. The E Cores should only be there to keep the computer idling and for very light duty; background tasks and programs that do not need that much power.
    All the work and the majority of the Core count should always be P Cores.

    Intel has always this policy that they didn't like to add CPUs unless it was an expensive Server or workstation CPU, and the only reason they dump, and the operative word here is Dump; is to appear to be competitive to AMD's core counts.

    Please correct me if I am wrong, but please give me a good reason why I'm wrong.
    Reply
  • magicturtle1
    AmirTHforum said:
    Ok, can someone please help me understand; I have asked this once before (I don't know if it was here or not), but the answer seems not to make sense to me.
    In my opinion, and I could be very wrong about this, the number of E Cores of any CPU, especially a Desktop CPU should not exceed 4 Cores. The E Cores should only be there to keep the computer idling and for very light duty; background tasks and programs that do not need that much power.
    All the work and the majority of the Core count should always be P Cores.

    Intel has always this policy that they didn't like to add CPUs unless it was an expensive Server or workstation CPU, and the only reason they dump, and the operative word here is Dump; is to appear to be competitive to AMD's core counts.

    Please correct me if I am wrong, but please give me a good reason why I'm wrong.
    As long as there are 6 to 8 p cores to do the heavy task that need faster speeds then it won't hurt gaming or responsiveness speeds. The more e-cores they had the better multi-core score they get from the same die area. Since 4 ecores are the same size as 1 p core. It's cheaper for them to compete with amd to through more e cores at multi threaded applications and have just enough p core to not hurt or or gaming performance
    Reply
  • bit_user
    AmirTHforum said:
    In my opinion, and I could be very wrong about this, the number of E Cores of any CPU, especially a Desktop CPU should not exceed 4 Cores. The E Cores should only be there to keep the computer idling and for very light duty; background tasks and programs that do not need that much power.
    All the work and the majority of the Core count should always be P Cores.
    A good way to think about the distinction between P-cores and E-cores is that the P-cores are for latency-sensitive stuff, while E-cores are for maximizing throughput.

    It's like: why doesn't a medieval army have all knights on horses? If mounted cavalry is so much better than foot soldiers, why not just have an entire army that's cavalry? The answer is obviously that foot soldiers are much cheaper to train, equip, and feed, so you can have lots of them. By sheer numbers, they can have a huge force of impact.

    Similarly, the P-cores are enormous and power-hungry to run at high frequencies, which is where they can provide the greatest benefit over E-cores. If you had a chip with all P-cores, not only couldn't you have as many of them, but you couldn't run them as fast, for all-core workloads.

    Let's look at the math. Here's how single-threaded performance compares on P vs. E cores, in Arrow Lake 285K:
    Source: https://chipsandcheese.com/p/analyzing-lion-coves-memory-subsystem
    So, a single E-core is 77.4% as fast at integer and 75.2% as fast at floating-point workloads. Yet, by having 16 of them, they provide about 1.5x the raw horsepower of the P-cores, while using half area of the P-cores (i.e. in a 8 vs. 16 configuration) and quite a bit less power.

    If the 285K didn't have those 16 E-cores, then the most it would realistically have instead is just another 4 P-cores. So, for the amount of area they use, the E-cores provide roughly 3x the performance of P-cores. People doing rendering, compilation, and other highly-threaded workloads are much better off with a large contingent of E-cores, rather than having just a few more P-cores.
    Reply