24-Core Intel Core Ultra 9 285 Falls Short of 8-Core Ryzen 7 9700X in Geekbench Leak

[Cpt.Jank]

The leaks and rumors surrounding Intel's upcoming Arrow Lake desktop CPU line-up are starting to heat up, with recent rumors tipping the existence of the Core Ultra 9 285K as the top-end chip in the upcoming launch. A new set of Geekbench 6 scores spotted by BenchLeaks on X, however, suggests the Core Ultra 9 285 non-K variant of this CPU might lag its Ryzen 9 counterparts significantly.

The Geekbench 6 test results, which were apparently achieved on an ASUS Prime Z890-P motherboard, reveal performance that falls short of even the current-generation AMD Ryzen 7 9700X, never mind any of the Ryzen 9 variants. The Geekbench 6 multicore score came in at an unimpressive 14,150, while the single-core score was a mere 3,081, falling short of the likes of the AMD Ryzen 7 9700X, which scored up to 19,381 and 3,624 in multi- and single-core tests, respectively. However, there appears to be more to this story—namely an odd test configuration that could heavily skew the test results, since the "stock" Intel Core Ultra 9 285K scores significantly higher in the Geekbench 6 charts than this particular 285 seems to.



Apparent engineering samples of the Intel Core Ultra 9 285K that made it onto the Geekbench 6 charts have registered multicore scores as high as 21,447, which puts the K-SKU CPU with the same core layout and very similar clock speeds significantly ahead. Digging into the results a little, it appears that the ASUS test platform powering the Core Ultra 9 285 CPU only had 8 GB of DDR5 RAM running at 5586 MT/s, while Core Ultra 9 285K tests that scored much higher had up to 32 GB of DDR5 RAM running at up to 5598 MT/s.

That said, the listed base clock speeds of the Core Ultra 9 285's E-cores are also significantly lower than those of the top-scoring Core Ultra 9 285K, at just 2.5 GHz, compared to the 285K's 3.7 GHz base frequency. This could also explain some of the discrepancy.

Geekbench isn't known to be very RAM-dependent, with the benchmark's largest working set being on the order of 1.6 GB, so there may be some validity to this benchmark score, however it wouldn't be surprising to see the CPU perform significantly better in subsequent benchmarks. At the very least, this Geekbench score validates previous rumors about the expected hardware configuration of the Intel Core Ultra 9 285 and 285K CPUs:

• 8 P-cores
• 16 E-cores
• 24 threads
• Base clock: 2.5 GHz (P-core base clocks will likely be higher)
• Boost clock: 5.586 GHz (likely only for the P-cores)

View at TechPowerUp Main Site | Source

 

[Space Lynx]

I know it's not the best bench, but I was not expecting that regardless. Hmm. I was expecting top of the line Arrow Lake to really dominate the benches across the board, best to wait for official benches I suppose.

 

[persondb]

If it's really running 8GB of DDR5 then it's almost certainly running it in a single channel configuration. This will likely affect the multicore score as several of individual tests depend a lot in memory bandwidth.

I do not believe there is any 4GB DIMM for ddr5.

Geekbench might have detected it wrongly though.

 

[_roman_]

Let's see official reviews.

My operating systems can not handle different processor architectures. Mainly the scheduler of the operating system kernel. I also doubt my toolchain will handle easily without tweaking those different cores while compiling code. The main reason why I went with a 6 core with 6 fake cores all with the same architecture. I went also with a recent product as in my point of view older product hardly get's any security fixes or any optimizations at all or any mainboard updates.

 

[Chaitanya]

At what wattage though?

 

[kawice]

It's misleading news header. It should be clearly stated that it's for 65W rated CPU.

Intel has never been energy efficient, look at mid-range and high end CPUs from Intel's 13 and 14 gen, their real TDP is 180W and 253W.

In gaming Ultra 9 285K will be best on the market. AMD can only compete with their 3DCache variants of their high end CPUs tier, but with their high core to core latency and invisible internal numa design (CCX), it's not really good for gaming.

 

[napata]

What has hardware news reporting become? Smh.

It also falls short of the Geekbench leak from yesterday about the 265K. I wonder why that isn't that the title?

I wonder what the Geekbench leak from tomorrow will have as its title.

 

[Hyderz]

Speculations speculations, the reviews is in 10 days?

 

[Nanochip]

This is the 285 not the 285K!!!

 

[docnorth]

As already mentioned it's the 285 non-K version and for some reason underperforming.. Even at strictly 65w, single core should be close to 285K levels, at least >3300.

 

[SL2]

Who cares.

Why is this even posted, it's just misleading. GBench is a waste of time, where an 8 core can have a multithread score that's 80 % of a 16 core.

 

[Solid State Brain]

At what wattage though?

Even if it was 65W, performance scaling with power is not linear. All CPUs operate much more efficiently when frequencies/voltages (which power limits affect) are kept low.

 

[Caring1]

Who cares.

Why is this even posted, it's just misleading. GBench is a waste of time, where an 8 core can have a multithread score that's 80 % of a 16 core.
View attachment 365563

Did you forget clock speeds that are taken into account in GB.

 

[SL2]

Did you forget clock speeds that are taken into account in GB.

Well that's part of the problem, isn't it?

You do know that the 9700X runs with a slower clock speed, right?

 

[Caring1]

There's obviously some other factor in their testing to make them run that close when the 9950X should be at least double the score

 

[SL2]

There's obviously some other factor in their testing to make them run that close when the 9950X should be at least double the score

Hence why it shouldn't be posted at all, which was my point from the beginning.

It's useless.

Here's another example for those who are unfamiliar with this pos bench lol

Multicore, from here: https://browser.geekbench.com/processor-benchmarks

 

[maximumterror]

Speculations speculations, the reviews is in 10 days?

I love tech reviews, but I don't really trust them because their numbers are different than mine

 

[SL2]

I love tech reviews, but I don't really trust them because their numbers are different than mine

And they don't match gkbench, or p*ssmark.
Or my dreams.

 

[Asni]

If it's really running 8GB of DDR5 then it's almost certainly running it in a single channel configuration. This will likely affect the multicore score as several of individual tests depend a lot in memory bandwidth.

I do not believe there is any 4GB DIMM for ddr5.

Geekbench might have detected it wrongly though.

I can confirm that GB6 is reading a single DDR5 module as dual channel due to the dual subchannel configuration. The benchmark was actually performed with a single 8GB DDR5-5600MT/s module.
According to this test, running GB in single channel mode is taking ~35% away.

Memory Showdown: Single vs Dual vs Quad Channel

What's the difference between single channel, dual channel, and quad channel memory configurations? Does it matter?

techguided.com

(note that this was DDR4 on 12600k).

 

[LittleBro]

It's misleading news header. It should be clearly stated that it's for 65W rated CPU.

9700X, too, is a 65W TDP CPU.

Intel has never been energy efficient, look at mid-range and high end CPUs from Intel's 13 and 14 gen, their real TDP is 180W and 253W.

Arrow Lake was said to lower TDP and increase performance. Intel stated, that Arrow Lake's highend will consume 100W less than Meteor Lake's highend while retaining clocks and performance.
Even 280W TDP is still extreme, but of course, it's no longer 380W as with 14900K(S).

AMD can only compete with their 3DCache variants of their high end CPUs tier, but with their high core to core latency and invisible internal numa design (CCX), it's not really good for gaming.

In many non-gaming tasks Zen 4/5 cores prove to be better than Intel. Just have a look at Phoronix.
In gaming, Intel has few digit % to 10% performance uplift over Zen 4/5 but at what cost? 2-3 times more TDP, temps around 110°C and extreme voltages that cause degradation?
AMDs biggest problem is their shitty memory controller and aged chipset silicon.

Personally, I don't have any expectations towards Arrow Lake now. Look how Lunar Lake ended up.

 

[dirtyferret]

With these results it's not hyperbole when I say "THIS CHANGES EVERYTHING"/s

 

[ThomasK]

In gaming Ultra 9 285K will be best on the market. AMD can only compete with their 3DCache variants of their high end CPUs tier, but with their high core to core latency and invisible internal numa design (CCX), it's not really good for gaming.

Where are the real world tests to back up your claim? Or are you using your crystal ball?

The single CCX designs have been dominating the gaming charts, whilst using a fraction of the power. Have you not seen?

 

[maxfly]

garbagebench

footsteps...door closing.

 

[Geofrancis]

benchmarks like geekbench are a joke, you can can make any CPU win just by changing the number of times it runs each test. Intel used to pay companies to do this back in the day when AMD started winning benchmarks they would just find the tasks that intel could do better and repeat them to change the score. so all you got was the total score so all you seen was intel winning. bapco was one, there was a few others too.

This is why im a fan of aplication benchmarks, synthetics are only good for single tasks that that dont change like calculating prime or pi.

 

[AnotherReader]

These results are highly suspect. Chips and Cheese got their hands on an ASUS laptop with the Core Ultra 7 258V. Its P cores show significant gains over Meteor Lake's P cores. Since these are laptops, the usual caveats apply, but I believe this Geekbench result is underestimating Arrow Lake's performance.