Intel Core Ultra 9 285K Tops PassMark Single-Thread Benchmark

[AleksandarK]

According to the latest PassMark benchmarks, the Intel Core Ultra 9 285K is the highest-performing single-thread CPU. The benchmark king title comes as PassMark's official account on X shared single-threaded performance number, with the upcoming Arrow Lake-S flagship SKU, Intel Core Ultra 9 285K, scoring 5,268 points in single-core results. This is fantastic news for gamers, as games mostly care about single-core performance. This CPU, having 8 P-cores and 16 E-cores, boasts 5.7 GHz P-core boost and 4.6 GHz E-core boost frequencies. The single-core tests put the new SKU at 11% lead compared to the previous-generation Intel Core i9-14900K processor.

However, the multithreaded cases are running more slowly. The PassMark multithreaded results put Intel Core Ultra 9 285K at 46,872 points, which is about 22% slower than the last-generation top SKU. While this may be a disappointment for some, it is partially expected, given that Arrow Lake stops the multithreaded designs in Intel CPU families. From now on, every CPU will be a combination of P and E-Cores, tuned for efficiency or performance depending on the use case. It is also possible that the CPU used inn PassMark's testing was an engineering sample, so until official launch, we have no concrete information about its definitive performance comparison.



View at TechPowerUp Main Site | Source

 

[Soul_]

For about 30 seconds, before it melts down . Jokes aside, I hope intel comes back swinging and keeps competition going.

 

[phanbuey]

between this and the 9800X3D, im super excited for CPU refresh.

 

[Chrispy_]

"stops the multi-threaded designs"

So Arrow lake is the first Intel CPU to abandon hyperthreading/SMT on P-cores? (E-cores already lack HT)

 

[Daven]

"stops the multi-threaded designs"

So Arrow lake is the first Intel CPU to abandon hyperthreading/SMT on P-cores? (E-cores already lack HT)

Yes

 

[phanbuey]

"stops the multi-threaded designs"

So Arrow lake is the first Intel CPU to abandon hyperthreading/SMT on P-cores? (E-cores already lack HT)

Yeah essentially the thought is - HT threads/logical cores are 30%-40% of a real 'P-core' performance since it relies on pipeline stalls to execute code, replace it instead with e cores that are 60% the performance of a P core instead and put that thread on E cores.

The currrent architecture has to juggle HT cores, e cores and P cores, and AFAIK that's not very efficient/overly complex. Disabling HT also generally lowers latency.

 

[dirtyferret]

I know i said this about the leaked Cinebench R23 rendering scores for the AMD Ryzen 9 9800X3D but this time it's definitely not hyperbole when I say, THIS CHANGES EVERYTHING!!!

 

[dgianstefani]

~10% better ST would be great, considering 14900K is already extremely fast ST, the X3D chips have similar gaming performance but that's because of cache. They're actually significantly slower in ST.

 

[RayneYoruka]

Intel's life cycle on a nutsell

 

[phints]

My fall is basically 9800X3D vs. 265K.

 

[persondb]

Yeah essentially the thought is - HT threads/logical cores are 30%-40% of a real 'P-core' performance since it relies on pipeline stalls to execute code, replace it instead with e cores that are 60% the performance of a P core instead and put that thread on E cores.

The currrent architecture has to juggle HT cores, e cores and P cores, and AFAIK that's not very efficient/overly complex. Disabling HT also generally lowers latency.

HT is able to exploit more than just pipeline stalls as you aren't able to issue an instruction(micro ops to be exact) to all Execution Units in one single. So you could use other threads to exploit that.

But it comes with area cost due to needing more logic and registers, duplicated logic and registers for some structures while partitioning others. As you can have things like Reorder Buffers which might be statically partitioned to each thread(say 256 in total split in two 128 entry ones) or competitively shared(each thread uses as much as it needs - but more logic is needed to implement that). In essence, HT/SMT implementations can vary in perfomance and a lot of attributes.

You generally do get more out of it though.

I believe that Intel has kinda abandoned it in their consumer chips more due to scheduler. Thinking on stuff like Intel Application Optimization, it seems like there is quite a bit of performance on the table from a number of applications not being scheduled properly. This is even worse in cases like Meteor Lake where you have 4 types of cores, the normal P-core core, the HT core, the E-core and the LP E-core.

Because from a pure performance perspective, it does not seem like designing without HT had any benefit to Lion Cove, except maybe(and not even sure about that) for mobile.

 

[SL2]

Also p*ssmark.

AMD Ryzen 9 9950X vs Intel i9-14900KS vs Ultra 9 285K vs Apple M2 Ultra 24 [cpubenchmark.net] by PassMark Software

 

[AnarchoPrimitiv]

For about 30 seconds, before it melts down . Jokes aside, I hope intel comes back swinging and keeps competition going.

I see a lot of people saying something similar....do they not realize that Intel still holds the overwhelming majority of x86 marketshare? There's no "back swinging" here....Intel still dominates the market, still has absolute top of mind awareness in the minds of the vast majority of consumers (ask random people in a best buy if they've ever even heard of AMD), still is much larger than AMD in every measure besides stock price (i.e. overall budget, R&D budget, total sales, number of employees, etc)....

Intel still domiates the relationships with OEMs, still keeps AMD out of the premium laptop models, etc.....Intel is still very much the dominant party in this duopoly, and in fact, it would be better for consumers in general if they lost more marketshare to AMD (based on the belief that an even 50/50 split in marketshare would be the beat for consumers).

 

[Lionheart]

Give us a 10/12 P core no E core solution Intel.

 

[kawice]

Drop in multithreading is expected, since hyperthreading give +30-35% in best case, -3-5% in worst case.
I'm more interested in Arrow Lake Refresh gain or next gen.

This is just rehashed, canceled Meteor Lake Refresh. Nothing to see here.

Give us a 10/12 P core no E core solution Intel.

There is a rumor that such CPUs will be released next year, Bartlett Lake, rumor does not mention if they will be for Desktop though.

 

[AnotherReader]

~10% better ST would be great, considering 14900K is already extremely fast ST, the X3D chips have similar gaming performance but that's because of cache. They're actually significantly slower in ST.

Higher single threaded performance isn't enough on its own to translate into higher gaming performance. Zen 5 is the most recent example. I suspect that the gaming performance of the Golden Cove line of cores owes a lot to the memory subsystem too. Don't forget that they have significantly larger L2 caches than Zen 4. The P cores in the 14900k have access to 52 MB of cache outside their L1 caches. In addition, Intel's integrated memory controller makes better use of faster RAM than AMD.

 

[TomWeng]

Apple M3 series looks great. Is anyone use MacBook Pro to play the games? lol

Also p*ssmark.

AMD Ryzen 9 9950X vs Intel i9-14900KS vs Ultra 9 285K vs Apple M2 Ultra 24 [cpubenchmark.net] by PassMark Software

View attachment 366504

Truly, dude

 

[A&P211]

I see a lot of people saying something similar....do they not realize that Intel still holds the overwhelming majority of x86 marketshare? There's no "back swinging" here....Intel still dominates the market, still has absolute top of mind awareness in the minds of the vast majority of consumers (ask random people in a best buy if they've ever even heard of AMD), still is much larger than AMD in every measure besides stock price (i.e. overall budget, R&D budget, total sales, number of employees, etc)....

Intel still domiates the relationships with OEMs, still keeps AMD out of the premium laptop models, etc.....Intel is still very much the dominant party in this duopoly, and in fact, it would be better for consumers in general if they lost more marketshare to AMD (based on the belief that an even 50/50 split in marketshare would be the beat for consumers).

They are also the top dog in energy usage.

 

[Marcus L]

the X3D chips have similar gaming performance but that's because of cache. They're actually significantly slower in ST.

And in other news, water is wet.... this has been known since 5800x3D as they generally run lower clock speeds than their non-x3D counterparts as you know and you know why, do you think you are able to comment on one Intel/NVIDIA thread without mentioning AMD, I have yet to see it, AMD live rent free in your head

Be interesting to see how MT plays out, 8+16 small vs AMD's 9950X 16 large/32T, and of course the 9000 series x3D will bring AMD to parity or there abouts with Arrow Lake in gaming as has been the case with the last 2 gens, so pretty much same old all round, I am going to hold to my trusty old 5700x and RX 6800 for a couple more years by the looks of it as there is no compelling reason to spend £600-700 for a platform upgrade, not withstanding GPU for negligible performance increase, I guess we have had it good for the last few gens barring GPU prices going through the roof, another reason I'll be sticking with what I have until I can get a meaningful performance uplift (1.5-2x) from my 6800 for <£300

 

[phanbuey]

...

This is just rehashed, canceled Meteor Lake Refresh. Nothing to see here.

...

IDK about that - I think in games it's going to be pretty exciting. If you play cinebench all day maybe not, but the latency reductions, faster ram, better scheduler, no HT (usually HT isn't that great for games) and the ST boost, I think we might be in the 15%+ boost over 14900K in gaming, which will be nice.

 

[Klemc]

The first thing they changed is remove the Inside from Intel-Inside.

Now wecan't know where it is anymore ;p

 

[AnotherReader]

IDK about that - I think in games it's going to be pretty exciting. If you play cinebench all day maybe not, but the latency reductions, faster ram, better scheduler, no HT (usually HT isn't that great for games) and the ST boost, I think we might be in the 15%+ boost over 14900K in gaming, which will be nice.

Lion Cove, the P core used in Arrow Lake, is much better than Meteor Lake's P Core. Given the larger caches for the desktop variant, I'm cautiously optimistic that we'll see higher gaming performance than Raptor Lake though 15% sounds unlikely. In any case, we won't have to wait long for reviews.

 

[efikkan]

~10% better ST would be great, considering 14900K is already extremely fast ST, the X3D chips have similar gaming performance but that's because of cache. They're actually significantly slower in ST.

I would expect the performance gains to vary quite a bit from application to application, perhaps there will be some significant regressions too, as there are some significant differences in how the CPU operates despite small changes in execution units. But nobody should care about these synthetic benchmarks for purchasing decisions.

As you can see in the link posted by AnotherReader, the execution units have been split across more execution ports, which allows for higher peak throughput, but also requires more scheduling for the front-end. I think these changes, along with the removed constraints from dropping HT will pay off a couple of generations down the road. But also this would make it easier for Intel to add even more execution units to high-end workstation and sever parts, even more than they currently do. (Who wouldn't like workstation parts with more FPUs etc at the cost of a few hundred MHz peak boost?)

Give us a 10/12 P core no E core solution Intel.

It's called Xeon W.

 

[mago4]

Yeah essentially the thought is - HT threads/logical cores are 30%-40% of a real 'P-core' performance since it relies on pipeline stalls to execute code, replace it instead with e cores that are 60% the performance of a P core instead and put that thread on E cores.

The currrent architecture has to juggle HT cores, e cores and P cores, and AFAIK that's not very efficient/overly complex. Disabling HT also generally lowers latency.

Should we get more E Cores then ?

 

[Daven]

do they not realize that Intel still holds the overwhelming majority of x86 marketshare?

And how’s that working out for Intel? Last time I checked Nvidia, Apple, Microsoft, Amazon and Google are raking in the money and none of them see x86 as necessary based on their current and future product plans.

Heck, Qualcomm might even buy Intel just to use Intel products as space heaters in ARM team developer cubicles during cold San Diego winters.

lololololol