Intel Core Ultra 9 285K

[Ravenmaster]

Well, that's Pat G on suicide watch

 

[Oberon]

Conclusion paragraph of the review you took a chart from.

 

[oZ65]

Conclusion paragraph of the review you took a chart from.

My guess is that he got drunk, like many in this forum.

 

[AnotherReader]

Yeah, everything this cpu is good at can be done better on a gpu.....

This does feel like Zen2 though decent application performance but meh gaming uplifts.


It preforms like a 7950x in w1z applications and gaming benchmark aggregate I fail to see how that is impressive after 2 years.... we are talking a couple percent better if we don't cheery pick what each is good at.....

Same problem the 9950X had really.

For me, the real sticking point is the low uplift for most multithreaded applications. I was expecting the scaling that we see in Blender, but for many, we get either no improvement or regressions, e.g. in file compression.

 

[JohH]

Where'd all the people from the Zen 2 days who only cared about rendering/encoding speed and power efficiency go??

It's not the same. Zen 2 3950X was nearly 2x the rendering performance of the competition (9900K) at the same power.
285K is sometimes +10% and sometimes -10% (depending on the renderer) at the same power.

Remember what Zen 2 looked like

https://www.techspot.com/articles-info/1940/bench/CB20-2-p.webp

Here's what the 285K looks like

https://tpucdn.com/review/intel-core-ultra-9-285k/images/cinebench-multi.png

The 3950X was an impressive step forward in MT even if it didn't translate to gaming. 285K is a small step forward in MT and it also doesn't translate to gaming.

 

[Oberon]

For me, the real sticking point is the low uplift for most multithreaded applications. I was expecting the scaling that we see in Blender, but for many, we get either no improvement or regressions, e.g. in file compression.
View attachment 368779

View attachment 368778

No SMT and compression loves thread count.

 

[R0H1T]

CUDIMMs should be compatible with Raptor Lake as well, and should similarly raise the clock frequency ceiling. Probably won't go as high as Arrow, but I think chances are good. Do you know if anyone tested it, now that some reviewers have a kit in their hands?

I don't think any regular Z890 boards will be doing 8000-8200MT on standard memory (without a clock driver) other than the Z890 Apex and its usual counterparts, though

Tested what exactly?

Intel Core Ultra 9 285K "Arrow Lake" Delivers Strong Linux Performance Review - Phoronix

www.phoronix.com

You're spending way too much for what's basically performance within margin of error!

 

[pokazene_maslo]

@W1zzard
I value very highly your reviews because they test the CPUs from almost all angles and they are highly trustworthy. However one angle is completely missing: long term load tests. Could you please add at least one page of tests that compare the CPUs on tasks that take at least 15 minutes for the CPU to complete? For example one 15 minute long game test, one 15 minute long video encoding, etc.? I'm fully aware that doing long term load test is very time consuming, however in the review most tests take less than 2 minutes for the CPU to complete. Such short tests highly favors CPUs that have very high short term boost. Or please consider releasing an extra review that compares some of the tested CPUs at long term load test. After all, you have been doing such specialized reviews that test some special aspect of the HW in the past. Like for example PCI-E scaling on the GPUs, memory speed scaling for the CPUs, etc.

Thank you!

 

[Dr. Dro]

Tested what exactly?

Intel Core Ultra 9 285K "Arrow Lake" Delivers Strong Linux Performance Review - Phoronix

www.phoronix.com

CUDIMM (the new DDR5 generation with a clock driver IC) on Raptor Lake and how far it scales vs. regular DDR5. This is how you're seeing a review with DDR5-8800

They should be, but will those platforms get BIOS updates?

Dubious.

I don't think it will be necessary, but if any motherboard will receive any love - it's the one I have

 

[AnotherReader]

No SMT and compression loves thread count.

That's a good point. However, even other applications don't scale as well as they should. Take NAMD which should scale pretty well because of the doubled vector throughput of the new E cores, but the uplift is relatively meager.

 

[TomWeng]

X3D is still the best gaming cpu in the world!

 

[Mr_Engineer]

Thank you for the thorough review as always @W1zzard

I guess my new rig is mostly decided and will be Ryzen 9800X3D especially if it's around 10% to 15% faster in gaming and productivity compared to the 7800X3D

 

[Darmok N Jalad]

The worrying thing for me is if silicon makers have finally hit a wall. Regardless of whether I want to upgrade or not I still love to see progress both AMD and Intel took 2 years to release new architectures which is much longer than past generation and seemingly only longer and longer between generation but both have very mixed performance and sub par gaming improvements to even regression.

I hate stagnation but the biggest current issues for both the 200 and 9000 series is pricing eventually both will be worth buying hopefully.

Well, there’s multiple things going on at the same time. At one point, performance gains were made by increasing clocks and improving IPC, then we added more cores. Now, adding more cores is diminishing returns. Higher clocks are not only harder to reach, but they also have a diminishing return. 400 more MHz on a chip like this is only 5%. That leaves architectural improvements, but some of those gains are offset by patching flaws. Now throw in P+E, chiplets, tiles, Thread Directors, preferred cores, parked cores, Windows Schedulers, and all that, and today’s CPUs are just begging for regressions and disappointments. There’s just so many more ways for things to go wrong.

@oxrufiioxo
Well, in terms of performance - probably not yet. But, as people mentioned already, the memory wall IS real. Zen 5 does suffer from inability to feed the cores fast enough. I am not sure how this can be overcome short of going to quad channel on desktop platforms and THAT would just drive the costs even higher, which I am not sure is what consumers want.

Perhaps some form of high-speed on package memory could supplement trips to the system RAM, like what Apple is doing with M-class SOCs? 8-16GB might help offset system costs, even if that means CPU prices go up some. That really seems like the future to me, where system RAM takes a step back for intensive workloads. Sure, eventually an app needs to hit system RAM, but perhaps the next major gain is to keep more tasks on-package. X3Ds do this already, and if they could clock as high as non-3DV chips, there would be no losses.

 

[kapone32]

The thing is, as usual, the X3D chips are only good for gaming. They're slower than the non X3D chips as well as the Intel competition in anything that isn't gaming, which happens to be what the vast majority of people in the world use CPUs for (not gaming). They're also more expensive.

Compare, say, a mainstream segment $330 9700X against the $310 245K, you're essentially getting 8% more application performance per dollar with the 245K, a more modern platform, and generally it's more efficient, while having slightly slower gaming prowess (with 6000 MT and early firmware). The 7800X3D is both more expensive and 10% slower in applications, but 20% faster in gaming (when the 245K is tested with slow memory), for $490.

What I'm seeing with the $590 285K is a CPU that compares favorably against its more expensive competition ($649 9950X), 30% more efficient in ST View attachment 368773

essentially the same in MT

View attachment 368772

25% less power in idle

View attachment 368774

...plus a better platform, but currently it's slightly slower in gaming despite being more efficient, when tested with memory 2000 MT slower than Intel's "sweetspot" 8000 MT.

View attachment 368775

Only good for Gaming? Are you not the one that bashed me because I did not get a 7950X3D because it was better at productivity? Anyone that has been on TPU for any length of time knows that you are very pro Intel though. This post is insane because these CPUs have nothing that stands out. Nothing. Power draw is still higher than AMD. Please don't even show X3D power draw there.

 

[dgianstefani]

This post is insane because these CPUs have nothing that stands out. Nothing. Power draw is still higher than AMD.

I can see you still have a certain level of wilful ignorance when presented with hard data.

 

[Dr. Dro]

Only good for Gaming? Are you not the one that bashed me because I did not get a 7950X3D because it was better at productivity? Anyone that has been on TPU for any length of time knows that you are very pro Intel though. This post is insane because these CPUs have nothing that stands out. Nothing. Power draw is still higher than AMD. Please don't even show X3D power draw there.

What he and multiple people on the forum told you is that the 7900X3D has inefficient topology. That didn't change. The 7950X3D doesn't have the same problem because it's 8+8, and the 9950X3D should finally do away with all major topology issues because it'll have 2 X3D CCDs. Zen 5 X3D will invalidate his point, especially if it's unlocked as rumored.

 

[AnotherReader]

Well, there’s multiple things going on at the same time. At one point, performance gains were made by increasing clocks and improving IPC, then we added more cores. Now, adding more cores is diminishing returns. Higher clocks are not only harder to reach, but they also have a diminishing return. 400 more MHz on a chip like this is only 5%. That leaves architectural improvements, but some of those gains are offset by patching flaws. Now throw in P+E, chiplets, tiles, Thread Directors, preferred cores, parked cores, Windows Schedulers, and all that, and today’s CPUs are just begging for regressions and disappointments. There’s just so many more ways for things to go wrong.

Perhaps some form of high-speed on package memory could supplement trips to the system RAM, like what Apple is doing with M-class SOCs? 8-16GB might help offset system costs, even if that means CPU prices go up some. That really seems like the future to me, where system RAM takes a step back for intensive workloads. Sure, eventually an app needs to hit system RAM, but perhaps the next major gain is to keep more tasks on-package. X3Ds do this already, and if they could clock as high as non-3DV chips, there would be no losses.

Apple's DRAM latency isn't stellar either (nearly 106 ns in that graph below). This is the raison d'être for that enormous L2 cache. AMD's MI300 shows the way, but I'm doubtful of such an implementation coming to us any time soon.

 

[dgianstefani]

What he and multiple people on the forum told you is that the 7900X3D has inefficient topology. That didn't change. The 7950X3D doesn't have the same problem because it's 8+8, and the 9950X3D should finally do away with all major topology issues because it'll have 2 X3D CCDs. Zen 5 X3D will invalidate his point, especially if it's unlocked as rumored.

Mostly true with 2 X3D, but there's still the issue of data having to do the expensive migration from CCD to IO die to CCD since there's no direct communication between the CCDs.

Zen 6 hopefully solves this with an active interposer.

 

[trparky]

Mostly true with 2 X3D, but there's still the issue of data having to do the expensive migration from CCD to IO die to CCD since there's no direct communication between the CCDs.

Zen 6 hopefully solves this with an active interposer.

If the rumors are true, then that's what's going to happen.

 

[Taraquin]

Mostly true with 2 X3D, but there's still the issue of data having to do the expensive migration from CCD to IO die to CCD since there's no direct communication between the CCDs.

Zen 6 hopefully solves this with an active interposer.

And I hope they seriously upgrades the I/O-die so it can run far better fclk

 

[Evrsr]

No, it didn't. Zen 5 is a significant improvement over Zen 4, but you wouldn't know it if you only considered gaming. What's clear is that the memory wall is making scaling ever more difficult.

Looking at the memory scaling results (and improvements without power limit), this replicates what Granite Rapids has achieved and seems to have a beast of a memory controller.

To me, it looks like the new Thread director is not playing well with the scheduler in the OS, just like with Alder Lake. We will have to see, there hasn't been a problem free CPU launch in quite some time.

 

[Psychoholic]

Should have held on to my 7800x3d and sold it now (Sold it when i got my 9950x)
Even though 9800X3d is out soon, could probably get quite a bit for a 7800x3d.

 

[oZ65]

I can see you still have a certain level of wilful ignorance when presented with hard data.

The day they add the clown emoji like on Steam will always be too late

 

[Evrsr]

@W1zzard
I value very highly your reviews because they test the CPUs from almost all angles and they are highly trustworthy. However one angle is completely missing: long term load tests. Could you please add at least one page of tests that compare the CPUs on tasks that take at least 15 minutes for the CPU to complete? For example one 15 minute long game test, one 15 minute long video encoding, etc.? I'm fully aware that doing long term load test is very time consuming, however in the review most tests take less than 2 minutes for the CPU to complete. Such short tests highly favors CPUs that have very high short term boost. Or please consider releasing an extra review that compares some of the tested CPUs at long term load test. After all, you have been doing such specialized reviews that test some special aspect of the HW in the past. Like for example PCI-E scaling on the GPUs, memory speed scaling for the CPUs, etc.

Thank you!

For most of these boards, there will be no drop as PL1=PL2=250W. Laptops is where you're likely to run into that.

 

[dgianstefani]

AMD's IMC is quite good - better even than Intel's at running high speed memory (just look at the APUs); the problem is routing the chiplet interconnect over the package, which severely limits the bandwidth and speed. Hopefully Zen 6 will finally be the iteration that moves to 2.5D packaging for the CCDs and IOD, which would very likely remove that bottleneck.

Ya it's better than Raptor Lake IMC, when it doesn't count, with APUs.

Agreed the issue is the interconnect.

Still surprised AMD didn't make a monolithic X3D CPU based off these APUs, instead of using the chiplet models with their inherent issues.