Intel Core Ultra 9 285K

[oxrufiioxo]

Rocket lake v2.0... Efficiency generally worse than Zen 4/5 even with 3nm vs 4/5nm...

Yeah, that's honestly the biggest bummer to me this has a node advantage and still ends up looking meh at best.

I can see why meteorlake skipped desktop it would have gotten murdered by everything in almost all workloads....

@phanbuey What a gaming monster.... thread didn't age well. I was optimistic as well and was hoping for 10% but even that low bar was off by 15% smh.

 

[Darmok N Jalad]

It’s really a mixed-bag launch at best. Efficiency is improved over RLR, but RLR was likely running an unsustainable voltage/frequency. Using the best node available (and better than the competition), but it can’t hit the clocks needed to come out on top. That doesn’t leave much opportunity to improve until the next node.

Lost in the noise of mediocre results is that this is a 1.0 desktop tile chip. That honestly is the biggest concern of all for me. I think we’re going to be spending a lot of this generation working through the bugs. Inconsistent performance, crashes, scheduling issues will all be present. I still don’t believe Windows has P+E figured out, and the performance regressions going to 24H2 only reinforce that.

Basically, even if this generation was winning the benchmarks, I’d be very hesitant to adopt it in our current “the customer is the QA” environment.

 

[529th]

When the walls fell, lol

CPU reviews have come a long way since early 2000s, great job TPU

 

[Pumper]

Power consumption is still way too high considering the performance, clocks, removal of SMT and the better node they're on, I don't know how they achieved this but it's not looking good.

It's OK once you take into account that Intel is idling at 25W lower than AM5. Take away those 25W from all the load power draw numbers and it won't look as bad, sometimes even better than AMD.

 

[Onasi]

Unimpressive, but I agree with W1zz on this one - this is Zen 1 all over again. A fully new architecture has a first iteration that’s not really fully great yet. I say let Intel cook now and see where they are 2 gens after at this point. I understand that having a double whammy of Zen 5 and now ArL being “disappointing” might be rather annoying to people who want to upgrade often, but it is what it is and, on the positive side, platform costs will go down in time. It’s whatever.

 

[Vincero]

The new CPUs are full of design changes, like removal of Hyper-Threading

@W1zzard - a shame there's no performance numbers using just the P-cores on the Raptor Lake and Arrow Lake CPUs for Cinebench - locking the core frequencies and comparing the ST/MT ratio would tell if they are pulling better numbers for the new P-cores without HT support vs the older ones with.

 

[DaedalusHelios]

No need to upgrade from my 7950X3D. I am glad I upgraded my monitor instead.

 

[Vincero]

It's OK once you take into account that Intel is idling at 25W lower than AM5. Take away those 25W from all the load power draw numbers and it won't look as bad, sometimes even better than AMD.

That doesn't make sense - surely that just points out how much worse the loaded efficiency is compared to AMD....

 

[Pumper]

That doesn't make sense - surely that just points out how much worse the loaded efficiency is compared to AMD....

How? Idle power draw number is for the full system. Load are for CPU only. AM5 platform is eating 25W more watts in all scenarios, it's just not represented in the load figures.

 

[Vincero]

How? Idle power draw number is for the full system. Load are for CPU only. AM5 platform is eating 25W more watts in all scenarios, it's just not represented in the load figures.

As per the review:

All power measurements on this page are based on a physical measurement of the voltage, current and power flowing through the 8-pin EPS CPU power connector(s), which makes them "CPU only," not "full system."

Now, technically there are some issues with that approach as its possible some of the ATX power lines from the ATX connector may supplement voltage lines to some parts of the CPU depending on implementation.

EDIT: I'm wrong...

Yeah, you're right - that 25W additional 'headroom' that AM5 uses doesn't just disappear. But on the flip side, Intel is compensating AMD by just blowing past that 25W figure when under full load.
The chiplet downside - not enough APU reviews for AM5 to compare with at TPU, but there is this:

Clearly AMDs AM5 APUs with a monolithic design do much better in this regard, and Intel's Foveros is a bit of a halfway house in that sense.
In a way this graph is a good reminder just how much power hungrier AM5 generally seems to be - I'm not entirely convinced the AMD promontory chipset chips are their best work in that regard.

 

[AnotherReader]

As per the review:

Now, technically there are some issues with that approach as its possible some of the ATX power lines from the ATX connector may supplement voltage lines to some parts of the CPU depending on implementation.

Idle power consumption is measured differently (emphasis added by myself).

Idle power usage is important for assessing energy efficiency, too. It reveals the power consumed when the system isn't actively in use—which is often the case for many computers. Unlike our other measurements, which report "CPU power only," these results are measured at the wall socket (220 V AC). The system is configured as detailed in the Test Setup section, i.e. with one SSD and one discrete graphics card installed.

 

[oxrufiioxo]

Unimpressive, but I agree with W1zz on this one - this is Zen 1 all over again. A fully new architecture has a first iteration that’s not really fully great yet. I say let Intel cook now and see where they are 2 gens after at this point. I understand that having a double whammy of Zen 5 and now ArL being “disappointing” might be rather annoying to people who want to upgrade often, but it is what it is and, on the positive side, platform costs will go down in time. It’s whatever.

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.

 

[rodrigorras]

Haven't Techpowerup tested AMD with the last Windows 24H2 updates???
Which brings 10% boost in AMD CPUS????

I don't see anything mentioned in the testbed

 

[Onasi]

@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.

 

[kapone32]

SamXe for me but for 449 tray edition

Well 7900X3D for me. I also remember the narrative putting the 7900X3D to $343 Canadian. I just watched the MSI Gaming livestream and the 9900X was one of the CPUs used.

 

[Space Lynx]

Haven't Techpowerup tested AMD with the last Windows 24H2 updates???
Which brings 10% boost in AMD CPUS????

I don't see anything mentioned in the testbed

You ask a good question, check out Techspot reviews, I think they did.

 

[AnotherReader]

Haven't Techpowerup tested AMD with the last Windows 24H2 updates???
Which brings 10% boost in AMD CPUS????

I don't see anything mentioned in the testbed

The review uses 23H2. The reasoning is elucidated in the conclusion.

When pairing Windows 24H2 with Arrow Lake, performance will be terrible—we've seen games running at 50% the FPS vs 23H2. One solution is to turn off Thread Director or disable the "Balanced" power profile, which is why we decided to use 23H2 for the time being. Last but not least, there are some driver issues and bluescreens when both a dGPU and iGPU are active at the same time.

 

[tfdsaf]

Seems quite good in most apps though, but there is a regression when it comes to certain games and apps, probably due to lack of hyperthreading. I think this is a step in the right direction for Intel in terms of being a lot more competitive in apps performance at a lower power draw, but we can clearly see that they do need to make some changes and optimizations for games.

I think its a decent processor line, but they would need to lower prices, same as the 9000 series vs the 7000 series. The thing is while it is a good future looking product, just like the 9000 sries from AMD the issues and certain regressions put a damp to it and they need to lower prices to account for that.

I think this new architecture is great for servers and AI and stuff like that, possibly for mobiles with the new e-cores and their very high performance, but for PC in terms of gaming there is a lot to be desired.

 

[AnotherReader]

@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.

Going quad channel would help with bandwidth limited workloads, but latency to DRAM would still be just as high and for many workloads that is the defining factor. There are two solutions: better prefetchers or more last level cache.

 

[oxrufiioxo]

@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.

I think it might be both latency to memory and bandwidth holding AMD back.

On the Intel side hard to say 8000+ didn't seem to help much but probably need mature low latency versions to know for sure.

I still think the IF is the biggest bottleneck on Zen5 and is in need of a massive upgrade for Zen6.

 

[ZenZimZaliben]

Wow, not a good showing. I hope the next revision improves this architecture, though it might be too late with AMD 9k X3d coming out.
Right now is the time to put some money down on shorting Intel. LOL. Welcome back to 2013 stock prices.

 

[mb194dc]

Another example of the tech stagnation we're facing and of the now massively diminishing returns each generation is bringing. A 5800x3d (or 5700x3d) will set your gaming performance for years more to come at realistic use case resolutions.

The comparison with ZEN 1 seems a bit silly to me. Intel had the consumer segment stuck on 4 cores before that, there were lots of performance gains and easy wins for AMD to get in productivity.

I can't see either of them thriving going forward because there's so little reason to upgrade now.

 

[oxrufiioxo]

The review uses 23H2. The reasoning is elucidated in the conclusion.

While I get that it also hampers Zen 4/5 performance somtimes by a little somtimes by a lot. Although the patched version of 23H2 isn't horrible I guess.

There are other reviews that tested amd on 24H2 and Arrowlake on 23H2 so it isn't an issue in the grand scheme of things but using somthing that benefits one architecture but not the other still isn't ideal.

I get it though w1z was put in a difficult position and wanted as much parity with platform as possible dealing with these half baked cpus.

 

[dgianstefani]

der8auer got 8800 MT working without going to gear 4. Some interesting performance results there when compared against lowest common denominator testing at 6000 MT, same RAM kit between rigs. Zen 4/5 have to run out of "sync" to get 8000 MT (1500/4000 MHz, compared to 2000/3000 MHz for 6000 MT), due to that, generally offers no gaming benefit.

It's a real shame he didn't also test at 6000 MT, to see how ARL scales with memory.

Definitely seems like some strange behaviour, depending on game. These 1% lows are crazy low, compared to Remnant 2 scaling against 14900K/Zen, where 14900K had slightly better averages and slightly worse 1% lows, and the 285K beat the 7800X3D, and considering P cores have more dedicated cache than Raptor Cove, buffering memory. I suspect it's early BIOS/software shenanigans. Looking forward to W1z's memory scaling articles, and later testing with more mature platform software and Win updates.

 

[Vincero]

Another example of the tech stagnation we're facing and of the now massively diminishing returns each generation is bringing. A 5800x3d (or 5700x3d) will set your gaming performance for years more to come at realistic use case resolutions.

The comparison with ZEN 1 seems a bit silly to me. Intel had the consumer segment stuck on 4 cores before that, there were lots of performance gains and easy wins for AMD to get in productivity.

I can't see either of them thriving going forward because there's so little reason to upgrade now.

I dunno - many feel AMD is just one good IMC implementation away from boosting performance... that and maybe addressing any IF / interconnect limitations between CCDs.

der8auer got 8800 MT working without going to gear 4. Some interesting performance results there when compared against lowest common denominator testing at 6000 MT, same RAM kit between rigs. Zen 4/5 have to run out of sync to get 8000 MT, due to that, generally offers no gaming benefit.

View attachment 368764

When you look at it like that, you have to admit Intel probably feels job done (in terms of the main aim) to bring much lower power use for a reasoable performance level. But it's not the norm/consistent.