News Intel Launches Arrow Lake Core Ultra 200S — big gains in productivity and power efficiency, but not in gaming

[bit_user]

I mean, if you want to call Devil's Canyon a separate product from Haswell, you could even argue it was three generations on that socket. Not that anyone really bought Broadwell 5775C!

That wouldn't be a very winning arguement. Haswell Refresh (Devil's Canyon) was concurrent with the Broadwell 5775C.

I don't really care that much. I know others do, but if I upgrade to a new CPU? Yeah, I'm getting a new motherboard to go with it.

Most of the time, I think there's a fairly narrow argument for upgrading from one gen to the next. The only way it makes sense for most people is if you do an entry-level budget build and then you swap the CPU for one that's a higher tier in the next generation. So, like going from an i5-13600K to an i9-14900K.

I have an i5-12600. At one point, I entertained the idea of upgrading to an i9-14900K, but I don't really need it and for my current plans with that machine, it would be a waste.

 

[bit_user]

according to TPU the 9950x is 3.5% faster than the 7950x in the application aggregate.

Those are launch-day benchmarks. We know they have issues. I'd put more faith in what the benchmarks show when it's compared to Arrow Lake, since they'll hopefully have redone the Zen 5 scores.

 

[TheHerald]

Those are launch-day benchmarks. We know they have issues. I'd put more faith in what the benchmarks show when it's compared to Arrow Lake, since they'll hopefully have redone the Zen 5 scores.

I don't expect even a 1% increase vs zen 4 but let's go with it. How much faster will it be? Even if the difference gets 3 times as big it's still roundabout the same as has refresh.

I don't know why people think they have issues. What kind of issues?

 

[bit_user]

I don't know why people think they have issues. What kind of issues?

Described here, sort of:

https://www.tomshardware.com/pc-com...built-in-benchmarks-versus-custom-game-scenes
​

Basically: core-parking issues, CCD-to-CCD latency, sticky scheduling state (i.e. when swapping CPUs without a Windows reinstall), and branch-prediction requiring admin privileges.

 

[thestryker]

Also, the hot spot is off-center. That and being smaller should pose more challenges for heatpipe-based coolers (or, at least the vast majority which lack a vapor chamber in the base).

Intel moved the socket a bit, and the reduced load ILM is going to be shipping with some motherboards which should have some impact on default behavior (though there will be new contact frames). MSI has already announced offset mounting available for their coolers so I suspect that will be more common as well.

 

[TheHerald]

Described here, sort of:

​

https://www.tomshardware.com/pc-com...built-in-benchmarks-versus-custom-game-scenes​

​

Basically: core-parking issues, CCD-to-CCD latency, sticky scheduling state (i.e. when swapping CPUs without a Windows reinstall), and branch-prediction requiring admin privileges.

None of that will really impact something like cbr / blender and the likes.

 

[King_V]

You know, I really kinda thought okay, a small drop in performance for the potentially vast reduction in power draw, sounds like it's worth it.

Yeah, I know they said "up to" 165W decline. Then I saw in "total system power draw" and I saw the table where the TDP etc., was and . . almost the same?

What kind of shenanigans are going on here? The message they're saying with the claimed power reductions, and the tech specs, are definitely at odds with each other, which is disappointing, to say the least.

 

[Mattzun]

The productivity gains at lower power settings look nice.
Most OEM workstations seem to be based on 14th gen 65 watt processors with DDR5 4400-5600 memory.
A combination of better low power performance and official support for faster RAM could be huge for standard workstations bought for productivity tasks

I can't see any business justification for spending resources on CPUs with faster gaming performance until faster GPUs hit the mainstream.
The 5090 is the only GPU where we might see significant differences between modern CPUs at 4k for the next two years.
A 13700K and a 7800X3D have roughly similar performance at 1440P and 4K on a 4090 and the 5080 is supposed to have comparable performance

Arrow Lake might be significantly better than 14th gen for gamers buying an OEM PC.
OEMs might actually ship with DDR5 6400 instead of DDR5 5200/5600 and the PCs are less likely to be thermally throttling.

 

[redgarl]

Anyone believing in an Intel powerpoint is a fool...

You need to remember that Intel is sucking up to 450W in MT, meaning that 25% is only a 110W power reduction. It is still a power hog.

Anyway, just by looking at the chart, 25% better efficiency doesn't cut it still.

 

[thestryker]

These seem very interesting, but not necessarily what I was hoping for. All of the technical aspects are right in line with expectations, but the absolute performance in the middle ground (greater than 1T, but not fully MT) seems somewhat lacking. To an extent this makes sense with the ~6% clock regression with claimed ~9% IPC gain.

Getting power consumption for lighter workloads under control is enough for me to not consider prior Intel generations, but AMD has been doing this since Zen 4. It'll be interesting to see how all the different levers for tweaking and overclocking impact performance. It also has full CUDIMM support so once these modules actually hit the market high speed memory should be easier to run.

At the end of the day this certainly doesn't seem to be a slam dunk product, but it's absolutely the right direction compared to prior Intel generations.

 

[TheHerald]

Anyone believing in an Intel powerpoint is a fool...

You need to remember that Intel is sucking up to 450W in MT, meaning that 25% is only a 110W power reduction. It is still a power hog.

Anyone not trusting the slides is a fool.

Any CPU is sucking up to 450w if you remove the power limits man. Stop being silly. The graph is quite clear, the 285k at 125w will be as fast as the 14900k at 250w.

 

[redgarl]

Jesus Christ... Why... just WHY... comparing against the 7950x3D in productivity... they are obviously misleading us. The 7950x3D is already 15-20% slower than the 9950x in MT alone.

Basically the 285 is on par with a 9950x while consuming more power if Intel 23% claim in efficiency is real.