Intel Core Ultra Arrow Lake Preview

[W1zzard]

Intel has just lifted the curtain on its highly anticipated Arrow Lake microarchitecture, we also learned about the processor models that are launching this month. In our preview we explain technical details, such as why Hyper-Threading was removed and how Foveros fuses multiple dies into a seamless, high-performance product.

Show full review

 

[BoggledBeagle]

This slide looks good:



Intel more energy efficient than AMD? Honestly hard to believe.

Speaking about energy savings, I wonder why they did not post the values for each of these games:

 

[Chaitanya]

Looks like socket still has similar questionable design choise as seen previously.

 

[oxrufiioxo]

Wow, rip gamers..... 2024 has been rough for gamers.

Hopefully the 9800X3D can save the day.

Kinda disappointing considering intel has a node advantage with this.


But it does have AI to save the day said nobody in their right mind ever lol.

 

[dgianstefani]

Wow, rip gamers..... 2024 has been rough for gamers.

Hopefully the 9800X3D can save the day.

Kinda disappointing considering intel has a node advantage with this.


But it does have AI to save the day said nobody in their right mind ever lol.

Similar gaming performance to current top performers, with an apparent efficiency and MT advantage over anything on the consumer desktop market is nice. Having great IO, along with an NPU is a perk too. I am a bit puzzled to see gen 1 Arc though.

I wouldn't be surprised to see the U5/U7 models have more performance in gaming than their i5/i7 predecessors too, despite 14900K-285K not having much of an uplift. Realistically most people aren't using $700 CPUs and $1600 GPUs in their gaming rigs, so advancing the mainstream is great, considering the competing X3D chips essentially start at the high end unless you're willing to buy an AM4 CPU in 2024.

 

[BoggledBeagle]

Wow, rip gamers.....

This is so funny. How many gamers are CPU limited to notice slightly worse gaming performance?

 

[oxrufiioxo]

Similar gaming performance to current top performers, with an apparent efficiency and MT advantage over anything on the consumer desktop market is nice. Having great IO, along with an NPU is a perk too. I am a bit puzzled to see gen 1 Arc though.

I wouldn't be surprised to see the U5/U7 models have more performance in gaming than their i5/i7 predecessors too, despite 14900K-285K not having much of an uplift. Realistically most people aren't using $700 CPUs and $1600 GPUs in their gaming rigs, so advancing the mainstream is great, considering the competing X3D chips essentially start at the high end unless you're willing to buy an AM4 CPU in 2024.

2 years later its pretty meh... Also it sounds like multiple AM4 X3D chips will be faster in gaming....

This is so funny. How many gamers are CPU limited to notice slightly worse gaming performance?

There are multiple games I'm already cpu limited in with X3D 7000 processors. I agree though most people probably are not.

 

[AnotherReader]

The gains in multithreaded performance are lower than expected. Given that IPC gain over Raptor Cove is rather modest, the lower clock speed and loss of SMT are limiting the gains from Skymont. Gaming performance is surprisingly low which should remind everyone that single threaded performance gains don't always translate into better fps; games tend to be more memory limited than other applications.

2 years later its pretty meh...



There are multiple games I'm already cpu limited in with X3D 7000 processors. I agree though most people probably are not.

Don't worry; Unreal engine 5 will ensure that your 4090 returns to being the bottleneck.

 

[Daven]

There's a few interesting tidbits from this release:

Going from 6 P-cores and 4 E-cores to 8 P-cores and 12 E-cores (2P+4E extra) costs an additional $85.
Going from 8 P-cores and 12 E-cores to 8 P-cores and 16 E-cores (4E only extra) costs an additional $195.
So the Ultra 7 is the best deal.

Given the very different architectures and packaging strategies between Ryzen (Zen 5) and Core Ultra (Arrow Lake-S), it's looking like they will be remarkably similar in power usage and performance.

Clock speed competition may finally be dead. With all the lower and lowered clocks between ARM, performance, dense and efficiency cores, I hope we never see a race to unstable, degrading clocks just to win some points in a benchmark over the competition. IPC, cache and good thread scheduling should be the goals for today's CPUs.

 

[oxrufiioxo]

The gains in multithreaded performance are lower than expected. Given that IPC gain over Raptor Cove is rather modest, the lower clock speed and loss of SMT are limiting the gains from Skymont. Gaming performance is surprisingly low which should remind everyone that single threaded performance gains don't always translate into better fps; games tend to be more memory limited than other applications.

It's gonna be an awkward situation of if you want MT you probably should grab a 285k and if you care about gaming you grab a 9800X3D...... so basically nothing has changed

Don't worry; Unreal engine 5 will ensure that your 4090 returns to being the bottleneck.

I guess that's true and 5.3/5.4 fixed cpu performance for the most part i can still be cpu limited in them with DLSS quality though at least some of the MP titles.

It's more disappointing to me because I was hoping for a banger of a gaming cpu for my backup system looks like it'll end up being a 9800X3D or whatever replaces it.

 

[kondamin]

I would like to see a power sipping 8 e core chip with 2 memory channels

 

[dgianstefani]

I would like to see a power sipping 8 e core chip with 2 memory channels

They sell them for NAS and cheap laptops already.

 

[AnotherReader]

It's gonna be an awkward situation of if you want MT you probably should grab a 285k and if you care about gaming you grab a 9800X3D...... so basically nothing has changed


I guess that's true and 5.3/5.4 fixed cpu performance for the most part i can still be cpu limited in them with DLSS quality though at least some of the MP titles.

It's more disappointing to me because I was hoping for a banger of a gaming cpu for my backup system looks like it'll end up being a 9800X3D or whatever replaces it.

The Core Ultra 7 265K is the sweet spot of the Arrow Lake lineup; it has all 8 P cores and is significantly cheaper than the 285K. For productivity in well multithreaded applications, the 285K should beat the 9950X but it's closer than I expected.

They sell them for NAS and cheap laptops already.

Are there any that are Skymont based? Skymont is a big upgrade over Gracemont.

 

[oxrufiioxo]

The Core Ultra 7 265K is the sweet spot of the Arrow Lake lineup; it has all 8 P cores and is significantly cheaper than the 285K. For productivity in well multithreaded applications, the 285K should beat the 9950X but it's closer than I expected.


Are there any that are Skymont based? Skymont is a big upgrade over Gracemont.

I'm not a fan of how intel nerfs the L3 on the 7 never have been but I guess we will see how that plays out also how much the 9800X3D smacks it at gaming if it's 10-15% faster that's a pretty large gap for gaming. These days we might not even get that difference with a new generation smh.

Also I didn't pay much more for my 7950X3D than what the ultra 7 will cost....

Are there any that are Skymont based? Skymont is a big upgrade over Gracemont.

With Zen6 it was server type workloads that got the huge upgrade with Intel the damn E cores two things IDGAF about lol for those that love both good for them smh.

 

[ir_cow]

DDR5-6400 JEDEC is looking nice . 6000-6400 will get cheaper as that will the highest volume.

Some more information from Der8auer video

6400 for 2-Slot MB with CUDIMM
5600 for 2-slot MB with UDIMM
5600 for 4-slot MB with UDIMM
4800 for 4-slot MB with UDIMM (Single Rank).

Looks like we will see a contact frame again because Intel didn't fix the bending problem. If you want to run high speed memory, better get a contact frame.

 

[AnotherReader]

I'm not a fan of how intel nerfs the L3 on the 7 never have been but I guess we will see how that plays out also how much the 9800X3D smacks it at gaming if it's 10-15% faster that's a pretty large gap for gaming. These days we might not even get that difference with a new generation smh.

Also I didn't pay much more for my 7950X3D than what the ultra 7 will cost....



With and it was server type workloads that got the huge upgrade with Intel the damn E cores two things IDGAF about lol for those that love both good for them smh.

Given the gap between the 14900K and the 7800X3D, I expect the Zen 5 X3D SKUs to be 10 to 15% faster than the 14900K and the equivalent 285K.

 

[oxrufiioxo]

Given the gap between the 14900K and the 7800X3D, I expect the Zen 5 X3D SKUs to be 10 to 15% faster than the 14900K and the equivalent 285K.

Same and that sucks in my book considering we've had faster/efficient gaming cpus for over a year now.

I'm curious what the latency to memory is on these and if that is the bottleneck apparently going with 8000+ mem brings it closer to the 14900k in gaming performance.... The way they tested it the 285k was 5% slower than the 14900K 31 games.

 

[kondamin]

They sell them for NAS and cheap laptops already.

WHat i mean is that I want an update to the n100 which is single channel memory

 

[Tomorrow]

Intel more energy efficient than AMD? Honestly hard to believe.

Not really considering it's essentially made by TSMC this time.

 

[AnotherReader]

Same and that sucks in my book considering we've had faster/efficient gaming cpus for over a year now.

I'm curious what the latency to memory is on these and if that is the bottleneck apparently going with 8000+ mem brings it closer to the 14900k in gaming performance.... The way they tested it the 285k was 5% slower than the 14900K 31 games.

Intel's packaging solution is more expensive and tightly bound than AMD's so I expect the latency hit for a DRAM hit to be higher than its monolithic predecessors, but lower than Zen 4/5. The ability to use faster RAM should improve performance in cases where it's limited by bandwidth and with 24 cores there should be quite a few of those cases.

 

[oxrufiioxo]

Intel's packaging solution is more expensive and tightly bound than AMD's so I expect the latency hit for a DRAM hit to be higher than its monolithic predecessors, but lower than Zen 4/5. The ability to use faster RAM should improve performance in cases where it's limited by bandwidth and with 24 cores there should be quite a few of those cases.

I think the architecture is super interesting for sure but having a node advantage and being more expensive to make while offering worse gaming performance and mild MT uplifts is disappointing.

It's weird becuase some things are super impressive in a vacuum. The power consumption, the tiles and how they are grouped, general perfomance considering they ditched HT, but on the other hand gaming and generational perfomance uplifts are pretty weak after 2 years and it's really only Zen5 being meh not making this look worse.

Honestly I just hope it actually works all of the time lol we don't need degradation gate 2.0 lol.

 

[Daven]

This has got to be the most nonsensical slide in the universe.

 

[oxrufiioxo]

This has got to be the most nonsensical slide in the universe.

View attachment 366974

It's marketing man they come up with all kinds of BS we will have reviews soon.

The slides for Zen5 were really bad hopefully these are at least better than that lol.

 

[hsew]

The prospect of reviving BCLK overclocking is actually pretty exciting.

 

[Steevo]

I will wait for an actual review before I spout off too much.


The number of admissions by Intel is telling.

Their nodes were fluff, TSMC is still more advanced and all the BS talk was just that, BS.

Power draw...... simply amazing what a better node will get you. No doubt Intel made some awesome discovering in power domain management and it's going to help a lot with this series.

Chiplet based design. The reality is they copied AMD, they used a tile like AMD did to mount GPUs with a silicon interposer, they can call it whatever but like the Barenaked Ladies sang, it's all been done before....

The most exciting parts are the removal of HT and the Ecores on the ring bus. I hope they put their big boy pants in and approached it with a "how can this F us" security mindset, I say this with the best of intentions as I think they have a winner with this design if it doesn't need 5 patches to slow it down and validate security on code.... They may have just placed themselves a generation ahead of AMD on architecture and planning. AMDs only move to counter is a little.BIG core where a LOT of resources can be turned off on a big core until needed or a set of stripped down cores with commonly used hardware. The proof is in the pudding that HT/SMT is is dying a rapid death and the silicon space it used is better suited to cache and or branch prediction.


The best part is the pricing, it tells almost the complete story.