Intel Unveils Arrow Lake Desktop Processors, Promising Power Efficiency Gains (pcworld.com) 46
Intel has announced its new Arrow Lake desktop processors, marking a significant shift in the company's approach to chip design and power efficiency. The Core Ultra 200S series, set to launch on October 24, 2024, introduces a disaggregated architecture manufactured using TSMC's advanced nodes.
The flagship Core Ultra 9 285K boasts 24 cores (8 performance, 16 efficiency) and can boost up to 5.7 GHz, priced at $589. Intel claims the new chips offer comparable performance to their predecessors while consuming significantly less power, with reductions of up to 136 watts in some gaming scenarios.
Arrow Lake utilizes a tiled design, combining compute, GPU, SoC, and I/O components manufactured by TSMC and packaged using Intel's Foveros technology. The compute tile is built on TSMC's N3B process, while the GPU tile uses TSMC's N5P, and the I/O and SoC tiles are on TSMC's N6. Intel's Roger Chandler stated, "Arrow Lake picks up the mantle of Raptor Lake's top-end gaming performance and delivers parity performance at about half the power."
Intel acknowledges that gaming performance may lag slightly behind the previous generation, with a 5% deficit in some benchmarks compared to the Core i9-14900K. The company is positioning Arrow Lake as a balanced solution, emphasizing power efficiency and content creation capabilities. The new processors require a new LGA 1851 socket and Z890 chipset, necessitating motherboard upgrades. Memory support extends to DDR5-6400, with XMP profiles potentially reaching DDR5-8000.
The flagship Core Ultra 9 285K boasts 24 cores (8 performance, 16 efficiency) and can boost up to 5.7 GHz, priced at $589. Intel claims the new chips offer comparable performance to their predecessors while consuming significantly less power, with reductions of up to 136 watts in some gaming scenarios.
Arrow Lake utilizes a tiled design, combining compute, GPU, SoC, and I/O components manufactured by TSMC and packaged using Intel's Foveros technology. The compute tile is built on TSMC's N3B process, while the GPU tile uses TSMC's N5P, and the I/O and SoC tiles are on TSMC's N6. Intel's Roger Chandler stated, "Arrow Lake picks up the mantle of Raptor Lake's top-end gaming performance and delivers parity performance at about half the power."
Intel acknowledges that gaming performance may lag slightly behind the previous generation, with a 5% deficit in some benchmarks compared to the Core i9-14900K. The company is positioning Arrow Lake as a balanced solution, emphasizing power efficiency and content creation capabilities. The new processors require a new LGA 1851 socket and Z890 chipset, necessitating motherboard upgrades. Memory support extends to DDR5-6400, with XMP profiles potentially reaching DDR5-8000.
intel is really giving up on the foundry business? (Score:2)
Re:intel is really giving up on the foundry busine (Score:4, Informative)
TSMC is now a monopoly... that can't be good. I thought intel had a good process after the stellar 13x series
Samsung also has two 3nm processes and has been shipping 3nm chips since 2022 with one of them and 2024 with the other. Intel could have gone with Samsung from what I understand, although there might have been technical reason behind the scenes reasons of which I am not aware. The reason might be volume as TSMC has 2 processes in production and 2 processes to begin production in 2024 and 2025.
Re:intel is really giving up on the foundry busine (Score:5, Informative)
TSMC is now a monopoly...
TSMC is a monopoly at the bleeding edge. But many chips don't need the latest process.
that can't be good.
It is not good, but it isn't clear what can be done about it. Subsidies for competitors aren't working.
I thought intel had a good process after the stellar 13x series
Intel is a node behind and lacks capacity, but has better yield.
Intel's best strategy is to spin off the foundry as a separate business. There is no good reason for design and foundries to be lashed together.
Re: (Score:2)
"There is no good reason for design and foundries to be lashed together."
When I was at Intel, design-process codesign got us one speed bin per quarter with no architectural changes.
Putting the design and the fab together means you can get exactly the process you need for the products you can actually sell.
Re: (Score:2)
Putting the design and the fab together means you can get exactly the process you need for the products you can actually sell.
Except when the fab can't deliver. Then, Intel calls TSMC.
The "tick-tock" process you describe worked well in the past but has failed for at least five years.
Re: (Score:2)
"There is ONE good reason for design and foundries to be lashed together."
Re: (Score:2)
13 series was stellar until Intel discovered that the chips were broken and spent the next 2 years covering it up and lying about it. Externally Intel has blamed this on improper "vdroop" predictive power delivery algorithms built into their processors, but it's easy to see how the cost of microcode update vs complete replacement would drive them to this conclusion.
The current move could be a result of understanding that their chips are instead corroding/degrading regardless of PM, but it could also be TSM
Re: (Score:2)
The performance of Intel 13th and 14th gen chips definitely drooped after the BIOS updates were installed.
Re: (Score:2)
Re: (Score:2)
That's called living in the fast lane. Most industries you don't know you've been fucked over for a decade.
Re: (Score:2)
If memory serves, Arrow Lake was intended for Intel's 20A process. They cancelled it to refocus on 18A, but it's not ready yet. So they had to move Arrow Lake to something else.
As for why they moved it to TSMC 3nm instead of Intel 3, which is supposedly in high volume production already, I'm guessing it was just a capacity thing. Intel booked a bunch of TSMC capacity for GPUs that didn't end up selling, so they probably had lots of unused capacity at TSMC but not in their own fabs.
Re: (Score:2)
No. Intel has not given up on the foundry business. They are investing heavily into the foundry business, trying to develop it into a separate business unit able to take in orders from outside designers in addition to Intel's own production.
That said... they don't have the capacity to produce the chips they are designing at this point.
Their old fabs are good for producing older chip designs -reliably and affordably. They have not built any fabs capable of producing current designs -they tried and failed.
Re: (Score:2)
TSMC is now a monopoly... that can't be good.
Intel is not closing down the fabs. It merely splits them up into a separate company.
This act doesn't diminish TSMC competition. In fact, it introduces more competition because until now Intel fabs were used almost exclusively by Intel, but now they'll compete in the open market.
Some gaming scenarios (Score:2)
That's kind of funny. Yes our processor uses much less power on a single threaded game.
Water Lake (Score:5, Funny)
Re:Water Lake (Score:5, Funny)
Re: (Score:3)
With Intel chips, people need to submerge the chip in a lake just to keep the chip from overheating.
Re: (Score:2)
Arrow Lake Ultra 9 CPUs are 125W base/250W max TDP vs ~350W max on the 14900k (depending on limits set by the motherboard, which sometimes were more generous than that). Intel chose to design for much better power efficiency in this generation. Given that 14th gen Intel was giving Threadripper desktop CPUs a run for their money, I think Intel made the right call.
Re: (Score:2)
No, Intel is NOT giving Threadripper a run for its money, unless you are looking at stupid single-threaded performance.
Re: (Score:2)
Intel hasn't been competitive with AMD performance since right after Piledriver, and they've been even worse from a price:performance standpoint.
Re: (Score:2)
Re: (Score:3)
I believe Intel names their chips after lakes in Oregon (they are an Oregon based company), so I would question the people who name lakes in Oregon.
Those CPUs are a pain for virtualisation. (Score:3)
Those multiple CPU types are a pain for virtualisation. I wonder if qemu is going to be eventually able to deal with them properly like give the host 1 performance CPU and 1 efficiency CPU. Or maybe give the guest vm a single or a few kvm64 CPU which will automagically split the load between those different core types on the host? Is anybody aware if any virtualisation solution is able to handle those multiple CPU types properly yet?
Re: (Score:2)
For Qemu, I would expect that you can do that on host-kernel level. Or is this Qemu-KVM you are talking about? That is more tricky...
Re: (Score:2)
Yeah most setup like proxmox use the kvm kernel module along with qemu otherwise performances are abysmal.
lsmod | grep kvm
kvm_amd 172032 64
kvm 1249280 31 kvm_amd
irqbypass 12288 1 kvm
ccp 131072 1 kvm_amd
# ps axw | grep qemu /usr/bin/kvm -id 111 -name www.example.com,debug-threads=on -no-shutdown-chardev socket,id=qmp,path=/var/run/qemu-server/111.qmp,server=on,wait=off -mon chardev=qmp,mode=control.......
2985809 ? Sl 116:33
Re: (Score:2)
I use qemu-no-kvm for some isolation jobs. But that is very different regarding CPU scheduling as it essentially is just a program being run. Performance is something like 1..5% native, but for some things that is just fine.
Hmm. If the current kernel CPU schedulers do not do this for qemu-kvm, it may take a bit for it to get implemented.
Re: (Score:2)
Yeah I use qemu for all kinds of stuff as well similar to what you stated. Funnily, back in ~2000. VMware had a kernel module even before the CPUs supported hypervisor modes but performance was much abysmal as well compared to CPU supporting hypervisor modes.
Intel promises vs. Intel reality: Not good (Score:3)
Better get something solid from AMD...
Re: (Score:2)
Re: (Score:3)
I recently moved on AMD for the first time for our proxmox cluster and I'm very happy so far:
AMD EPYC 4244P 6-Core Processor (1 Socket)
6 cores and 12 threads. That CPU just came out this year. It runs 60+ vms smoothly with 128GB RAM, 2x1GB nvme for OS raid 1, backups and what not on ext4 and 2x4GB nvme for a zfs mirror pool for the vm images.
Re: (Score:2)
Ooooh, that is an impressive load for such a small CPU to run well.
Re: (Score:2)
Well, you need to do some fine tuning, mainly sync=disabled on the zfs pool rpool/data used by vms and lxc containers and arc cache=metadata for zfs.
I only have sync=default for the database guests. Apart from that, no L2, no ZIL nor anything else.
I even enabled deduplication on the whole zfs pool without any problems, it just makes good use of the CPUs who never went above 80% so far globally. arc is set to 32GB max and uses no more than 24GB so far on one host, the others being below that.
But yes, that cp
Re: (Score:2)
We used to run everything on a 72 core Intel CPU but we used spinning rust so the nvme move might have been an important factor too.
Those servers had 256GB RAM and a 92GB arc cache set to cache everything and performance sucked compared to what we have now mainly due to IO writes. The CPUs were barely used ever since they were always waiting on IO.
Anyway, I am very happy with the move to the new hardware. We all slowly learn I guess... :)
With such a scenario (virtualisation) using Intel/AMD, the first concern should always be IO not CPUs.
Re: (Score:2)
I don't know why anyone would buy Intel these days. Even if they do work and last more than a year, AMD parts are just better price and performance.
Re: Intel promises vs. Intel reality: Not good (Score:2)
If an Intel CPU works for you, then why not? Intel's death wouldn't benefit anyone, even AMD.
Re: (Score:2)
Because supporting a crook and unethical company is immoral?
And ultimately bad for everyone?
Just saying.
Re: (Score:2)
That's the problem, they don't work. They self destruct after a while. Maybe they fixed it, who knows. They have not demonstrated competence lately.
OEMs are even more screwed because OEM CPUs only get a 1 year warranty, so most of the ones dying are not going to be replaced by Intel. The OEM will get the whole machine back for repair/replacement.
Re: (Score:2)
Re: (Score:2)
Yucki (Score:3)
Eww the same performance as before with the need to get a new board and cpu to offset a few dollars in power consumption. Unless you plan to keep it for a long time there is no payback on this stupid move. Why upgrade? For the planet? Seesh.
Re: (Score:3)
As I stated above those different CPU cores introduce new complexities as well. Doesn't work well yet with virtualisation hosts as far as I can tell. Also more complexity usually introduces new problems. Heck, I think I'd still prefer packing a machine with MIPS or ARM cores if you are looking for low power consumption rather than going with different types of cores.
Re: (Score:1)
Re: (Score:3)
What does that leave Desktop with? -5% performance. Woohoo. I'm sorry, I run a 4090, I don't care if my CPU sips less power at idle or under load.
Yeah, that juxtaposition of top-end gaming performance and power savings with no performance gains felt really odd.
Intel's Roger Chandler stated, "Arrow Lake picks up the mantle of Raptor Lake's top-end gaming performance and delivers parity performance at about half the power."
On top end gaming performance, they're bragging about performance parity? Power savings while running GPU's that use 450W??? Sure, that's fine, but it's not a huge selling point.
Who’s willing to trust them? (Score:1)
Still waiting... (Score:1)