The Intel Execution Thread [2021]

edit: I was under a wrong impression, see postings below I am under the impression, that Cypress Cove is much more Ice Lake than Intel wants us to believe: I hope they did not forget to backport the cache to inclusive again with the smaller L2. With Tiger Lake, L2 and L3 caches are non-inclusive. That's normally a trade-off between size, latency and power (density). With 10 nm, they wanted to boost frequencies and had the transistors density to invest in larger sizes to make up for higher latencies.

With Cypress Cove, that balance fell apart in places.

 

Maybe the NDA compliant reviews will help understanding the performances.

 

Not sure what you mean by "more than Intel wants us to believe", since it's literally Sunny Cove cores from Ice Lake ported over to 14nm.

 

Oh, never mind then. I was under the impression, some news outlets were (more or less consistently) reporting on Cypress Cove being a backport from Willow Cove. Maybe I was googling too old rumors.

 

Yep, too old rumors. Pretty much first real leaks already confirmed the caches to match Sunny Cove, which instantly ruled out Willow Coves

 

The problem I have with this is that the time when Intel - Intel god dammit, of all the companies! - could have made such excuses as "it's not an original design" and "well, what can we do with a backport" has long passed.
They have an army of engineers and all they can do between 2015 and 2021 is a bad backport of a part from 2019? Why even do it then? RKL's only potential reason to exist would be to be on par with Zen3 in <=8C parts. If it's not then why even launch it? A bigger die (lower margins, less dies on the market), less cores in the high end, higher power draw - and the result is more or less similar to what CML has been showing a year prior.
This is a fail on all fronts really. And seeing this I can't help but just assume that ADL and MTL will be the same - and nothing will really change up until Pat will clean the house and make a new CPU core from scratch. Which if they couldn't do a good backport in three years means what? 2030?
I dunno what Intel is doing but they need to seriously get their shit together.

 

I guess with the graphics and AVX-512, they got more valuable chips coming out of 14nm until better manufacturing is free.

 

Adl won't have to be backported, so I guess it will be fine. And rocket lake is already old, since it's based on ice lake ?

I don't know, I blame the people wanting this thing out more than the engineers having to tape this monster together.

 

It's actually a bit interesting. It's based on Ice Lake which was launched back in 2019, however what makes it worse is that Ice Lake itself was delayed due to issues with 10nm process, so Ice Lake itself from a design stand point is likely not really a 2019 design/release and more likely to have been 2018. What makes this even more interesting is that wasn't the first 10nm delay either, so it'd be interesting to see what Ice Lake's planned targets were at inception versus what actually happened.

But this really high lighting the core issue that Intel's process problems are causing. Had this been on a working a 10nm released in say 2018 and therefore going against Zen+ and Zen2? Completely different narrative in the CPU space.

 

Well if Intel can actually get their Rocket Lake CPUs on the market for DIY consumers at MSRP, then it'll be infinitely better than the AMD's higher-end 5000 CPUs that can't be found anywhere.
At least it's a new desktop architecture that uses PCIe 4.0 and not the 10 year-old PCIe 3.0.

Though at this point I wonder if most DIY-ers willing to invest on a PC to last wouldn't just be better off waiting for late 2021 / early 2022 to get a PCIe 5.0 + DDR5 platform.

 

The "better of waiting" game is always valid but you never buy a thing in the end

At one point the question is "does it respond to my needs ?" imo.

 

This is correct, but playing the waiting game is unavoidable at the moment. In practice, for most people the time period that counts is not between "right now" and when the new hardware releases. It's between the "when the current hardware become available" and when the new hardware releases.

Rocket Lake will supposedly release in late March, but the truth is that it is a very late DDR4 + PCIe 4.0 platform and Intel is supposed to bring an early DDR5 + PCIe 5.0 platform at 10nm with Alder Lake in H2 2021.

Unless Alder Lake is delayed, Rocket Lake is not just a late adopter of old tech, it's also supposed to have a very short life to be succeeded by a new and significantly more evolved platform pretty soon.

 

Yes I agree, Rocket Lake timing to the market is weird. I hope (for Intel and consumers waiting) it doesn't mean that ADL is late..

 

Fully expect ADL to lose to Zen3 while launching against Zen4 at this point. Intel seem to be unable to make good CPU these days.

 

In the laptop space, Tiger Lake is already competitive with Zen 3 in single core performance, and may not be far off in multicore (depending on the base clocks of the upcoming 8 core parts). So even a modest boost in IPC may be sufficient to put them past Zen 3.

 

The terribleness of the Intel options seems overblown.

https://www.techpowerup.com/review/amd-ryzen-9-5950x/17.html

 

Right, the problem is that Zen4 will have up to 16 big cores while ADL will have only 8 of those which will hopefully be competitive with Zen3. Not a hard extrapolation from the data we have.

They aren't "terrible", they are simply worse than what AMD make in every way (with the very useful exception of AVX512) and this is rather embarassing for Intel considering that it's been five years already since they have started to lose ground to Zen.

 

Yes, so they will lose in multicore performance at the high end (the status quo since Zen 2). They are probably betting on battery life, single threaded performance, and gaming performance. Optimistically, the rumoured 20% improvement in IPC (in line with the improvement from Skylake to Ice Lake) could see them being competitive in the last two elements at least.

 

The problem is I doubt that anyone cares about "battery life" when it comes to 16C CPUs but a lot of people care about the multicore performance on these parts.
This just highlights the issue which Intel has at the moment, and it's not the process even but the inability to make a product which the market wants.
They've been riding on their 4C parts for so long that they seemingly forget that it's not them who decide how a good product should look like.

 

So you assertion is that no one should bother buying anything higher than a 4c8t ~4.4ghz CPU because at 1440p over all benchmarked games it doesn't make any difference? Yeah let's just go back to 2015 and let Intel continue stagnating the CPU market since there's no benefit.