Monday, July 27th 2020
Intel Rocket Lake CPUs Will Bring up to 10% IPC Improvement and 5 GHz Clocks
Intel is struggling with its node development and it looks like next-generation consumer systems are going to be stuck on 14 nm for a bit more. Preparing for that, Intel will finally break free from Skylake-based architectures and launch something new. The replacement for the current Comet Lake generation is set to be called Rocket Lake and today we have obtained some more information about it. Thanks to popular hardware leaker rogame (_rogame), we know a few stuff about Rocket Lake. Starting off, it is known that Rocket Lake features the backport of 10 nm Willow Cove core, called Cypress Cove. That Cypress Cove is supposed to bring only 10% IPC improvements, according to the latest rumors.
With 10% IPC improvement the company will at least offer some more competitive product than it currently does, however, that should be much slower than 10 nm Tiger Lake processors which feature the original Willow Cove design. It shows that backporting of the design doesn't just bring loses of the node benefits like smaller design and less heat, but rather means that only a fraction of the performance can be extracted. Another point that rogame made is that Rocket Lake will run up to 5 GHz in boost, and it will run hot, which is expected.
Source:
_rogame
With 10% IPC improvement the company will at least offer some more competitive product than it currently does, however, that should be much slower than 10 nm Tiger Lake processors which feature the original Willow Cove design. It shows that backporting of the design doesn't just bring loses of the node benefits like smaller design and less heat, but rather means that only a fraction of the performance can be extracted. Another point that rogame made is that Rocket Lake will run up to 5 GHz in boost, and it will run hot, which is expected.
45 Comments on Intel Rocket Lake CPUs Will Bring up to 10% IPC Improvement and 5 GHz Clocks
This holds true for many ultrabooks, too, but spot-on, dude.
Did I miss something (outside of my typo when I said 'Zen4')?
I missed something... don't mind me!
That said, I don't think it will have any issues with Zen 3 in clocks or IPC. It will, yet again, be within reach/faster due to clock speeds in non heavily multi-threaded benchmarks. It will still cost more and use more power. Rinse...repeat. :)
Also, Zen 3 will be on the market before Rocket Lake so Intel would truly be in a position of attempting to retake the single-thread performance crown at that point. We haven't seen that scenario in 15 years.I find the leaks suspicious in light of this as well.
I find it disappointing that most other people in here are so gullible and believes pretty much any piece of "news" posted on Twitter etc. Nobody outside Intel have access to these chips yet to do a proper test. So regardless of what the actual IPC gains are, this piece of news is bogus.
Which is to say that GPUs would've gotten that level of performance eventually but Crysis lead & drove a race to the top much like what Vista & Apple forced, so if you're saying without Intel we'd never have had that kind of performance or efficiency leap then you're obviously forgetting the entry of ARM as well. Around the same time Intel decided to enter the mobile & tablet arena, with Bay Trail IIRC, so again Intel was led there & we'd have had great battery life & ultra portability maybe with ARM driven notebooks instead of Intel's o_O
The point being Intel has not been innovating over the last decade or so, they have mostly been followers rather than leaders in the tech arena. Their biggest leap or achievement of the last decade IMO would be Sandy Bridge.
There's no doubt intel has world class engineers on every level of their business... but the fab guys clearly dropped the ball, and whether the design guys can make up for that or not is one question.. but someone has to point out the crater left by that earlier ball dropping... I think intel has to take a close look at if they can survive being a top tier designer and fabricator. Maybe one division has outgrown the other. Top tier design needs the latest and greatest fabs, maybe thats not in house anymore. Hell maybe this leads to high end intel cpus being made at TSMC and Apple winds up partnering with intel's fabs to make their ARM on laptop/desktop work. Theres no other competitor in ARM in that product stack so maybe Apple wouldn't need to push for the latest node there. In this political climate, Apple is big enough to want to make something stateside and intel has 5 fabs here and Apple has the money to invest in them. So maybe its time for these two divisions inside intel to start relying less on each other.
It's a fun thought at least..
As of right now, TSMC would be the only other foundry capable of producing high-performance CPUs, but their high power production lines are all booked up. Intel could probably only get a few thousand wafers per month, not enough to make server chips, but could have been "enough" to cover a couple of K-models in to upper mainstream.
But going forward, dropping their own foundries is not an option for Intel. Even TSMC don't have nearly enough production lines optimized for high-power chips. So if Intel were to primarily rely on other foundries, they would have to reserve significant capacity like 4-5 years in advance.
While it is clear to most that Intel didn't put enough resources into R&D of 10nm, Intel also failed to build enough production capacity. Now that the yield issues are resolved, Intel still can't push nearly enough wafers to meet demand. This is partly due to higher product demands, but also because lithography takes more time than anticipated. Intel is in no short supply of money, so I think they should double-down on foundries and design their chips to target multiple (even external) nodes. The potential lost revenue is way more than the cost of doing so, and they can always sell spare capacity.Intel have not been afraid to work with ARM in the past. In fact, a while back the 14nm supplies were limited because a portion of the capacity were reserved for modems for Apple, and Intel apparently had assumed they would be moved on to 10nm by then.
But production capacity is likely not about failure to build but a planning question. 10nm - even with delayed 7nm - is going to have a short lifespan. Intel kept 14nm running as much as possible (bringing money in, shortages) and from rumors and some of their statements actually wants to skip 10nm in as many fabs as it can. Refitting from 10nm to 7nm might be faster but refitting from 14nm to 10nm also takes considerable time. I cannot say whether this is a good strategy. If they actually succeed in getting competitive 7nm process done it probably is.
I believe it was about two years ago Intel launched another low-power optimized version of their 22nm family. 22nm and 14nm will be kept online for many years to come, as will 10nm and 7nm. Most production lines like this are kept until they are no longer functional, as most of their cost are in the equipment tied to that node. And the demand for cost effective nodes from third parties is only growing, so Intel can probably recoup some of their expenses of 10nm there.