Some sources have told us some very interesting things going on inside Intel, in respect to their future products, nodes, and direction. We do want to emphasize that we totally trust our sources, but as with all rumors take this information with a grain of salt; some of this information could change (we are talking about things planned for 2022 after all), but we expect most of it to hold up. What we have been told primarily relates to Intel’s 10nm and 7nm nodes, their upcoming graphics, and Intel’s general strategy for the period leading up to 7nm.
There will be no desktop CPUs that use the 10nm node
Today, we can say our source also says 10nm will not be seeing the desktop. Intel has reportedly given up on 10nm as a process and accepts that it will never be in high volume. It’s fine for smaller mobile products, but for CPUs that are expected to have 10 cores at a minimum along with high clock speeds, it’s just not good enough. Any plans for 10nm on the desktop were either speculative or simply never existed to begin with. But if by some chance there are 10nm desktop CPUs, they will be at best mid range since Intel cannot achieve high end performance due to low clock speeds on the 10nm node.
Intel is of course denying this but previously the company said 10nm development was going great and they’re reaffirming that it is still going great; it’s clear to anyone familiar with 10nm that it’s not at all going great, however. At best Intel could offer a token 10nm CPU for the desktop, a la Cannon Lake, just for PR. The volume and performance just isn’t good enough for the desktop.
Our source states Intel is basically going to be buying market share until they can get 7nm desktop CPUs out and the company has reportedly allocated billions of dollars to this end. This would seem to be the case especially when considering this leaked Intel slide Jim showed in his most recent video. Hopefully this does not entail Intel rerunning its mid 2000s strategy of simply paying OEMs to not use AMD. But if this is Intel’s new strategy, AMD might be in some trouble, though the company’s diverse set of customers (from Sony and Microsoft to Samsung) could reduce the impact of Intel’s “financial horsepower.”
In the course of its transformation from a PC centric company to a data focused one, Intel will apparently assign very little value to the desktop beyond using it for marketing purposes and improving the brand. Intel is primarily concerned about mobile right now, which is a big money maker for Intel, whereas the desktop is no longer growing fast enough or sometimes at all. This isn’t a huge surprise but this is probably a significant factor in why Intel’s desktop platform will be seeing rehashed 14nm Sky Lake derivatives until 7nm arrives.
Intel’s mobile ambitions with Tiger Lake
As said before, Intel is focused on the mobile market right now, especially laptops. Intel sees Nvidia and AMD as its primary threats in this arena, and to this end Tiger Lake has been designed to edge both of them out of the laptop.
Tiger Lake’s main appeal will be its graphical performance. The CPU part is apparently somewhat disappointing, but the graphics are actually impressive; it should be twice as fast as its predecessor and good enough for the 1080p resolution at 60 FPS. If Intel could do this in just one package, with competitive power efficiency, and at a low price, it could certainly damage Nvidia, which Intel considers to be its main rival according to our source (something that is reflected within Intel’s recent presentation for investors).
Tiger Lake certainly won’t be fast, of course. But laptops have never really been focused on speed anyways. What Intel wants is the money Nvidia is making in the gaming laptop business (and they do make quite a bit of money there), just like AMD. We should expect the dynamic in this market to change significantly.
On the other hand, Tiger Lake is a 10nm processor which apparently suffers from the same issues Ice Lake does, but it will be improved to some degree. We also weren’t told when exactly when it would launch, but we would expect a window in 2020. While Intel does not consider AMD to be much of a threat to Tiger Lake, Renoir is approaching sooner rather than later in 2020 and Intel is already struggling against 12nm based Picasso APUs. Tiger Lake is something Intel does not want to delay on, at any rate.
AMD is putting the pressure on Intel’s server CPUs
Our source had very little to say on Intel’s server and datacenter CPUs, but we did hear about how AMD is affecting things, namely with Rome. For example, we were told that more than a fifth of an OEM’s server business was Rome CPUs and that AMD is very well positioned right now in the datacenter.
One of the reasons why this is the case is because AMD is making Rome its primary concern. While there may not be a silicon shortage, desktop CPUs are apparently getting the short end of the stick in regards to supply since Rome is in such high demand. Also concerning desktop CPUs, the 3950X is getting delayed because AMD is having a hard time justifying setting aside dies for the 16 core CPU. Rome is in much higher priority.
We also got some figures on Milan and Zen 3’s performance. Apparently Zen 3 will deliver a significant IPC bump on average and a slight improvement in clock speeds (perhaps a few hundred MHz). We can’t say with certainty exactly how much more IPC Zen 3 is set to deliver (yet), but according to Red Gaming Tech, who has had some decent leaks in the past, it is greater than 8%. Again, take with a grain of salt, especially considering some of Zen 2’s features were actually borrowed from the Zen 3 design, so the room for IPC improvements has been diminished.
10nm is getting pushed aside in favor of 7nm
One of the most important things we were told was how Intel was emphasizing development of the 7nm and the products that would come with it. Similar to how AMD bet on TSMC’s and GlobalFoundries’ 7nm nodes (the latter of which never saw the light of day), Intel is also apparently going all in on their own 7nm node, by buying a great deal of EUV equipment in preparation for a ramp in 2021 and also by preparing a plethora of 7nm based processors.
10nm has been such a problematic node for Intel that the company has decided 7nm development will be a much better use of resources. There are apparently many issues with 10nm Intel still hasn’t solved, and instead of using all that time to fix 10nm, Intel would rather speed up development on 7nm.
As stated before, 10nm will almost certainly never see the desktop due to its poor performance and low volume. 7nm on the other hand will be on the desktop rather quickly and we were told to expect it in 2022 with the possible codename Meteor Lake. With Rocket Lake being the last 14nm desktop CPU for the high end, we should see a 7nm desktop CPU immediately succeed it. We were told nothing in respect to performance, however.
As for server, our source couldn’t tell us very much, but we did confirm Sapphire Rapids could be the last 10nm server CPU (we were also told it could be 14nm but that that is very unlikely) and it seems Granite Rapids will succeed it on 7nm. We were also told Intel is hoping Foveros (their multi die stacking technology present in Lakefield, a mobile CPU) will do something for their server business, but that is more of a theory than a reality at the moment. Intel will want to have a competent 7nm server CPU to face off against Genoa and whatever the Zen 5 based successor will be called.
Finally, for mobile, we were just told Alder Lake would be the last 10nm mobile CPU. Considering Intel has a great emphasis on mobile in its strategy, however, we should expect more competent 7nm CPUs for mobile than for the desktop.
According to our source, 7nm is launching in 2022. We are aware that Intel is saying 2021 for their 7nm Xe GPU, but we were told 7nm would launch in 2022 with a ramp in 2021. Additionally, there’s very little talk about the node at Intel (something that is not limited to just 7nm, even 14nm architectures are hard to find out about presently). We don’t know exactly what 7nm will look like, but since Intel is hoping it will make them competitive again, we should expect a decent improvement versus 10nm at least.
Intel’s strategy: wage a price war and wait for 7nm
Finally, we have Intel’s general strategy for the next few years until they can get 7nm out, and this strategy involves cutting prices (which we’ve already seen with Cascade Lake), utilizing their “financial horsepower”, and preparing for 7nm products. This strategy isn’t supposed to be very effective against AMD but instead is supposed to head off Nvidia which is Intel’s greatest threat in graphics and mobile. This strategy also naturally puts Intel in a position where they will be making far less money than they have recently.
With AMD’s current form, Intel cannot simply cut prices and win a price war, especially in regards to desktop and server CPUs. Intel might be able to make something work with GPUs where AMD has the thinnest of margins, but overall AMD is mostly safe thanks to extremely thick margins on Rome and Matisse as well as having a very diverse set of customers.
Nvidia, however, is a different beast. They are apparently very vulnerable in mobile and in the datacenter which is why Intel is targeting Nvidia first. Tiger Lake is supposed to capture Nvidia’s traditionally profitable low end GPU market in mobile (as well as prevent AMD from gaining a foothold in laptops), mostly by offering these products at a ridiculously low price. Nvidia has also failed to make many good friends in the market, which is why companies like Google are making their own AI processors to get away from Nvidia’s exorbitant prices.
A price war in the name of competition might be highly effective against Nvidia, which was used to moderate success by AMD a few times in the past. Since Intel can afford to spend billions of dollars doing this (unlike AMD), Intel could keep the pressure on Nvidia for an extremely long period of time, and this is all thanks to Intel’s “financial horsepower.” By offering ridiculously cheap graphics with decent performance, Intel is hoping to capture some of Nvidia’s most profitable segments of the market.
On the other hand, our source believes AMD has a great chance in the market up until 2022 when Intel will supposedly be getting back on its feet with 7nm CPUs and GPUs. Even with a price war, AMD still has several advantages with their server CPUs (supply volume and especially power efficiency), decent resistance in mobile with Renoir coming in 2020, and has many customers Intel could not possibly touch (such as the console makers and Samsung); for these reasons AMD will weather the storm much better than Nvidia is expected to.
After 7nm comes out, it remains to be seen whether or not Intel can live up to its own expectations on bringing its CPU, GPU, AI, and other technologies together. Our source doesn’t have a very positive outlook on Intel’s current internal affairs, but things can change between now and then.
We went through quite a bit of information just there. But these are the key points:
No real plans to bring 10nm to the desktop due to low volume and performance; 14nm will remain the primary process for desktop
Intel is putting more resources into pushing 7nm to production than it is into fixing 10nm
But Tiger Lake, which is 10nm, will be important for Intel in the mobile market thanks to its good gaming performance
Intel will be focusing on mobile in particularly during this period, with the primary objective being to defeat Nvidia; AMD is not a large enough threat at the moment
Intel has no competitive server CPUs on 10nm and is waiting for 7nm to try and retake the lead; this is also due to Milan being more powerful than expected
Again, take with a grain of salt as with all rumors. We have great confidence that our sources are correct on this, of course, but we don’t expect to be believed at face value nor can we guarantee that Intel won’t make some adjustments to its roadmaps (like adding a token 10nm desktop CPU). Intel may very well launch a 10nm CPU on a desktop platform, but we can absolutely guarantee you it won’t be a whole product stack that scales from the low end to the high end.
If we hear more from our source on this matter, we will update accordingly.