Ever since AMD (AMD) launched its 7nm laptops products earlier this year with eight cores, the voices about AMD taking a definite CPU leadership have increased, such as this recent article as just one example of many, lately also amplified by Apple’s (NASDAQ:AAPL) transition from Intel (NASDAQ:INTC) to its own silicon.
Intel, on its part, finally launched 10nm Ice Lake (in laptops) for the holidays last year, with four cores. But on the one hand, the product had to face competition from the new 14nm Comet Lakes with six cores Intel also launched, while on the other hand, even (per-core) performance didn’t improve much due to highly regressed frequencies from the newer 10nm process with its history of delays.
It is mindboggling that the author's premise is that the worst case scenario is already baked into the share price, and then doesn't even mention what the real worst case scenario is. In fact, AMD is barely mentioned in the article at all. The real worst case scenario is that AMD quickly takes most of Intel's market share, and Intel EPS turns meaningfully negative. This is entirely possible given the significantly better products AMD provides. I don't know how likely it is that most of Intel's share evaporates in the next 2 years, but it is a scenario that is on the table. The Worst-Case Scenario For Intel Is In The Price Intel cannot "insert" more cores into their 14nm+++++ cause they would blow the power budget. Intel cannot "insert" more cores into their 10nm++ cause its broken and has poor yields, it also clocks way lower than both Intels own 14nm and AMD's 7nm. Please tell me how Intel is supposed to "Push the pedal" as you write, with 14nm they already pushed the pedal trough the car mat, with 10nm++ they cannot push the frequency cause that node is still utterly broken for anything but low power laptops as it is now. And no, when you put core to core Intel does not win, AMD wins on the most important performance metric for the target segment - Power efficiency / Performance per watt. AMD Is Not Half An Intel
However, much of this is changing with Tiger Lake. While it retains its 4 cores (for now), leaked benchmarks suggest that it still manages to compete against AMD’s parts with double the number of cores. Quite a feat. It will also leverage Intel’s new Xe graphics architecture, which it will also use to deliver high-end gaming GPUs in 2021. Overall, then, Tiger Lake will be roughly on-par in performance (winning in some workloads, losing in others) despite half the number of cores, while taking a definite lead in graphics.
Looking at the second quote above, this article will counter the claims made in that comment: that 10nm is broken, that it clocks lower (than AMD's 7nm), that Intel can't insert more (10nm) cores, and that it loses in power efficiency. (AMD bulls asked - see italicized part - this article answers.)
So all in all, Tiger Lake is set to be a major update across performance, performance per watt, battery life, connectivity, display, media and graphics. Combine blowout per-core performance with blowout integrated graphics, and Intel should soon reclaim CPU leadership.
Looking at it from the AMD side, much of its ‘leadership’ simply comes from its higher core count, not per-core performance (a better metric of actual technology leadership).
From an investment perspective, looking at the recent trends, AMD investors are already taking AMD's purported "leadership" to the bank big time, and given the recent 7nm delay from Intel, seem to be assuming this will be the case for years to come. Instead, Tiger Lake will launch on September 2.
Note: this article is about laptop leadership, not the (much smaller) desktop segment.
Ice LakeAs a refresher, Intel launched Ice Lake (for laptops) last year as its first 10nm CPU. On the one hand, benchmarks confirmed Intel’s 18% IPC improvement claim: at the same clock speed, it has 18% higher performance (on average).
It also came with other platform upgrades such as Wi-Fi, media, display and graphics capabilities.
On the other hand, clock speeds regressed quite significantly to 3.9GHz (single core) on the highest-end 15W part, compared to frequencies as high as 4.9GHz on the 14nm parts. Of course, Intel’s 10nm node was the culprit for this, delivering lower performance than the optimized 14nm++.
After writing, Arm-based server start-up NUVIA published a blog with their analysis of the current CPU competitive environment (with their announcement of their own Phoenix CPU). It shows that Ice Lake and Zen 2 are actually very similar in both performance and performance per watt. However, Ice Lake has the edge as its performance curve extends further out.
Tiger Lake vs. Renoir
Tiger Lake is set to change this, on Intel's much-improved 10nm+ process (formerly called 10nm++), with clock speeds as high as 4.7GHz or 4.8GHz confirmed in leaked benchmarks. This will also make it readily faster than AMD’s Renoir which tops out at ~4.2GHz. Additionally, Tiger Lake will further improve Ice Lake’s performance-per-clock leadership, with an estimated further 5-10% improvement.
So how does it perform?
In this Geekbench comparison, the AMD Ryzen 7 4800U (Renoir) @ 1.80 GHz with 8 cores, 16 threads goes up against the 11th Gen Intel Core i7-1165G7 (Tiger Lake) @ 2.80 GHz with 4 cores, 8 threads. Both are the top parts of their line-ups. The reported frequencies are base clocks.
In single-core, the score is 5210 against 6737, a clear 30% lead for Tiger Lake. What this means is that when putting one Intel CPU core against one AMD core, Intel reliably takes the win. This should remove in one go any doubt over who actually has leadership technology. (In particular, higher clocks seem to contribute about ~15% of Tiger Lake's lead, with higher performance per clock contributing the other ~15%.)
Moving on to multi-core, remember that AMD has twice as many cores. This should give it an unsurmountable a priori advantage, in principle. Instead, due to the higher per-core performance, the score is 28651 against 23414, a 22% lead for Renoir.
This actually makes sense to a large degree: in laptops, performance is constrained by thermals (performance per watt), so, while not confirmed, the 8 cores might have clock down significantly (at least in sustained workloads).
(The multi-score of Tiger Lake is 3.5x the single-core score, close to 4x, while for Renoir the multi-score is 5.5x the single-core score. This indicates in the multi-score, Zen runs at ~2.8GHz while Tiger Lake runs at ~4.1GHz. )
This actually makes the score of Tiger Lake even more impressive: since power efficiency increases at lower clocks, this should provide another tailwind for the AMD cores to limit the extent to which they have to clock lower. Hence, this indicates Tiger Lake leads in performance/watt (efficiency).
Further, it should be noted that CPUs are much more than just the actual CPUs cores. For one, they contain integrated graphics. Here, AMD made the bold move to actually reduce the amount of hardware compared to its 16nm predecessor, from 11 to 8 compute units. For a company that prides itself with its console business and had historically always used its faster integrated graphics (“APUs”) as a selling point, this arguably went a step too far.
Meanwhile, Intel has claimed Tiger Lake’s new Xe architecture combined with 1.5x more execution units will allow it to double performance. Intel performance chief Ryan Shrout recently showed Tiger Lake achieving 30fps in Battlefield V. Tests have confirmed that this indeed doubled performance of Ice Lake, which itself had already doubled performance compared to 14nm GPUs.
When it comes to CPU performance, Tiger Lake seems set to be taking a firm lead over Renoir by as much as 30% per core due to a combination of leadership performance per clock and higher frequencies. Renoir takes a small victory in multi-threaded test simply by virtue of having twice as many cores, but its lead was far from 2x the performance.
In GPU performance, Tiger Lake seems to be taking a lead across the board. So when noting that Renoir actually reduced the amount of transistors spent on the GPU compared to its predecessor, it seems that AMD decided to go for 8-core CPUs to leapfrog Intel in peak CPU performance, but obviously this has come at the cost of GPU performance, so this was by no means a free lunch for AMD.
Chip designers only get a certain transistor budget, and AMD has made it clear it wants to improve its gross margins, so despite using TSMC’s 7nm process, it seems to have caused AMD to make critical trade-offs, namely losing its long-established GPU leadership for higher multi-core CPU performance. But compared to Tiger Lake, the difference is, at least in this test, surprisingly small despite having twice as many cores.
Looking at it from a technology perspective, Intel never actually lost its technology leadership, as Ice Lake showed last year: Sunny Cove is a more advanced micro-architecture than Zen 2. However, Intel could not translate this to product leadership as its frequencies were heavily regressed.
Since AMD launched its 7nm CPUs, sentiment concerning CPU leadership has started to shift significantly in its favor. Make no mistake, though, much of AMD’s performance advantage come from its strategy to throw more cores at the problem, not necessarily to deliver better technology (performance per core, and performance per clock).
That last statement is evidenced by Tiger Lake’s CPU competitiveness. As shown by leaked benchmarks, it indeed goes a long way to compete against AMD’s parts with double the number of cores. Achieving roughly similar (on-par) overall performance with 2x as many cores isn’t very telling of any leadership (and should be considered quite a feat for Intel).
Moreover, it has come at the cost of GPU performance. In Renoir, AMD actually reduced the number of graphics compute units from 11 to 8. Just as Intel is launching its new Xe graphics architecture with which it will also enter the discrete GPU space in due time.
Overall, more balanced CPU performance (crushing in single-thread and competitive in multi-thread) and much faster graphics, as marquee features of Tiger Lake, arguably bring overall processor leadership firmly back to Intel. And I haven't even discussed Thunderbolt, Wi-Fi 6, CPU+GPU DLBoost, security, media and display.
Looking further out, while AMD bulls may now point to Zen 3 being on the horizon (although for laptops that might be well into 2021), Intel bulls could point to Alder Lake being on the horizon just the same. Moreover, even before Alder Lake, there have also been leaks of an 8-core Tiger Lake (for the high-performance laptop segment), launching in Q1'21.
As I noted in the bullet points, ADM’s products simply contain ‘more’ cores, not ‘better’ cores. The latter point should concern AMD investors: the 8-core Tiger Lake leak shows that a core count advantage can simply be replicated. Indeed, in the server segment as well, Sapphire Rapids might have 72 cores according to a recent leak (vs. AMD's 64). Hence, this indicates that AMD does not have any moat.
(The only thing that AMD has going for it, obviously, is that all of Intel's products have been shifted (delayed) by at least a year due to the 10nm delays.)
So summing up, contrary to a recent, more subjective analysis of the Intel vs. AMD stock market events, here I let the numbers speak for themselves: one Intel core is ~30% faster than one AMD core. The multi-core test also shows that this speed-up has not come at the expense of power consumption/efficiency.
Put differently, as Tiger Lake will demonstrate, Intel should (continue to) be seen as the leader in CPUs, just like investors are valuing Nvidia as the leader in GPUs. Roar.