Once again, Intel’s banking on a high clock speed to drum up excitement.
![What if Comet Lake, but longer? This wafer litho image gives us a bird's eye view of the layout of one of the Comet Lake S-series CPUs, featuring two more cores than prior generations.](https://cdn.arstechnica.net/wp-content/uploads/2020/05/COMET_LAKE_S_WAFER_STATIC3_CLOSE_UP_RGB_V01_HIRES-800x334.jpg)
So far, 2020 is a tough year for Intel CPU fans—in both senses of the word. The newest generation of Chipzilla’s desktop CPUs have arrived, and Intel is scrambling to find ways (mostly involving overclocking) to make them look good compared to AMD’s 7nm Zen 2 parts.
The 10th generation Core, Pentium, and Celeron parts follow the trend established by Intel’s recent laptop H-series launch: they’re old process technology tweaked to within an inch of its life, and Intel still isn’t delivering any hard performance numbers which could be compared directly to the competition’s.
Performance
For the most part, Intel’s pre-launch benchmark data looks like what they provided us for Comet Lake H-series laptop CPUs—razor-sharp focus on unqualified raw clockspeed, and a healthy smear of Vaseline on the lens when looking at real performance. Once again, we’re seeing single-core turbo speeds on the highest SKUs in excess of 5GHz—and a noticeable veer away from hard performance data that might be directly compared to AMD’s 7nm Ryzen CPUs.
To Intel’s credit, we do at least have some generation-on-generation performance deltas this time. We still have almost no hard performance numbers, however—let alone direct comparisons with AMD Ryzen. Intel is also still very keen on comparisons to a “three year old PC.” Presumably, they’re hoping that people who aren’t up-to-date on the Intel vs AMD rivalry will just think, “yes, that sounds nice, let’s go ahead and upgrade this year.”
Intel’s marketing this time around leans heavily on their optimization collaboration with game development companies as well as—if not more than—the new CPUs’ raw performance. But without direct head-to-head comparisons against Ryzen processors, it’s impossible to ascertain whether the game companies’ glowing testimonials to Intel’s help apply to Intel vs AMD, or merely to newer generations vs older.
The marketing focus on gaming and gamers with this series of CPUs is both extreme and predictable—single-threaded performance is very nearly the last place Intel can hope to compete directly against its AMD competition, and it’s hard to find anyone who cares much about that statistic outside the gaming world.
Modern content creation and compiling workloads tend to be massively multi-threaded and focused more on throughput than latency. Gaming—particularly the highest levels of competitive gaming—tends to be more heavily influenced by input and response latency. If a single core can chew through a game’s main loop faster, an elite player may be able to shed a precious extra few milliseconds of reaction time.
Trying to evaluate the real gaming benefit of the fastest possible CPUs is made more difficult since none of the typical industry metrics measure it well—frames per second is a throughput metric, not a latency metric. Making matters worse, FPS is typically shown as a mean average, rather than a median or mode. It focuses heavily on the graphics dimension of the gameplay loop while effectively ignoring input and logic.
Quibbles about metrics aside, some of the new Comet Lake S-series presumably got better framerates than their Ryzen equivalents—on the three games Intel tested, at least. We’re forced to guess this, based on the fact that Intel lists a Ryzen 9 3950X configuration on one of the “fine print” slides.
We can also guess that this wasn’t a particularly compelling difference and it may not hold up widely among most games. Another of the fine print slides notes that Intel’s USA marketing tagline “World’s Fastest Gaming Processor” cannot be used in a staggering list of countries—in those countries, it’s replaced with “Elite Real World Performance” or “Intel’s Fastest Gaming Processor.”
Yes, yes, we know—we’re reading marketing tea leaves, here. But without hard data, they’re all we have to go on.
Architecture
First and foremost, if you want a 10th generation Intel desktop CPU, you’re going to need to buy a new motherboard. The new CPUs are designed for an LGA1200 socket, rather than LGA1151. These won’t be a drop-in replacement for anyone.
Beyond that, we’re mostly just seeing overclocking tweaks. Turbo Boost Max 3.0 dynamically selects and boosts the fastest two cores on a given individual CPU, and there are lots of new knobs for hardcore overclockers to twiddle in the Extreme Tuning Utility—such as the curious ability to enable or disable hyper-threading on individual cores, not just the CPU as a whole.
The i9 series (and only the i9 series) also gets a new Thermal Velocity Boost feature, which can eke out an additional 100MHz for short periods of time while the CPU is generally running cool.
Intel has also thinned the CPU die on some models, allowing for a thicker integrated heat spreader and therefore potentially better cooling to the hottest-running areas on-die. We don’t have much specific information on the new thinning process—and we don’t know for sure which models it applies to.
The officially supported RAM speed has also increased a bit, from DDR4-2666 to DDR4-2933. It still lags behind Ryzen 3000’s DDR4-3200.
Intel’s 10th generation desktop offers a whopping 32 models on launch day. Yes, 32. It’s difficult to make sense of this gigantic smorgasbord, but we’ll do our best. First and foremost, let’s talk about modifiers: K series CPUs are overclocked and unlocked, F series have no integrated GPU, and T series are tuned for lower TDP.
Which variants are available depends on which base SKU you’re looking at, with some base models having all five possible variants, and some being available in the base version only.
Celeron and Pentium Gold
Celeron and Pentium Gold are the cheapest processors in Intel’s 2020 desktop lineup, ranging in price from $42 to $86. These are all dual-core parts; Pentium G offers hyper-threading, while Celeron does not.
Celeron and Pentium G operate at base clock only, with no Turbo boost or any of the associated technologies. They all offer integrated graphics, and the base models offer fixed clockspeeds ranging from 3.4GHz to 4.2GHz.
The two Celeron models and the lowest-end Pentium Gold are the only CPUs in the lineup to offer UHD Graphics 610—the entire rest of the lineup, all the way up to the i9 models, offers the faster UHD 630 iGPU.
i3 and i5
There are three base models apiece of i3 and i5 CPU; the i3 models offer four cores plus hyperthreading, and the i5 models offer six cores plus hyperthreading. Base clocks range from 3.6GHz to 4.1GHz, with boost clock ranging from 4.3GHz to 4.8GHz.
None of the i3 or i5 models offer Turbo Boost Max or Thermal Velocity Boost. The i3s have neither K nor F variants; the lowest i5 offers an F-series GPU-free variant, and the fastest i5 offers both K and KF variants.
These middle-of-the-road CPUs range in price from $122 for the i3-10100, to $262 for the i5-10600K.
i7 and i9
The highest end is, oddly, the least complex section of Intel’s desktop lineup this year, with only one i7 and one i9 base model. Both i7-10700 and i9-10900 come in all five possible variants, though—base, F, K, KF, and T. i9-10900 is a 10-core/20-thread CPU, and i7-10900 is 8-core/16-thread. Prices range from $298 for the i7-10700F, to $488 for the i9-10900K.
The base clock rates for non-K models are 2.8GHz for the i9, and 2.9GHz for the i7, with Turbo Boost Max 3.0 of “up to 4.8GHz” for the i7, and “up to 5.1GHz” for the i9. K-series variants bring the base clock up another 900MHz for both i7 and i9, and TBM 3.0 comes up another 300MHz for the i7 K-series, and 100MHz for i9 K-series.
The i9 can eke out an additional 100MHz boost clock from its Thermal Velocity Boost technology, bringing its top single-core clockrate to “up to 5.2GHz” for the base model and “up to 5.3GHz” for the K-series.
Thermal Design Power
Intel’s 2020 desktop CPU lineup comes in four distinct TDP ratings: 35W, 58W, 65W, and 125W.
All T-series CPUs, regardless of their base family, are rated at 35W TDP. The T-series CPU variants share core and thread counts with their non-T base models, but these variants underclock them significantly to reduce cooling and power budget. The price for a T-series part is currently identical to that of its base model.
Pentium Gold and Celeron CPUs—not counting the T variants—are all rated for 58W TDP. i3, i5, and i7 CPUs—not counting the T variants, or the K and KF variants—are all rated for 65W TDP. Finally, all K and KF models are rated for 125W TDP.
Please remember these are “Intel TDP” specifications; they’re sufficient to run all cores at base clock and no further. Significant additional cooling budget is strongly advised for anyone who expects to get the most performance out of these processors, even if they don’t plan to overclock.
Anandtech reports some motherboard manufacturers are planning to support 500W or more of CPU power consumption, in the attempt to keep all-core turbo frequencies as high as possible, for as long as possible.
Conclusions
There aren’t any real surprises here. Just as it did with the H-series laptop parts, Intel is banking on high raw clockspeed in its increasingly elderly 14nm lineup to impress potential buyers, while it tries to refine its newer 10nm process.
If you’re just a die-hard Intel person, you can expect to see minor generation-on-generation performance gains and enough of an improvement from Intel’s newest favorite metric—the “three year old PC”—to justify an upgrade if you’ve got the money burning a hole in your pocket. It won’t be a cheap upgrade, though, since you’ll need a new motherboard to go with your new processor—and maybe newer, faster RAM as well.
Professional gamers might—and we wish to stress might—see a small potential win in Intel’s extremely high single-core clock speeds, here, with a corresponding slight decrease in latency in demanding games’ core loop. It’s difficult to recommend Intel’s desktop lineup over AMD’s this year in any other capacity, and honestly we have trouble keeping a straight face over the potential minor gains in gaming response either.
Let’s focus for a moment on Intel i7-10700K vs AMD Ryzen 7 3700X:
CPU | Price | Core/Thread Count | Performance vs i7-9700K |
Intel i7-10700K | $374 | 8 cores / 16 threads | “up to 18%” faster (HEVC Software Encoding) |
AMD Ryzen 7 3700X | $330 | 8 cores / 16 threads | 115% faster (Cinebench R20 multi-threaded) |
We’ll be the first to admit that this comparison is very sloppy: HEVC video software encoding and Cinebench R20 CPU rendering are absolutely not the same workload. But this is as close as we can get with the sparse data that Intel has given us—and it’s probably close enough to give a pretty good overview of what’s going on here.
Intel’s marketing slides claim an “up to 18 percent” improvement gen-on-gen for HEVC software encoding from i7-9700K to i7-10700K. When we reviewed published Cinebench R20 scores for Ryzen 7 3700X vs i7-9700K, we saw a 115 percent improvement there. We don’t think paying $44 more for the Intel part makes any kind of sense here, particularly when it isn’t a drop-in replacement.
Once again, we look forward to the chance to get real, concrete performance numbers once the hardware ships—but it’s difficult to expect that testing will bring good news when it happens.
Contact Information:
JIM SALTER
Tags:
, Wire, Artificial Intelligence Newswire, United States, English
Contact Information:
JIM SALTER