Intel this week introduced a slew of new processors, including its new 9th Generation Intel Core desktop processors, a new Core X-series aimed at premium content creation, and a 28-core Xeon for workstations, along with a new chipset.
It’s a very broad and powerful lineup, reflecting Intel’s goals of raising average selling prices by segmenting the market and acting as a counterpoint to AMD’s recent Ryzen desktop processors, including the 32-core Threadripper. But with increased segmentation comes increased complication, and it’s going to be harder for most people to determine which processor makes the most sense for their applications.
Consider today’s announcements. Anand Srivatsa, VP of Intel’s Client Computing Group, said Intel’s goal was to have leadership performance across all segments and product lines with no compromise. To do so, the company introduced a portfolio of desktop products geared to gamers and content creators, which included three different lines with different architectures, memory structures, and even varying manufacturing processes (both 14nm and 14++ nm designs; 10nm processors aren’t slated for another year or so).
Intel’s focus was on what it says are 130 million “creators.” Srivatsa said that 44 percent of professionals refresh their systems every 2 years, while 34 percent of prosumers and mainstream creators refresh their machines every 2-3 years—much faster than the rate of most desktop users.
First up was the top performer, the Xeon W-3175W workstation processor. This is a monster chip with 28 cores and 56 threads, designed to get a single core turbo frequency of 4.3 GHz out of the box, with overclocking possible (which is unusual for a Xeon chip). It offers 125 GB/second of memory bandwidth, with 6-channel DDR4 memory support plus up to 512GB of 2666MHz memory; ECC memory support; 38.5 MB of cache; and support for AVX-512 instructions. This is the latest in the Skylake-SP series, built on Intel’s 14nm process, and is slated to ship in December. Intel hasn’t yet provided pricing.
Intel positioned this processor as being appropriate for applications such as film creation, with Tangent Studio demonstrating how it used Intel Embree ray-tracing kernels to create its new Netflix movie NextGen. A studio rep explained how this lets directors make lighting and creative calls early, but added that it can still take up to 4 hours to render each frame of a movie, and that there are four different versions that have to be created for each frame, so the total rendering time is still enormous.
Intel didn’t formally introduce new motherboards, but there were several new Xeon boards on display.
Next up was the Core X-series, which is specifically aimed at high-end content creators, and includes models in the Core i7, Core i9, and Core i9 Extreme families, with a total of seven models from the i7-9800X with 8 cores and 16 threads, to the i9-9980XE with 18 cores and 35 threads. This series offers Intel Turbo Boost Max technology, which allows an individual core to run as fast as 4.5 GHz (the base speed varies from 3.8 GHz on the Core i7-9800X to 3.0 GHz for the i9-9980XE). These processors support 4 channels of DDR4-2666 memory, and the top of the line includes 24.75MB of cache.
These processors are known as Basin Falls, effectively a refresh of last year’s Skylake-X platform, with similar core counts and prices, but typically with higher frequency. (For instance, the i9-9980XE has a base frequency of 3.0 and a turbo of 4.5 GHz, while the current i9-7980XE has a base frequency of 2.5 GHz and a turbo of 4.4 GHz). They run on the existing X299 chipset, with a TDP of 165 Watts. All of the models support a total of 68 platform PCIe lanes, with 44 on the processor and 24 on the platform control hub (PCH). These run on the 14++ process, which seems to have relaxed some of the design rules (meaning it isn’t quite as dense as the early 14nm process) in order to enable higher speeds.
One new feature for this is solder thermal interface material (TIM), which sits between the CPU die and the heat spreader, and offers increased thermal conductivity; in other words, it distributes the heat farther and more evenly, which enables improved overclocking.
Demos here included photogrammetry, using a Smithsonian American Art Museum capture of lots of photos on a 10-core version (as the app won’t scale much beyond that).
Another demo showed the Unreal gaming engine running on an 18-core 9980XE using both the multi-threaded capability for rendering and higher single-threaded performance to show what it would be like to play the game.
Pricing ranges from $589 to $1,979 (quantity 1000). These will be shipping in November.
9th Generation Core
Finally, Intel announced its 9th Generation Core processor line, with Core i5, Core i7, and Core i9 variants, all designed for overclocking. The Core i5-9600K has six cores and six threads, with up to 4.6GHz maximum core frequency, 9MB of cache, and a list price of $262 (quantity 1000). The Core i7-9700K has eight cores and eight threads, with up to 4.9 GHz, 12 MB of cache, and a list prices of $374. The top-end Core i9-9900K offers eight cores and 16 threads, with up to 5.0GHz frequency and 16MB of cache, with a list price of $488. Srivatsa claimed that the i9-9900K is the first 5.0 GHz chip in broad volume, which he said was “breaking the laws of physics.” (It looks fast, but no, it isn’t that fast.)
Not only does the top end have more cores than last year (eight vs. six in the i7-8700K), but frequencies are higher, it has more cache (2 MB per core) and it supports AVX-512 instructions. The processors are designed for overclocking, and include Solder TIM.
The demonstrations here featured playing and streaming two games simultaneously in two virtual machines on one processor, and running games with faster speeds. Srivatsa said all the major game designers would support this line, including Acer Predator, Asus Republic of Gamers, Lenovo Legion, HP Omen, and Dell Alienware.
These chips are known as Coffee Lake-S, and are also manufactured on the 14++ process. They use a new chipset called the Z390, but can also work with the existing Z300 series boards. They are available for pre-orders today and should be available October 19.
What’s interesting to me here are the performance numbers. For the 9th Gen chips, Intel generally claimed 10 or 11 percent better gaming (frames per second) or multitasking, compared with last year’s chips (the 8700K), although they made a point of talking about how this was around 40 percent faster than a three-year-old PC (running a 6700K). But for some things, the performance is much better: Intel claimed 34 percent faster video editing on Adobe Premiere Pro, compared with last year’s chip, and 97 percent faster compared with a three-year-old system.
For the X-series, Intel engineers talked a lot about the chip’s mesh architecture, which connects all the cores with memory and IO controllers using rows and columns; the unified memory architecture is said to result in lower and more predictable latency. For comparison, Intel showed benchmarks of its 18-core Core i9-9980XE compared to AMD’s 32-core Threadripper Ryzen 2990WX. (As always, I take vendor benchmark numbers with a grain of salt, and you should too.)
With that said, this showed up to 27 percent better performance in Maya, which has many functions that are only lightly threaded (suggesting an 8-core version would actually be faster than the 18-core version); up to 108 percent better performance in Premiere Pro, because it really is memory sensitive; and up to 13 percent faster performance in Unreal, because this does use all of the cores. They stressed that there was no one number that covered it all.
And that’s where it all gets complicated to me.
For those of us who buy machines for enterprise and business use, it was pretty clear just a few years ago: Core i7 for the high-end, Core i5 in the middle, and Core i3 for the entry level, with a couple of the workstation parts designed for those running very specialized applications. (Before that, you could pick based on the clock speed). But now that processors aren’t advancing in speed as quickly, it’s grown more difficult.
Today, just about every mainstream app—Office, Acrobat, the web browser, and even basic video conferencing—runs perfectly fine on the whole Core line (and AMD’s Ryzen line, for that matter). But for higher-end applications, you need to know whether the application really requires more cores, more bandwidth, higher frequency cores, etc. And my guess is that most IT departments just don’t know.
But it’s even more complicated than that. Some functions in some applications use lots of cores well; others use only one or two threads. (One example: Excel is multi-threaded and scales well, but Visual Basic for Applications within it doesn’t.) To emphasize: it’s become much more difficult to pick the processor that’s right for your application. For an individual, that may be fine—people in a community can figure out what the best processor may be for their particular application. But for an enterprise buyer who supports a number of different applications, you’re almost certainly going to end up with a product that is a bit of a compromise. And that’s the price we pay for so much segmentation.