The LGA2066 platform and the Skylake-X family of processors were introduced by Intel more than a year and a half ago. Initially, this solution was aimed by the company at the HEDT segment, that is, at high-performance systems for users who create and process content, because Skylake-X contained a significantly larger number of computing cores compared to the usual representatives of the Kaby Lake and Coffee Lake families.
However, in the time that has passed since the advent of Skylake-X, the landscape in the processor market has changed significantly, and today quite affordable CPUs can have six or even eight computing cores, and promising mass CPUs, which should be released during this year, can get ten or even twelve cores. Does that make Skylake-X chips useless? Most likely no. Firstly, among the representatives of this series there are offers with 16 and 18 cores, and there will definitely not be mass options like them on the market in the near future. Secondly, the LGA2066 platform has other advantages that distinguish it from conventional consumer processors, such as superiority in the number of memory channels and available PCI Express lanes.
Therefore, the cosmetic update of the Skylake-X lineup, which the microprocessor giant carried out at the end of last year, seemed quite natural - it fit perfectly into the schedule of annual announcements adopted by Intel. However, the manufacturer's attitude to its HEDT-new products was a little surprising: the company not only did not revise prices, but also refused to provide samples of processors to the IT press, confining itself to a formal presentation and the subsequent start of sales.
Apparently, the company considered the new Skylake-X secondary and uninteresting products, but we fundamentally disagree with this formulation. Yes, the number of processing cores for representatives of this model range has not increased during the update process. However, other interesting improvements are embodied in them: the new items received increased clock frequencies, an increased L3 cache size, and an improved internal thermal interface. Therefore, we nevertheless decided to pay some attention to the updated Skylake-X, in which the Regard computer store helped us a lot, agreeing to provide a couple of new LGA2066 ten-core processors for research: Core i9-9820X and Core i9-9900X.
In addition, from the very moment of the announcement of Skylake-X Refresh, the question haunted us: why did Intel choose a name confusingly similar to the popular mass eight-core Core i9-9900K for the older ten-core HEDT processor? What does this mean? And now we have a chance to find out...
Skylake-X Refresh lineup
Intel announced the appearance of new LGA2066 processors with model numbers from the nine thousandth series back in October last year. The new products included seven models: six processors of the Core i9 series with a number of cores from 10 to 18 and an eight-core Core i7 model of a conditionally entry level. There are no more six-core and quad-core processors for LGA2066 in the new generation, which is not at all surprising against the backdrop of the rapid growth of the capabilities of the LGA1151v2 platform.
| Cores/threads | Base frequency, GHz | Turbo frequency, GHz | L3 cache, MB | Memory | TDP, W | Price | |
|---|---|---|---|---|---|---|---|
| Core i9-9980XE | 18/36 | 3,0 | 4,5 | 24,75 | DDR4-2666 | 165 | $ 1 979 |
| Core i9-9960X | 16/32 | 3,1 | 4,5 | 22,0 | DDR4-2666 | 165 | $ 1 684 |
| Core i9-9940X | 14/28 | 3,3 | 4,5 | 19,25 | DDR4-2666 | 165 | $ 1 387 |
| Core i9-9920X | 12/24 | 3,5 | 4,5 | 19,25 | DDR4-2666 | 165 | $ 1 189 |
| Core i9-9900X | 10/20 | 3,5 | 4,5 | 19,25 | DDR4-2666 | 165 | $989 |
| Core i9-9820X | 10/20 | 3,3 | 4,2 | 16,5 | DDR4-2666 | 165 | $889 |
| Core i7-9800X | 8/16 | 3,8 | 4,5 | 16,5 | DDR4-2666 | 165 | $589 |
The most noticeable change in the processors listed in the table, when compared with the previous Skylake-X models of the seven thousandth series, was the increase in clock frequencies. The nominal frequencies have risen by 200-600 MHz, and the maximum frequencies achieved when the turbo mode is turned on have increased by 200-300 MHz. In addition, the younger representatives of the series have increased the amount of cache memory in the third level. Previously, it was calculated based on the rule "1,375 MB per core", and now each core can account for up to about 2 MB of cache. Last but not least, the PCI Express controller of the octa-core Core i7-9800X has been completely unlocked, giving this processor access to all 44 lanes that were previously only available in processors with 10 or more cores.
However, all these pleasant changes led to an increase in heat release. While the first generation of Skylake-X had a thermal package limited to 140W, the TDP characteristic has been increased to 165W in the new processors. In other words, for the increased frequencies that are assigned to new processors without any fundamental changes in the 14-nm technological process used for their production, one has to pay with expanded energy and thermal limits.
True, Intel itself at the same time claims that the introduction of the production technology of the third version with the conditional name 14 ++ nm, which is now used to manufacture Coffee Lake and Coffee Lake Refresh processors, made it possible to raise the speed characteristics. And if not for this, then the heat dissipation could be even higher. But there is no reason to fear that the new Skylake-X may be prone to overheating. The improved thermal interface material under the heat distribution cover should reduce the operating temperatures of new processors. The place of the previously used polymer thermal paste was taken by solder with obviously higher thermal conductivity.
But everything mentioned above is just the tip of the iceberg. The fact is that the change in specifications, and primarily the increase in the volume of third-level cache memory, has a much more unexpected basis. Now, for the release of HEDT processors, Intel has begun to use somewhat different "sense" semiconductor crystals.
This means the following: HEDT processors have always been a desktop variety of server chips. Traditionally, Intel has taken lower Xeon modifications, adapted the memory controller and some other characteristics to them, and ported them to the desktop environment. At the same time, while Intel produced three variants of semiconductor crystals for its server products: LCC (Low Core Count) with 10 cores, HCC (High Core Count) with 18 cores and XCC (eXtreme Core Count) with 28 cores, in desktop HEDT processors only got the simplest versions of crystals. So, in the first generation Skylake-X processors with 6, 8 and 10 cores, an LCC crystal was used, and in modifications with 12, 14, 16 and 18 cores, an HCC crystal was used.
In the updated Skylake-X, which we are talking about today, the younger version of the LCC crystal is no longer used. All new 7-series HEDT processors, including 9800- and 9-core variants, are based on the HCC chip. That is, even in the Core i9900-18X or Core iXNUMX-XNUMXX, there are potentially XNUMX cores, but a significant part of them are hardware-locked at the production stage.
Such a decision, strange at first glance, was made just for the sake of increasing the amount of cache memory in new processors. The internal structure of Skylake-X assumes that a portion of the cache memory of 1,375 MB is allocated to each computing core. And if the same Core i9-9900X had used a junior LCC crystal, this processor would certainly not have been able to get more than 13,75 MB of L3 cache. The larger HCC die is more flexible in this regard, with a total of 24,75 MB of cache, and this increased amount is partially used in the eight- and ten-core processors of the new wave.
As a result, all Skylake-X became unified in design, but the downside of this unification was the widespread use of a very large semiconductor crystal with an area of about 485 mm2, which is more than two and a half times the area of \u2066b\u9bthe crystal of the eight-core Coffee Lake Refresh. This means that any of the LGA9900 processors of the nine thousandth series has a significantly higher cost compared to the same Core i9-9800K. But despite this, the eight-core Core i100-9X in the official price list is priced at only $9900 more than the Core i18-XNUMXK. Therefore, it is reasonable to assume that the production of eight- and ten-core processors based on XNUMX-core chips still makes some economic sense for Intel, for example, the company uses this opportunity to sell semiconductor chips with a large number of manufacturing defects, which until now could not find a worthy application.
Learn more about the ten-core Core i9-9900X and Core i9-9820X
For testing, we took two ten-core "new wave" processors - Core i9-9900X and Core i9-9820X. While these CPUs have moved to a new HCC die, they haven't changed that much compared to the Core i9-7900X. Usually, when Intel released second generations of processors for past versions of the HEDT platform, they transferred them to a newer microarchitecture, but this has not happened now. The changes affected only numerical parameters, but qualitatively, in the face of the Core i9-9900X and Core i9-9820X, we have almost the same thing that the ten-core Core i9-7900X of the 2017 model offered.
But on the other hand, the second generation Skylake-X has no compatibility problems: they work perfectly in existing LGA2066 motherboards based on the Intel X299 chipset. Like their predecessors, they have a four-channel DDR4 memory controller, and the PCI Express 3.0 controller built into them supports 44 lines, which in theory can be split into an arbitrary number of slots - from three to eleven.
However, the HCC semiconductor crystal underlying the Core i9-9900X and Core i9-9820X is somewhat different from the crystals used in the older Skylake-X before. Although the formal stepping retained the M0 number that was typical of the original Skylake-X versions with more than 12 cores, Intel has now begun to use modified lithographic masks in the manufacturing process due to the use of a more mature 14++ nm process instead of the previous 14+ nm process. The key difference between the technologies is a slightly larger step between the gates of the transistors, which, as we have already seen in the example of Coffee Lake, has a positive effect on the frequency potential.
At the microarchitecture level, there are no changes at all. Surprisingly, the new Skylake-X Refresh processors did not even include any hardware fixes to combat the Meltdown and Specter vulnerabilities. And this is very strange against the background of the fact that in the parallel product Coffee Lake Refresh, which was released at the same time, certain patches have already appeared. For example, modern LGA1151v2 processors are protected from Meltdown (Variant 3) and L1TF (Variant 5) attacks at the hardware level.
But the most annoying thing is not even that. The main reason for frustration is the absence of any changes in the scheme of combining processor components into a single whole. Skylake-X Refresh continues to use a peer-to-peer mesh mesh network overlaid on top of an array of processor cores. This inter-core interconnection works well with a significant increase in the number of cores in server processors, but for HEDT products with a not too high number of cores, it is much worse than a traditional ring bus, causing a dramatic increase in latency. One way to combat this negative effect would be to speed up Mesh connections, but here everything remains the same. The frequency of interconnects in both past and current Skylake-X is set at 2,4 GHz, so the L3 cache and memory controller in LGA2066 processors have noticeably worse latency against the background of massive Coffee Lake Refresh. True, this is partly offset by the extended second-level cache, which Skylake-X has a volume of 1 MB per core, and not four times less.
All this can be easily illustrated with a graph of the latency of the memory subsystem of Skylake-X processors of the current generation in comparison with Coffee Lake Refresh. It clearly shows the lack of improvements in the latency situation of the new HEDT processors.
But the new ten-core processors can boast of progress in clock speeds and cache size. For example, the Core i9-9900X has a 3 MB L19,25 victim cache, which is 40% larger than the previous ten-core processor, the Core i9-7900X. At the same time, the base frequency of the new model has increased from 3,3 to 3,5 GHz, but the maximum frequency of the Core i9-9900X in turbo mode can reach the same 4,5 GHz that was available to the ten-core of the previous generation. In both cases, reaching the 4,5 GHz mark requires the use of Turbo Boost Max 3.0 technology, with the traditional turbo mode, the maximum frequency for the Core i9-9900X is 4,4 GHz.
However, in practice, the situation with the frequencies of the Core i9-9900X is somewhat different. When loaded on all cores, the processor operates at a frequency of 4,1 GHz.
If this load uses AVX instructions, the processor frequency drops to 3,8 GHz.
And the most resource-intensive 512-bit instructions from the new AVX-512 set at full load on all cores make the processor slow down to 3,4 GHz, which, it should be noted, is even lower than the nominal frequency declared in the specifications.
If we talk about the ten-core Core i9-9820X, which is one step lower, then it differs from its older brother mainly in the amount of third-level cache, which is cut down to 16,5 MB. The rated frequencies are also a bit lower, but we must not forget that all Intel HEDT processors have free multipliers, which will allow enthusiasts to ignore this shortcoming.
Nevertheless, the nominal frequency of the Core i9-9820X is 3,3 GHz, and the maximum frequency in turbo mode is 4,1 or 4,2 GHz, depending on whether we are talking about Turbo Boost 2.0 or Turbo Boost Max 3.0 technology.
In practice, when operating the processor with default settings and the load on all cores, the Core i9-9820X is able to operate at a frequency of 4,0 GHz.
If the load uses AVX instructions, the processor reduces the operating frequency to 3,8 GHz.
And in AVX-512 mode, the frequency of the Core i9-9820X drops to nominal - 3,3 GHz.
Speaking about how the new ten-core LGA2066 processors are better than the old Core i9-7900X, one cannot help but recall the transition that has taken place in them to the use of a more efficient internal thermal interface. The heat-dissipating cover is now soldered to the die in a similar fashion to Coffee Lake Refresh. Intel says that due to this, new processors cool more efficiently and operate at lower temperatures, but there are two problems. First, the solder used by Intel is not very popular among overclockers because it is less efficient than liquid metal. And secondly, now the scalping procedure has turned out to be inaccessible for most enthusiasts: it has become very difficult to remove the cover without damaging the processor, so it is extremely difficult to improve the existing thermal interface.
At the end of the story about the characteristics of the Core i9-9900X and Core i9-9820X, it is worth mentioning the prices. Here, Intel did not show any imagination and set the cost of the older ten-core Core i9-9900X at the same $989 that it asked for the Core i9-7900X processor belonging to the previous generation. But the Core i9-9820X costs $100 less, which makes it a more attractive proposition for enthusiasts, because the 15% smaller L3 cache is unlikely to have any significant impact on performance, and the nominal clock speeds for true high-performance enthusiasts do not have any values.
Source: 3dnews.ru