There’s never been a better time to be a high-end system builder. Intel’s Skylake Server core made its way onto uber-desktops back in June as the Skylake-X family of chips, and AMD returned the serve with its Ryzen Threadripper CPUs and the X399 platform. Now it’s Intel’s turn to raise the stakes again.
On the bench today, we have the 16-core, 32-thread Core i9-7960X and the newest Extreme Edition CPU: the 18-core, 36-thread Core i9-7980XE. The Core i9-7960X is a core-for-core, thread-for-thread match against the Ryzen Threadripper 1950X, while the Core i9-7980XE lays claim to what is perhaps the highest core and thread count available in a “consumer” CPU today.
Of course, neither of these chips come cheap. The Core i9-7980XE lives up to its Extreme Edition lineage with an eye-popping $1999 sticker price, and the Core i9-7960X isn’t far behind at $1699. These price tags put the highest-core-count Skylake-X CPUs in a somewhat uncomfortable spot for a couple reasons. For one, that kind of money for a CPU is well within workstation-class territory, but neither of these CPUs do anything to address Threadrippers’ higher CPU PCIe lane complement or ECC RAM support. Recall that Threadripper CPUs and the X399 platform have both ECC RAM support and 60 PCIe 3.0 lanes directly connected to the CPU. Those resources are available from every Threadripper, too.
In an apparent response to AMD’s aggressive marketing of its platform advantages, Intel has begun aggregating the number of PCIe lanes available from both the chipset and CPU in its marketing materials. That aggregation is a bit disingenuous, though, because it doesn’t account for the fact that PCIe lanes from the X299 chipset have to traverse the DMI 3.0 link and its roughly 32 Gbps of bandwidth before reaching the host CPU. Some X399 peripheral controllers do need to travel over a similar link on Threadripper systems, but the jockeying for bandwidth from chipset to CPU should be a lot less rowdy there. Even if Threadripper CPUs don’t outperform the Skylake-X competition, the robustness of the X399 platform for workstation-class uses remains a point in its favor.
A block diagram of the Skylake Server core. Source: Intel
Although we’ve already discussed the Skylake Server architecture in detail in our review of the Core i9-7900X, the implementation of that architecture in the Core i9-7960X and i9-7980XE is worth exploring a bit more.
It’s no secret that Intel has long repurposed server hardware for its high-end desktop processors. The company has made multiple Xeon dies with varying core counts to fit the needs of the businesses it serves, but until now, the company has never had to repurpose its higher-core-count Xeons for duty on the desktop.
We weren’t briefed on the various Xeon Scalable Processor dies for the Skylake Server rollout this time around, but this Tom’s Hardware report leads us to conclude that the Core i9-7960X and Core i9-7980XE are bringing Intel’s high-core-count (or HCC) Skylake Server die down from the data center. Other Skylake-X Core i7s and Core i9s use the 10-core low-core-count, or LCC, die as their foundation. This may be the first time that Intel has ever had to bring its HCC Xeon die to its high-end desktop platform. Competition is a wonderful thing.
Like all Core i9 CPUs, these high-core-count chips boast two AVX-512 execution units: a dedicated AVX-512 unit per core on port five of the unified scheduler, and another created through the fusion of the two AVX-256 units on ports zero and one of the unified scheduler. Recall that the Core i7-7800X and Core i7-7820X are only equipped with one AVX-512 unit per core: the one created through fusion of the dual AVX-256 units. The dedicated AVX-512 unit on port five is disabled on those chips for market-segmentation reasons. Some users have reported that both AVX-512 execution paths are available even on Core i7 products, but we’re reporting the official line until we’ve had time to do some directed testing.
With the release of 12-, 14-, 16-, and 18-core CPUs, the Core i9 lineup is complete. You can see that the highest-end CPUs get a 25W TDP bump over the Core i9-7900X and friends, to 165W. Even with the more generous TDP, some have expressed concern about the clock speeds these higher-core-count Core i9s can hit under load. Happily, my experience with this duo suggests Intel’s 2.8 GHz base clock for the i9-7960X and 2.6 GHz base clock for the i9-7980XE are extremely pessimistic for enthusiast desktops with adequate cooling. Intel rates the Core i9-7980XE for 3.4 GHz non-AVX Turbo operation with all cores active, and I can confirm that the chip can hold that speed under a 280-mm liquid cooler. The i9-7960X is rated for an all-core Turbo speed of 3.6 GHz. Here’s the full Turbo Boost 2.0 table for each Skylake-X CPU, straight from the horse’s mouth:
Typical AVX workloads (albeit not AVX-512) caused the i9-7980XE to fall to just 3.2 GHz per core, but as our performance results will show, that drop hardly matters in the big picture. Regardless, I wouldn’t worry about seeing clock speeds under 3 GHz outside of intensive AVX-512 workloads. Given the paucity of programs using those code paths, the average enthusiast shouldn’t have any clock-speed worries at stock speeds.
Like most other Skylake-X CPUs, the i9-7960X and i9-7980XE offer an improved Turbo Boost Max 3.0 implementation compared to Broadwell-E. On these high-core-count CPUs, one should see up to 4.4 GHz speeds on two favored cores.
Intel’s rebalancing of the cache hierarchy on Skylake Server chips means the i9-7960X and i9-7980XE have massive private L2 caches at their disposal. Each core gets 1MB of L2 to work with, for a total of 16MB on the i9-7960X and 18MB on the i9-7980XE. Recall also that the bandwidth between the L1 and L2 caches on these chips has been increased to 128 bytes per cycle for reads and 64 bytes per cycle for writes. At the same time, the L3 cache per core now serves as a victim cache for the L2 above it, and L3 per core has been cut to 1.375 MB. Contrast that with the 2.5 MB of shared L3 per core on Broadwell Xeons. The new L3 allocation leads to 22MB of L3 across all cores on the i9-7960X and 24.75 MB on the i9-7980XE.
Now that we’ve revisited the essentials of Skylake-X, it’s time to get to testing.
Our testing methods
As always, we did our best to deliver clean benchmarking numbers. We ran each benchmark at least three times and took the median of those results. Our test systems were configured as follows:
|AMD Ryzen Threadripper 1950X||AMD Ryzen Threadripper 1920X|
|CPU cooler=||Thermaltake Water 3.0 Ultimate 360-mm liquid cooler|
|Motherboard||Gigabyte X399 Aorus Gaming 7|
|Memory type||G.Skill Trident Z DDR4-3600 (rated) SDRAM|
|Memory speed||3600 MT/s (actual)|
|Memory timings||16-16-16-36 2T|
|System drive||Intel 750 Series 400GB NVMe SSD|
|Intel Core i9-7980XE||Intel Core i9-7960X||Intel Core i9-7900X||Intel Core i7-7820X|
|CPU cooler||Corsair H115i 280-mm liquid cooler|
|Motherboard||Asus Prime X299-Deluxe|
|Memory type||G.Skill Trident Z DDR4-3600 (rated) SDRAM|
|Memory speed||3600 MT/s (actual)|
|Memory timings||16-16-16-36 2T (DDR4-3600)|
|System drive||Samsung 850 Pro 512GB|
|Intel Core i7-6950X|
|CPU cooler||Cooler Master MasterLiquid Pro 280 280-mm liquid cooler|
|Motherboard||Gigabyte GA-X99-Designare EX|
|Memory type||G.Skill Trident Z DDR4-3200 (rated) SDRAM|
|Memory speed||3200 MT/s (actual)|
|Memory timings||16-18-18-38 2T|
|System drive||Samsung 960 EVO 500GB|
They all shared the following common elements:
|Storage||2x Corsair Neutron XT 480GB SSD
1x HyperX 480GB SSD
|Discrete graphics||Nvidia GeForce GTX 1080 Ti Founders Edition|
|Graphics driver version||GeForce 385.69|
|OS||Windows 10 Pro with Creators Update|
|Power supply||Seasonic Prime Platinum 1000W|
Our thanks to Intel and AMD for all of the CPUs we used in our testing. Our thanks to AMD, Intel, Gigabyte, Corsair, Cooler Master, and G.Skill for helping us to outfit our test rigs with some of the finest hardware available, as well.
Our X299 test rig
Some additional notes on our testing methods:
- Unless otherwise noted, we ran our gaming tests at 2560×1440 at a refresh rate of 144 Hz. V-sync was disabled in the driver control panel.
- For our Intel test system, we used the Balanced power plan, as we have for many years. Our AMD test bed was configured to use the Ryzen Balanced power plan that ships with AMD’s chipset drivers.
- All motherboards were tested using the most recent firmware available from the board vendor, including pre-release versions provided exclusively to the press where necessary.
- All available Windows updates were installed on each test system before testing commenced. The most recent version of each software application available from each vendor was used in our testing, as well.
Our testing methods are generally publicly available and reproducible. If you have questions, feel free to post a comment on this article or join us in the forums.
Memory subsystem performance
Let’s kick off our testing with a quick look at main memory performance. Both the Ryzen Threadripper CPUs and every Core i9 in this test are running with the same DDR4-3600 kit, so we can easily make apples-to-apples comparisons about their performance using AIDA64’s built-in memory benchmarks.
Intel’s memory controller on the high-core-count Skylake-X die seems to favor read performance over writes. Given the potential hunger for data of these chips’ AVX-512 units, that’s probably the right balance to strike. Copy bandwidth is slightly higher than on the low-core-count chips, but only slightly. The Ryzen Threadripper duo beats out the Skylake-X CPUs in memory writes and copies, but read bandwidth lags behind Skylake-X chips of equal or higher core counts.
Even though they have many more cores and threads—and thus a broader mesh to traverse—than low-core-count Skylake-X chips, the i9-7960X and i9-7980XE deliver memory access latencies in line with their less resource-endowed cousins. That’s a testament to the scalable nature of the Skylake Server mesh architecture.
Some quick synthetic math tests
To get a quick sense of how these chips stack up, we turn to AIDA64’s synthetic benchmarks. Photoworxx stresses the integer SIMD units of these chips with AVX. FPU Julia tests single-precision floating-point throughput and uses AVX instructions (though not AVX-512), while FPU Mandel puts those same instructions to work in the service of double-precision throughput.
Hm. It seems Photoworxx may not be fully optimized for CPUs with this many cores. Perhaps the benchmark will be optimized for these chips in the future, at which point we’ll have to retest.
As we’d expect, throwing more cores at the AIDA64 Hash benchmark produces basically linear increases in bandwidth for our Skylake-X chips.
Ryzen Threadripper chips do outpace the Intel competition in this benchmark, but that’s because the Zen architecture has what seems to be little-publicized support for Intel’s SHA Extensions. These extensions permit hardware acceleration of some of the SHA family of algorithms, and CPU Hash uses SHA-1 as its algorithm of choice. SHA-1 isn’t particularly useful in practice any longer, but SHA-256 is, and the folks at SiSoft report similar speedups for that algorithm. AVX implementations of other SHA versions might help Intel processors close the gap, though.
Thanks to their large complements of wider AVX units compared to Threadripper CPUs, the i9-7960X and i9-7980XE come close to doubling the throughput of the 1950X in the Julia test, and they still maintain a healthy lead in the Mandel test. That’s excellent performance. Now, let’s see how these chips handle games.
Hitman‘s DirectX 12 renderer can stress every part of a system, so we cranked the game’s graphics settings at 1920×1080 and got to testing.
Hitman‘s DX12 mode favors lots of cores, but performance seems to regress once we move beyond 10 cores or so. Neither the i9-7960X nor the i9-7980XE can top the i7-7820X, and the AMD CPUs fall behind their many-core Intel counterparts. Despite their high average frame rates, however, the i9-7960X and i9-7980XE fall toward the back of the pack in our 99th-percentile frame time measure of delivered smoothness. Only the i7-7820X does worse, if a 99th-percentile frame time of 16.2 ms can even be called bad to begin with.
That 99th-percentile frame time weirdness for the i7-7820X is a theme of this review. You will want to pay attention to the chip’s frame-time plots in the graphs above. For reasons we still haven’t been able to crack, this chip exhibits its own particular brand of stutter that’s clearly evident in frame-time graphs. We can analyze just how much of our one-minute test run those little spikes occupy using our advanced “time-spent-beyond” metrics, so let’s get to it.
These “time spent beyond X” graphs are meant to show “badness,” those instances where animation may be less than fluid—or at least less than perfect. The formulas behind these graphs add up the amount of time our graphics card spends beyond certain frame-time thresholds, each with an important implication for gaming smoothness. The 50-ms threshold is the most notable one, since it corresponds to a 20-FPS average. We figure if you’re not rendering any faster than 20 FPS, even for a moment, then the user is likely to perceive a slowdown. 33 ms correlates to 30 FPS, or a 30-Hz refresh rate. Go lower than that with vsync on, and you’re into the bad voodoo of quantization slowdowns. 16.7 ms correlates to 60 FPS, that golden mark that we’d like to achieve (or surpass) for each and every frame. 8.3 ms corresponds to 120 FPS, an even more demanding standard that fans of high-refresh-rate monitors will want to pay close attention to. Finally, we’ve recently begun including an even more demanding 6.94-ms mark that corresponds to the 144-Hz maximum rate typical of today’s high-refresh-rate gaming displays.
Even though its frame-time plot is spikier than we’d like, the i7-7820X’s weirdness shows up as less than a tenth of a second spent beyond 16.7 ms. Even at the more demanding 8.3-ms mark, the i7-7820X’s spikes add up to just about a second and a half of our one-minute test run. Both the Core i9-7960X and i7-7980XE do worse, and the Ryzen 7 1800X is far behind.
Even though we’d prefer not to see the i7-7820X’s spikiness at all, our time-spent-beyond-X graphs suggest those spikes only represent a tiny proportion of frames rendered. I certainly didn’t notice hitchiness or other unpleasantness while gaming on the chip in Hitman. Let’s see whether that continues to be the case.
Deus Ex: Mankind Divided (DX11)
With its rich and geometrically complex environments, Deus Ex: Mankind Divided can prove a challenge for any CPU at high enough refresh rates. We applied our preferred recipe of in-game settings to put the squeeze on the CPU and got to it.
Deus Ex gives all of our Intel CPUs a chance to shine versus the red team’s competitors. Neither the i9-7960X or the i9-7980XE take the top spot, but they still deliver higher performance than AMD’s chips do. The Core i7-7820X’s spikiness continues to be bad enough to drop its 99th-percentile frame time performance back with the Ryzens, even though its average-FPS figure suggests a fine experience.
None of our contenders leave the graphics card waiting for work long enough to cause substantial time spent beyond 16.7 ms. Even at the 8.3-ms threshold, the i9-7960X, i9-7980XE, and the i7-7820X all hold up the graphics card for about half as long as the Ryzen chips do. That fine performance continues at the 6.94-ms mark. Even with its unusual spikiness, the i7-7820X is hardly delivering a bad time.
Although Crysis 3 is nearly four years old now, its lavishly detailed environments and demanding physics engine can still stress every part of a system. To put each of our CPUs to the test, we took a one-minute run through the grassy area at the beginning of the “Welcome to the Jungle” level with settings cranked at 1920×1080.
We generally expect Crysis 3 performance to scale with the number of cores available, but the game doesn’t actually run better as the core counts climb past 10. The i9-7960X and i9-7980XE land midpack in our average-FPS figures, and they turn in the “worst” 99th-percentile frame times of this bunch. When you’re still delivering 99% of frames at an instantaneous rate of about 73 FPS, it’s hard to call the overall experience bad with any of these chips.
None of our chips spend any time past the troublesome 50-ms or 33.3-ms marks, thankfully. We’d expect the spikiness of the Core i7-7820X to begin showing up at the 16.7-ms mark, and it kind of does. Thing is, all those spikes only add up to four milliseconds of a one-minute test run spent past 16.7 ms. Moving down to the 8.3-ms mark, the Core i7-7820X is spending less than a second of our one-minute test run holding up the powerful GeForce GTX 1080 Ti, while the higher-core-count Core i9s log about three to four seconds of tough frames at that mark. I can’t say I notice these spikes as anything more than fleeting unsmoothness in gameplay, and while we’d obviously like a flatter line in our frame-time plots, it’s hard to argue that the i7-7820X is delivering a subpar gaming experience.
Even at the tough 6.94-ms threshold, the i7-7820X spends a perfectly respectable three seconds of our test run holding up the graphics card. More troublesome is the fact that the i9-7960X spends about six seconds of our test run below 144 FPS, while the i9-7980XE spends about eight seconds in a similar predicament. Whatever issue is causing that hold-up, it’s not exhibited by the lower-core-count Intel chips or Ryzen Threadrippers.
Watch Dogs 2
Watch Dogs 2 can seemingly occupy every thread one can throw at it, so it’s a perfect CPU test. We turned up the eye candy and walked through the forested paths around the game’s Coit Tower landmark to get our chips sweating.
Here’s a case wehere Intel’s latest and greatest prevail. All of our Skylake-X chips hang just a couple of frames per second apart on average, and they all cluster toward the top of the 99th-percentile frame time chart—all of them, of course, save the i7-7820X.
Once again, only a few frames show up past the 16.7-ms mark for any of these chips, so we have to consider stricter stuff. At the 8.3-ms mark, the Skylake-X chips—even the Core i7-7820X—all spend several seconds less than the Ryzen competition on tough frames that hold up the graphics card. Once again, the i7-7820X’s spikiness looks bad on a graph, but its actual impact on gameplay simply isn’t a big deal.
Grand Theft Auto V
Grand Theft Auto V can still put the hurt on CPUs as well as graphics cards, so we ran through our usual test run with the game’s settings turned all the way up at 1920×1080. Unlike most of the games we’ve tested so far, GTA V favors a single thread or two heavily, and there’s no way around it with Vulkan or DirectX 12.
GTA V doesn’t much like running on many-core chips, and the complex fabrics and meshes of this server-class hardware doesn’t seem to help. Past 10 cores and 20 threads, performance begins to regress in our average-FPS charts on both AMD and Intel CPUs, and the i9-7960X and i9-7980XE fall toward the back of the pack in both average FPS and in our 99th-percentile frame time metrics. As usual, the i7-7820X is off in the weeds a bit, too.
As we’d expect from those 99th-percentile frame time figures, the i9-7960X spends about seven seconds of our one-minute test run past 8.3 ms on tough frames, and the i9-7980XE fares even worse. As usual, the i7-7820X’s frame-delivery weirdness just doesn’t add up to that much time past 8.3 ms. At 6.94 ms, the i7-7900X and i7-6950X leave everything else in the dust.
Streaming performance with Hitman and OBS
One of the uses that Intel and AMD have hyped the most for their highest-end desktop processors this year is single-PC gaming and streaming. The most avid Twitch streamers, as we understand it, have tended to set up dedicated PCs for video ingestion and processing to avoid affecting game performance, but the advent of these many-core CPUs may have opened up a world where it might be more convenient to run one’s stream off a single PC.
Although one might wonder why people are still making a hullabaloo about CPU encoding performance when hardware-accelerated game streaming is available from both major GPU software packages, the fact of the matter seems to be that the most demanding professionals still choose to use software encoding. The reason for this is that Twitch and other streaming services have a restrictive bit rate for streamed content. GPU-accelerated services like GeForce Share (nee Shadowplay) and Radeon ReLive make it easy to stream without affecting gaming performance that much, but they might not offer the highest-quality viewing experience to fans within a given bit rate. For achieving the best results possible, the name of the game is still software encoding with x264.
Given the exploding popularity of Twitch and similar services, this review was as good a time as any for us to start digging into single-PC gaming and streaming performance. I’m obviously not a professional streamer by any stretch of the imagination, but I did learn enough to be dangerous with the popular OBS Studio tool for this review. I only offer that caveat because a seasoned professional in this space might use different settings and software to achieve their preferred results. I’m not running a webcam or overlay, for example, and those add-ons could further affect CPU performance.
I played with OBS’ various x264 settings to achieve what looked (to my eye) like the best visual quality possible without unduly bogging down our less-powerful test rigs. To my snobbish eyes, that was the “fast” x264 profile. For perspective, my TR colleague Zak Killian says he has to use the “veryfast” or “ultrafast” presets to make gaming and streaming possible at once on his Core i7-4790K system. I otherwise configured OBS using Twitch’s recommended guidelines.
It’s worth noting that we’re not considering x264 encoder performance in isolation for this review. While metrics like dropped frames are certainly important to the viewer experience, we don’t have the methods to effectively process or present that data yet. We did monitor stream quality during our testing and ensured that our particular encoder settings weren’t producing choppy or otherwise ugly stream delivery, but we didn’t dive deep into OBS log files or anything to that effect. We might consider these metrics in future articles, but for now, we’re worried about the gameplay experience these CPUs deliver to the streamer.
Throw a single “fast” x264 stream onto these chips, and only the i9-7960X and i9-7980XE remain unaffected by the moderate performance hit that affects the other high-end desktop chips in this lineup. In fact, their 99th-percentile frame times even improve slightly compared to their gaming-only performance. We suspect the extra load keeps the game’s threads from being scattered all over these high-core-count CPUs.
The Ryzen Threadripper CPUs aren’t delivering the highest peak frame rates under this load, but their 99th-percentile frame times are right in line with those of the Skylake-X chips. The Core i7-7820X handily outperforms the Ryzen 7 1800X in our average-FPS metric, but its 99th-percentile frame time is right down there with that of the 1800X. Let’s see just how much time these chips spend at the ragged edge with our time-spent-beyond-X graphs.
Once again, it’s most informative to click over to the 8.3-ms threshold straight away for our streaming tests. Even though it shares a 99th-percentile frame time with the Ryzen 7 1800X, the i7-7820X spends less than a third of the time past 8.3 ms that the 1800X does, at about seven and a half seconds compared to 24. That figure pales next to the rest of the Skylake-X family’s great performance, of course, but it’s still better than even the Ryzen Threadripper 1920X and 1950X can manage. As for the Core i9 CPUs, they’re simply in a different class of performance than the Threadrippers under this streaming workload. The i9-7900X, i9-7960X, and i9-7980XE are barely fazed by these test settings (and neither is the i7-6950X, for what it’s worth).
At least in this simple testing, the Core i9 family seems to be the best thing going for fluid frame rates from a streaming PC. Our testing suggests single-stream game performance with Intel CPUs doesn’t really increase past 10 cores, however. Given that fact, you could probably load down the i9-7960X and i9-7980XE even further without affecting game performance much, perhaps with even higher-quality x264 settings, multiple streams, or with on-the-fly encoding to archival video. Although it might be cheaper to build separate, dedicated PCs for gaming and streaming, there is an undenible convenience to having everything on one box for those who can afford it.
With that, we conclude our gaming benchmarks. As expected, 1920×1080 gaming is something these many-core chips can do in a pinch, but it’s not their forte in the least. Much cheaper CPUs can deliver far better gaming-only performance at this resolution. Game streamers should find a lot to love from any Core i9 (or Ryzen Threadripper) CPU, though we’ll need to load these high-end chips down even further to see where they break in future testing.
Our tests also showed that while the Core i7-7820X does have something of a frame-time consistency problem, its effect on delivered performance is actually pretty minor. If Intel can get to the bottom of the i7-7820X’s stutter in games, it’ll have the best value in high-end desktop processors today. For now, the chip remains on the knife-edge of greatness.
Now that we’ve seen whether these CPUs have game, it’s time to put away the toys and slip into something more business-casual.
Most Core i9 and Threadripper owners will not be browsing Facebook all day, but these three benchmarks offer an idea of how these chips will perform in lightly-threaded web browsing tasks. Unfortunately for the high-core-count Core i9s, the picture is a mixed one. The i9-7960X and i9-7980XE trail especially far behind the pack in the JetStream benchmark, end up about where we’d expect in Octane, and bookend our Kraken results. It seems some of the latency-sensitive tests that make up these test suites don’t play well with the high-core-count Skylake-X die.
Compiling code with GCC
Our resident code monkey, Bruno Ferreira, helped us put together this code-compiling test. Qtbench records the time needed to compile the Qt SDK using the GCC compiler. The number of jobs dispatched by the Qtbench script is configurable, and we set the number of threads to match the hardware thread count for each CPU.
The Core i9 CPUs come out on top in this test, but we’re clearly high on the curve of diminishing returns with Qtbench and 36 threads.
File compression with 7-zip
Our high-core-count Core i9s earn a massive lead in the compression portion of this test, but the i9-7960X, Threadripper 1950X, and i9-7980XE finish in a dead heat in the decompression portion of this benchmark.
Disk encryption with Veracrypt
In the accelerated portion of our Veracrypt test, both the i9-7960X and i9-7980XE lead the pack by a wide margin. The unaccelerated Twofish algorithm lets Ryzen shine, though.
The evergreen Cinebench benchmark is powered by Maxon’s Cinema 4D rendering engine. It’s multithreaded and comes with a 64-bit executable. The test runs with a single thread and then with as many threads as possible.
On a single thread, Skylake-X chips take a moderate lead over the Ryzen competition, as we’d expect. Turbo Boost Max 3.0 probably helps, but so does the Skylake architecture’s higher IPC to begin with.
Loose every core of these chips on Cinebench, and the i9-7960X outpaces the Threadripper 1950X thread-for-thread. The i9-7980XE is even further ahead, as it ought to be.
Blender is a widely-used, open-source 3D modeling and rendering application. The app can take advantage of AVX2 instructions on compatible CPUs. We chose the “bmw27” test file from Blender’s selection of benchmark scenes to put our CPUs through their paces.
Blender loves it some threads, but neither of the high-core-count Core i9s open much of a lead over the Threadripper 1950X in this test. It seems there’s a wall of diminishing returns around 16 cores in this benchmark.
Here’s a new benchmark for our test suite. Corona, as its developers put it, is a “high-performance (un)biased photorealistic renderer, available for Autodesk 3ds Max and as a standalone CLI application, and in development for Maxon Cinema 4D.”
The company has made a standalone benchmark with its rendering engine inside, so it was a no-brainer to give it a spin on these CPUs. The benchmark reports results in millions of rays cast per second, and we’ve converted that figure to megarays for readability.
If Blender doesn’t see much benefit from the move to these many-core chips, Corona sure does. Both the i9-7960X and i9-7980XE jump far ahead of the Threadripper 1950X here.
Handbrake is a popular video-transcoding app that recently hit version 1.0. To see how it performs on these chips, we’re switching things up from some of our past reviews. Here, we converted a roughly two-minute 4K source file from an iPhone 6S into a 1920×1080, 30 FPS MKV using the HEVC algorithm implemented in the x265 open-source encoder. We otherwise left the preset at its default settings.
The i9-7960X and i9-7980XE shave 40 seconds off the Threadripper 1950X’s time, but the x265 encoder doesn’t seem to gain anything from having 18 cores to play with.
CFD with STARS Euler3D
Euler3D tackles the difficult problem of simulating fluid dynamics. It tends to be very memory-bandwidth intensive. You can read more about it right here. We configured Euler3D to use every thread available from each of our CPUs.
It should be noted that the publicly-available Euler3D benchmark is compiled using Intel’s Fortran tools, a decision that its originators discuss in depth on the project page. Code produced this way may not perform at its best on Ryzen CPUs as a result, but this binary is apparently representative of the software that would be available in the field. A more neutral compiler might make for a better benchmark, but it may also not be representative of real-world results with real-world software, and we are generally concerned with real-world performance. With all that in mind, it should be no surprise that the Core i9-7960X and i9-7980XE are the undisputed champions of Euler3D, and by no small margin.
Digital audio workstation performance
One of the neatest additions to our test suite of late is the duo of DAWBench project files: DSP 2017 and VI 2017. The DSP benchmark tests the raw number of VST plugins a system can handle, while the complex VI project simulates a virtual instrument and sampling workload.
We used the latest version of the Reaper DAW for Windows as the platform for our tests. To simulate a demanding workload, we tested each CPU with a 24-bit depth and 96-KHz sampling rate, and at two ASIO buffer depths: a punishing 64 and a slightly-less-punishing 128. In response to popular demand, we’re also testing the same buffer depths at a sampling rate of 48 KHz. We added VSTs or notes of polyphony to each session until we started hearing popping or other audio artifacts. We used Focusrite’s Scarlett 2i2 audio interface and the latest version of the company’s own ASIO driver for monitoring purposes.
A very special thanks is in order here for Native Instruments, who kindly provided us with the Kontakt licenses necessary to run the DAWBench VI project file. We greatly appreciate NI’s support—this benchmark would not have been possible without the help of the folks there. Be sure to check out their many fine digital audio products.
At 96 KHz and a buffer depth of 64, we are most likely testing per-core throughput more than anything. By that measure, it’s perhaps not surprising that chips with the most L2 cache and the highest clocks perform the best. Relax the buffer depth to 128, and performance improves on all these chips. For the i9-7960X and i9-7980XE, though, the performance increase is simply jaw-dropping. These are by far the best-performing chips in DAWBench VI at a high sampling rate.
The DAWBench DSP test is a much more even playing field for our test subjects at 96 KHz. The Threadripper 1950X can keep up with the i9-7980XE here. In fact, the i9-7960X emerges as our champion at both buffer depths.
Even if we halve our sampling rate, the blue team retains a wide lead. The Core i9-7960X basically doubles the Threadripper 1950X’s performance at a buffer depth of 64, a figure that maps nicely onto the vast gulf in floating-point throughput between Zen and Skylake-X we saw early in this piece. The only thing keeping the gulf from being wider at 128 samples is that we maxed out the number of voices of polyphony available in DAWBench VI. Kind of astounding, really.
Once again, DAWBench DSP evens out the playing field for these chips. The Threadripper 1950X has no trouble keeping up with the highest-end Core i9s here.
Our expanded DAWBench testing continues to suggest that Ryzen Threadripper CPUs are great chips for work that involves truckloads of DSPs, but virtual instrument performance is still far and away the domain of Intel chips. If you need competence in both domains, Skylake-X CPUs are the way to go.
Power consumption and efficiency
We can get a rough idea of how efficient these chips are by monitoring system power draw in Blender. Our observations have shown that Blender consumes about the same amount of wattage at every stage of the bmw27 benchmark, so it’s an ideal guinea pig for this kind of calculation. First, let’s revisit the amount of time it takes for each of these chips to render our Blender “bmw27” test scene:
Next, we check system power draw at the wall using our trusty Watts Up power meter:
Strangely, load power consumption for the Ryzen Threadripper duo is up compared to our initial review of those chips. We’re using a new version of Blender, updated firmware on all of our Ryzen boards, and a higher memory speed, so those changes might explain the higher power consumption. These results were repeatable, so we can probably chalk them up to the changes in our test environment since our initial review. In any case, the high-core-count Skylake-X chips consume less power under load compared to the Threadrippers, and that augurs well for their efficiency in our final reckoning.
To estimate the task energy consumed in joules for our rendering workload, we simply multiply the time each chip needs to complete the scene by its load power. We can then plot that task energy figure against time-to-completion to get this intuitive view . The best results in our efficiency scatter tend toward the lower left of the chart, where power consumption is lowest and time to completion is shortest.
At least with our particular test rigs, the Core i9-7960X and i9-7980XE consume moderately less power than the Ryzen Threadripper 1950X while delivering slightly better performance. At least at stock speeds, the common hand-wringing about power consumption and efficiency with Skylake-X seems a bit overblown to us. Overclocking these chips is a different story, of course, but not one we’ll be exploring today.
It’s time once again to condense all of our test results into our famous value scatter plots. We use a geometric mean of all of our real-world results to ensure that no one test has an undue impact on the overall index. First up, let’s look at gaming performance.
The short take here: don’t drop $2000 on a CPU for 1920×1080 gaming. The Core i9-7960X and i9-7980XE are not going to deliver world-beating performance for high-refresh-rate gaming at low resolutions, and you can build a system that’s far better at that purpose for far less money. If you have a GTX 1080 Ti like the one we used in our test rig, you should really be hooking it up to a 2560×1440 or 3840×2160 monitor to begin with.
Taking these results for what they’re worth, however, the Core i9-7960X and i9-7980XE don’t deliver an appreciably better gaming experience than the Ryzen Threadripper 1920X and Threadripper 1950X. The Core i9s do shine in game streaming and multitasking, however, and we’d highly recommend you be planning to do more than gaming alone before you drop this kind of coin on a processor.
Shocker: the Core i9-7960X and Core i9-7980XE are easily the two fastest CPUs we’ve ever tested for productivity workloads. Perhaps because it has to trade a bit of all-core clock speed for core count, though, the i9-7980XE does little to separate itself from its 16-core sibling in most of our tests. There are a few cases in our testing where the extra cores help the i9-7980XE distinguish itself from the i9-7960X a bit, but not many. We gotta admit: part of the problem is finding enough things to do with 18 cores on the desktop to begin with.
Even with that caveat in mind, the i9-7980XE doesn’t seem worth the whopping $300 extra over the i9-7960X unless you absolutely must have every last drop of performance possible from an X299 system. That’s nothing new for high-end Intel CPUs, though. We’ve long suggested that taking one step back from the top will get you most of the performance of the top-of-the-line chip for a lot less money, and that’s as true of the i9-7960X as it’s ever been.
Another mark against the professional-grade performance of the i9-7960X and i9-7980XE is that the X299 platform has both feet firmly planted on the consumer side of the fence. X299 motherboards don’t support ECC RAM, and they have a hard limit of 128GB of memory. Intel will sell you ECC support and a higher RAM ceiling on one of its Xeon-W CPUs, but the price tags of those chips are truly eye-watering when compared to their X299 counterparts. AMD’s X399 platform offers both ECC support and a higher theoretical RAM capacity than X299, no questions asked. I have a hunch that we haven’t seen the upper limit of Threadripper CPUs’ core counts, either. Folks whose workloads aren’t bolstered by what Core i9 CPUs have to offer will still find Threadrippers a great value.
Even for those who can benefit from the virtues of Skylake-X, it’s kind of crazy how much dough one has to spend to get a chip that consistently beats the Ryzen Threadripper 1950X. The Core i9-7960X is 70% more expensive than the Threadripper, and the i9-7980XE is twice as expensive. If you look at our final performance index alone, the i9-7960X is about 24% faster than the Threadripper 1950X overall. Given how little its two extra cores seem to add to the equation, the i9-7980XE will never be a good value.
Let’s be real, though: once you’re shopping in the realm of $1000-and-up CPUs, a general sense of value isn’t a major concern. Return on investment seems like a stronger case. If your PC needs to run software as fast as possible and perhaps make you money while doing it, Core i9 systems could let you do more in less time or simply let you do more, period. Our value index necessarily has to ignore the fact that buyers in this range are likely looking at heavy-duty chips for workload-specific reasons. Some parts of our benchmark suite run twice as fast or more than twice as fast on Core i9s compared to the 1950X, and if those cases match yours, the extra cost of a Core i9 might be easier to stomach.
While we’re at it, I have to repeat my continued bittersweet feelings about the Core i7-7820X here. This chip would easily be the best bang-for-the-buck CPU in high-end desktops right now, but for the fact that it continues to exhibit some minor frame-delivery weirdness that other Skylake-X chips don’t. To be fair, you likely won’t notice the vanishing bits of unevenness the i7-7820X introduces to gameplay, but even a hint of unsmoothness shouldn’t be an issue on a CPU this expensive.
If you’re willing to tolerate that possibility, however, the i7-7820X is possibly the smartest way onto the X299 platform. For about $180 more than a Ryzen 7 1800X right now, you get higher performance across the board, more PCIe lanes, and quad-channel memory support. It’s hard to argue with those improvements.
Even with my concerns about pricing and platform parsimony, there is no denying that the Core i9-7960X and i9-7980XE are both awesome CPUs, as well. In both performance and power efficiency, the i9-7960X and i9-7980XE are best-in-class. On top of these chips’ great performance, it’s just flat-out cool that Intel is shipping the high-core-count version of its Xeon silicon on a consumer platform for the first time. (Thanks, AMD.) These are truly cutting-edge processors, and if you have the scratch for one, you won’t be disappointed.
Like this review? Help us continue to perform the dozens of hours of hands-on testing that go into articles like this one by becoming a TR subscriber today. Subscribers get exclusive benefits as part of their membership, and you can contribute any amount you’d like to support us. Best of all, we were doing crowdfunding before it was cool, so your entire contribution (less transaction fees) helps support our continuing work. Thanks in advance for your support.