Intel Skylake-X (i9 7900X) and Kaby Lake-X (i7 7740X) CPU Review

It has been a long time coming since there was true competition in the CPU market. Intel, for the last several years and CPU generations (since 2006/Conroe-based chips and later), have been pretty dominant over their competitor, AMD. Instructions Per Clock (IPC) has always been heads above the rest. AMD’s lack of comparable IPC seems to have moved them in a direction of adding more cores and threads to compete, and at a lower price. Bang for your buck was second to none from Thuban all the way to Vishera based AMD CPUs. AMD finally stepped up to the plate and, for all intents and purposes, knocked one out of the park with their Zen architecture. Not only did they manage to catch up in IPC just a few percent behind in most single threaded tasks, but they poured on the cores and kept the price low. A wonderful thing for consumers and competition.

Cue a response from the boys in blue, Intel. While we have seen grumblings in our, and many other forums about Intel in a panic mode, I personally have to disagree. I am certain (guessing) Intel moved up some timetables, to get a more competitive product out, but I don’t picture the sleeping giant in a panic quite yet. To throw an analogy out there (no, not a car analogy!), it’s like putting in a pinch hitter to go against the lefty they put in from the bullpen. In other words, they are putting in the best man they have available for the job. Sure he may not have had time to stretch and warm up, but you can bet he was already on the team prior… he was moved into action a bit sooner than they may have wanted.

Anyway, enough about the state of the union, and on to Intel’s counter punch: Skylake-X and Kaby Lake-X. What’s that you say? Kaby Lake-X? Yep, that’s right, starting off on the low end, Intel is, curiously, bridging the gap between their mainstream and HEDT platforms by putting a quad-core with Hyperthreading (4c/8t) at the bottom of the product stack named, 7740X. In other words, a 7700K but based on the socket 2066 platform. There are more differences and will jump into those a bit later. Currently, at the top (there are other CPUs coming out with more threads, note), is the i9-7900X. The 10c/20t 7900X will now fall in the middle of the Skylake-X product stack with 12c/24t (i9-7920X), 14c/28t (i9-7940X), 16c/32t (i9-7960X), and finally the flagship 18c/36t part (i9-7980XE).

We have the 7900X and 7740X in our hands. We will compare it to the 6950X and the 7700K for both stock improvements and IPC. We will put it through our CPU testing suite as well as a few games to see how these CPUs perform! Please note, I was a bit limited on time with the CPUs arriving on the 17th.

Specifications and Features

Below is a list of specifications of our two new CPUs. On the left is the 7900X, while on the right is the 7740X. Both processors are based off the 14nm Tri-Gate 3D production, using a 64-bit instruction set. Both CPUs also support DDR4 2666 as the standard and up to 128 GB of RAM. The 7740X will be dual channel however and not able to reap the benefits of quad channel memory.

The 7900X comes with 10 cores and 20 threads with a base clockspeed of 3.3 GHz. This is a 300 MHz bump up over the i7-6950X. The max turbo frequency is going to be a stellar 4.5 GHz. With an improved Turbo Boost 3.0, this time around the max boost is done on TWO cores instead of one. I don’t have any information on the regular boost at this time. But I would imagine it to be cascading to less cores as the clocks go up just as was done in the past. The TDP remains the same at 140 W. The 7900K sports 44 on-CPU PCIe 3.0 lanes, with its little brothers getting less (28 lanes for the lower SL-X based CPUs). Outside of being based on Skylake, the 7900K adds AVX-512 instruction sets to the mix, the first of its kind for Intel.

One thing of note, that isn’t listed in the specifications is the 13.75 MB of L3 cache available. This is actually LESS than what was available on 6950X! At first glance this seems like a problem, however, Intel has “rebalanced the cache hierarchy” to have it work more efficiently with less latency. A bit more on that later.

The Kaby Lake-X based part, the 7740X, a 7700K built on socket 2066 bones, gets its base clock increased to 4.3 GHz (from 4.2 on the Kaby Lake part) and the boost remains the same at 4.5 GHz. There is increased TDP of 112 W compared to 91 W for KL. PCIe lane availability remains the same with 16 PCIe 3.0 available when using this CPU. Also something to note again is this base CPU SKU will still be dual channel. Not that it will be a loss, but its still mixing church and state in the HEDT platform as it has only run triple or quad channel since the X58/Nehalem days. Another important fact is the K L-X parts have their IGP disabled. You will need a discrete GPU for the platform.

The larger PCB and additional pads are said to give the 7740X some more overclocking headroom, but only time will tell on that one. There were a couple of early leaks on the 7900K showing it running Cinebench with an AIO cooler at 4.7 GHz 1.25 V. Temps were banging off the 100C throttling point though. If you want to push much past 4.5 GHz, seems as though custom water is in store, and perhaps a delid/re-TIM to go to the extreme. Yes, I said a delid and re-TIM. They went ahead and did NOT solder the IHS on either CPU. Maybe the 12C+ parts will, not sure. But it seems curious that the solder is gone, though we are still seeing pretty good overclocks. Not sure why they went this route and it will certainly irritate some, but the reality is the processor can run just fine within its specifications and seemingly well outside of them with what is currently done.

Below is a picture from the Intel press documents on the entire product stack. At this time, the 12-18 core parts will not be available. We can expect to see them in the coming months.

Some of the ‘little guys’ at eight and six core, as well as the two Kaby Lake-X offerings.

Key Features (Intel Core X Fact sheet):

• New! Intel® Core™ i9 Extreme Edition processor featuring 18 cores and 36 threads
• New! Intel’s most scalable high-end desktop platform ever with 18, 16, 14, 12, 10, 8, 6, and 4- core options
• New! Intel® X299 chipset with improved I/O capabilities
• New! LGA 2066 socket for Intel® Core™ X-series processor family
• Additional system performance and amazing responsiveness with Intel® Optane™ memory support
• Improved Intel® Turbo Boost Max Technology 3.0 (select SKUs) for single- and multithreaded performance
• Up to 44 lanes of PCIe 3.0 directly connected to the CPU, to expand your system with fast SSDs, multiple discrete graphics cards and ultrafast Thunderbolt™ 3 technology
• Up to 4 channel DDR4-2666 memory support, support for the Intel® Extreme Memory Profile (Intel® XMP) specification, revision 2.0 for DDR4
• Fully unlocked for performance tuning
• Rebalanced smart cache hierarchy
• Up to 10 percent faster multithreaded performance over previous generation
• Up to 15 percent faster single-thread performance over previous generation

Below are some slides I picked out from an Intel presentation on the new CPU and chipset. The slides give us an idea of where Intel is coming from. There is parental bias I am sure, but it touches on how powerful and scalable it is, usage models, and even overclocking!

Obviously the High End DeskTop (HEDT) segment is made for those power users of a PC. Outside of the curious inclusion of Kaby Lake-X based CPUs with the fewest threads the platform has seen on it in years, each generation has given us faster CPUs, more threads, and generally more efficiently while doing so. Intel claims 10% faster multithreaded performance than the previous generation and up to 15% faster single-threaded. We will dig in later to see if there are IPC increases or if it is due to clockspeed only.

Intel still frames these CPUs in the Extreme Performance, Extreme Mega-Tasking, and dubbing it The Ultimate Platform. Seems like there is a use for everyone now. Speaking of uses, that 7740X – When you think about it, can be a cheaper stepping stone into the X299/Basin Falls platform with the potential to upgrade to more cores later/when needed. Gaming and Livestreaming is no problem here with all the threads there are to throw around.

Overclocking? They mention that too, offering Memory controller trim voltage control for higher and faster clocks, as well as PEG/DMI overclocking. Skylake-X processors still have the per-core overclocking and voltage control. If you are really pushing on the CPU and need some insurance in case you push too far, Intel offers the performance tuning protection plan for a small fee.

This slideshow requires JavaScript.

The slide below talks about the changes made to Intel’s Turbo Boost Max Technology 3.0 on the Skylake-X chips. Its latest implementation now chooses the best two cores on the die to reach maximum boost. This will help out with any dual threaded or single threaded application.

We talked about the 13 MB of cache available on the 7900X SKU earlier and how that was a bit more than half of the 25 MB on the 6950X. Intel has changed how it works internally in order to achieve lower latency and overall better performance. In essence, you are getting 5x more L2 (10 MB vs 2 MB) and less L3 (13.75 MB vs 25 MB). They have gone from 256 K of L2 private-local cache to 1 MB per core of the same. In theory, this can reduce how often the slower L3 cache is hit since more L2 is available.

A die shot of the 7900k from the Intel slide deck.

Product Tour

Below are some pictures of the CPUs themselves. A couple of things to note: First the ever so slight difference between IHS on the two processors. Next, the thickness of the PCB. In the last picture you are seeing a head on view of the 7740X (right) and 7700K (left) PCBs… moving to K L-X it now uses the thicker PCB.

Benchmarks and Test Systems

The data we have gathered will give us a great idea of its performance both at stock (no turbo), and matching clockspeeds to see IPC performance increases between Broadwell-E and Skylake-X as well as Kaby Lake and Kaby Lake-X.

Below are some pictures of the products used in the review. A special thanks goes out to our friends at G. Skill and ASUS for providing the RAM and Motherboard for review. G. SKill sent us a non-RGB (my choice) 4×8 GB DDR4 3600 MHz CL16 kit (not listed at their website yet in quad form), while ASUS sent the venerable Prime X299-Deluxe. I had zero problems on the platform in my testing. ASUS already had another BIOS ready to go (likely one today/tomorrow as well) and the RAM worked with simply enabling XMP. I am sure there will be some teething issues, but in my extremely limited time with the platform both at stock and overclocked, I didn’t have an issue.

This slideshow requires JavaScript.

And the test system picture:

Benchmarks Used

All benchmarks were run with the motherboard being set to optimized defaults (outside of some memory settings which had to be configured manually). When “stock” is mentioned along with the clockspeed, it does not reflect the boost clocks, only the base clocks. I tested this way as it seems motherboards are different in how they work out of the box. The MSI I have sets all cores at the max turbo (4.5 GHz) for the CPU while other boards follow the Intel Turbo specification. So, it is more of a “run what you brung” type of testing for stock speeds.

After testing, we then shifted to comparing the 7700K and 7740X at the same clockspeeds and memory settings. This testing will flesh out the difference in Instructions Per Clock (IPC) between the two.

CPU Tests

• AIDA64 Engineer (Memory, CPU, and FPU Tests)
• Cinebench R11.5 and R15
• x265 1080p Benchmark (HWBOT)
• POV Ray
• Intel XTU
• SuperPi 1M/32M
• WPrime 32M/1024MB
• 7Zip

All CPU tests were run at their default settings unless otherwise noted.

Gaming Tests

All game tests were run at 1920×1080 and 2560×1440. Please see our testing procedures for details on in-game settings.

• Rise of the Tomb Raider
• Crysis 3
• Dirt: Rally
• Ashes of the Singularity: Escalations

CPU Tests

Stock testing using the AIDA64 Engineer software is below. Please note, again, samples were heavily delayed in getting to the Finalwire team so these tests are not optimized for the AVX-512 instruction set and when it’s used/not used. So some numbers can be a bit different here when they get the final tweaks put in. As it stands I used their latest beta they suggested for the most accurate results at the time. We can see the results are pretty close together here with the higher clocked, newer chips taking a 2-3% lead across most of these tests. Others, like PhotoWorxx, we can see likely hasn’t been optimized as it is coming through a bit slower on the new CPUs.

FPU is showing results a bit all over the place from the Skylake-X chip losing by 10% to beating out its predecessor by over 15%. The 7740K results are close as expected since it’s the same CPU with higher clocks.

The memory test shows a tale of different memory speeds as some of these were recorded at 3200 MHz (6950x) while the rest were 3600 Mhz on the new platform.

Moving on to some real world benchmarks, outside of an outlier in x265, results are as expected and pretty close together.

Things were all over the map on the Hwbot benchmarks WPrime and SuperPi with the newer CPUs not putting their best foot forward on my test system and OS, especially with the 7900K. It’s 300 Mhz faster but doesn’t show it here for whatever reason.

Game Results

Game results are not going to be surprising to most. We set the CPUs at the exact same speed (6950X/7900X at 4 GHz and the 7740X and 7700K at 4.5 GHz) with the only difference being the ram speed and timings. Results are very close indeed. With the clockspeeds these can all achieve, I wouldn’t expect much of a difference unless the game tested can use more than eight threads.

Head to Head Results

Head to head results are listed in the slideshow below and show us, overall, of slight improvement over their predecessors again. The 7900X shows improvements of 10% or so in many AIDA tests, while others (the less optimized ones perhaps?) show no gain at all. The Kaby Lake-X CPU shows little to no improvement as expected since its the same thing but on a different platform.

Overclocking

I’ll be honest here… I wish I had more time with the platform. But I spent most of it just getting all the data you see above then updating the graphs, etc. To that end, with the 7900X, I was able to get it to 4.5 GHz using 1.3 V. I likely could have used less, but, just set it and went. We can see I was just about at my thermal limits with stress testing so we are just about topped out here without a delid. Most 6950Xs couldn’t reach 4.5 GHz (low 4’s for the most part) so we do see improved overclocking in Skylake-X versus Broadwell-E. These clocks also reach well past what any Ryzen can do as well.

Power Consumption and Temperatures

Power consumption for the 7900X was about in line with what I was expecting with the platform for the most part. Prime 95 (v 28.10) Small FFT really made the CPU scream peaking at 465 W at the wall. If you are pushing the 10 core far, you will want a robust motherboard (which the ASUS PRIME X299-Deluxe made a solid showing… if only with some coil whine) and robust cooling. The temps you see for 7900X are on a large custom loop. So the TIM is certainly good enough for normal use, but leaves something to be desired when pushing clocks. There is no doubt I wish there was solder on these high end parts still as they could be pushed even farther I would imagine.

Nothing out of the ordinary with the 7740X all around. Power use increased a bit on the newer, more power hungry platform, but remember the chip itself went from 91 W to 112 W as well.

Conclusion

So where did we end up with these CPUs? In a nutshell, the 7900X brought some minor IPC improvements in some areas in our our testing, while the 7740X is simply a faster 7700K on a more robust platform.Our testing, due to a not fully optimized AIDA64 test application, really blurred the lines in trying to get a clear read on performance. Platform immaturity also didn’t help our cause either. If you look around the web, we can see though that IPC was barely increased, so its mostly clocks and other tweaks that show improvement if any. Some. due to the difference in L3 cache, even performed worse, but this is use specific and not general performance.

Overclocking headroom is said to be improved on the 7740X due to the better/large PCB/substrate its on. Truth be told, I couldn’t confirm that in the time we had with the chips. By the time this is published, I would have had two full days with them. Since the Kaby Lake-X is old hat, I focused more on the 7900X with my overclocking. We see that reached 4.5 GHz and stable enough to run some all thread benchmarks and 35+ minutes of the standard AIDA64 stress test so its reasonably stable there. The concern, or really I should say limit, was with the temperatures. Intel has chosen to bypass the Indium solder for this stuff that doesn’t work as well as many enthusiasts would like it to. I guess delidding is in the cards if you really really need to push these things as volts are just not getting in the way, at least on my sample.

Intel improved upon its Turbo Boost 3.0 for the Skylake-X SKUs with the boost going up to 4.5 GHz, higher by 500 MHz than the 6950X, and now using TWO cores instead of one. From baseclocks to full Turbo Boost 3.0 is a large 1.2 GHz increase and will help with single threaded apps needing core speed for best results. Another notable improvement is the cache management. We saw incremental improvements in many tests with some of that having to do with the 4x increase in L2 private cache. Don’t judge a book by its cover with the 7900X having a lot less L3 on tap… the L2 is going to do its job in many cases negating the need to utilize the L3 as frequently. However in tasks (read: some compiles) it could use that beefier cache. Another item to consider on the Skylake-X chips is the increase in PCIe lanes available to 44 on the 7900X and up. The hex and quad cores come in with less at 28 CPU attached lanes, while the Kaby Lake-X processors sit with 16.

We haven’t talked too much about pricing in the article, but the 7900X comes in at $999 while the 7740X comes in at $339 (1K units). If we remember the 6950X release, that deca-core monster started out life as a $1700+ CPU on day one. Over time, and since AMD decided to get competitive on the performance side of things, that has since dropped to $999 also. So it starts out a lot cheaper, a great thing for consumers. The real competition on this platform for many of us though are the Hex/Octo parts, the 7800X and 7830X. These two are priced at $389 and $599 respectively. Will it isnt taking the price /performance crown from AMD’s hex/octo’s it is offering faster clockspeeds and slightly better IPC…and for this, you need to pay for the best. Pretty simple. But with that said, the pricing structure feels like its a lot more in order compared to pre-Ryzen days.

Overall, Intel has managed to bring to market another top performing formidable line of CPUs in the HEDT side of the house. This time, with a bit of influence, pricing is a bit more down to earth and competitive. You should find the 7900X on down in stores today.

Joe Shields (Earthdog)