Review Index:
Feedback

The Intel Core i9-7980XE and 7960X Review: Skylake-X at $1999 and 18-cores

Author:
Subject: Processors
Manufacturer: Intel

Specifications and Architecture

It has been an interesting 2017 for Intel. Though still the dominant market share leader in consumer processors of all shapes and sizes, from DIY PCs to notebooks to servers, it has come under attack with pressure from AMD unlike any it has felt in nearly a decade. It started with the release of AMD Ryzen 7 and a family of processors aimed at the mainstream user and enthusiast markets. That followed by the EPYC processor release moving in on Intel’s turf of the enterprise markets. And most recently, Ryzen Threadripper took a swing (and hit) at the HEDT (high-end desktop) market that Intel had created and held its own since the days of the Nehalem-based Core i7-920 CPU.

View Full Size

Between the time Threadripper was announced and when it shipped, Intel made an interesting move. It decided to launch and announce its updated family of HEDT processors dubbed Skylake-X. Only available in a 10-core model at first, the Core i9-7900X was the fastest tested processor in our labs, at the time. But it was rather quickly overtaken by the likes of the Threadripper 1950X that ran with 16-cores and 32-threads of processing. Intel had already revealed that its HEDT lineup would go to 18-core options, though availability and exact clock speeds remained in hiding until recently.

  i9-7980XE i9-7960X i9-7940X i9-7920X i9-7900X  i7-7820X i7-7800X TR 1950X TR 1920X TR 1900X
Architecture Skylake-X Skylake-X Skylake-X Skylake-X Skylake-X Skylake-X Skylake-X Zen Zen Zen
Process Tech 14nm+ 14nm+ 14nm+ 14nm+ 14nm+ 14nm+ 14nm+ 14nm 14nm 14nm
Cores/Threads 18/36 16/32 14/28 12/24 10/20 8/16 6/12 16/32 12/24 8/16
Base Clock 2.6 GHz 2.8 GHz 3.1 GHz 2.9 GHz 3.3 GHz 3.6 GHz 3.5 GHz 3.4 GHz 3.5 GHz 3.8 GHz
Turbo Boost 2.0 4.2 GHz 4.2 GHz 4.3 GHz 4.3 GHz 4.3 GHz 4.3 GHz 4.0 GHz 4.0 GHz 4.0 GHz 4.0 GHz
Turbo Boost Max 3.0 4.4 GHz 4.4 GHz 4.4 GHz 4.4 GHz 4.5 GHz 4.5 GHz N/A N/A N/A N/A
Cache 24.75MB 22MB 19.25MB 16.5MB 13.75MB 11MB 8.25MB 40MB 38MB ?
Memory Support DDR4-2666 Quad Channel DDR4-2666 Quad Channel DDR4-2666 Quad Channel DDR4-2666 Quad Channel DDR4-2666
Quad Channel
DDR4-2666
Quad Channel
DDR4-2666
Quad Channel
DDR4-2666
Quad Channel
DDR4-2666 Quad Channel DDR4-2666 Quad Channel
PCIe Lanes 44 44 44 44 44 28 28 64 64 64
TDP 165 watts 165 watts 165 watts 140 watts 140 watts 140 watts 140 watts 180 watts 180 watts 180 watts?
Socket 2066 2066 2066 2066 2066 2066 2066 TR4 TR4 TR4
Price $1999 $1699 $1399 $1199 $999 $599 $389 $999 $799 $549

Today we are now looking at both the Intel Core i9-7980XE and the Core i9-7960X, 18-core and 16-core processors, respectively. The goal from Intel is clear with the release: retake the crown as the highest performing consumer processor on the market. It will do that, but it does so at $700-1000 over the price of the Threadripper 1950X.

Continue reading our review of the Intel Core i9-7980XE and Core i9-7960X!

Architectural Refresher

There is very little new to discuss about the Core i9-7980XE and the Core i9-7960X processors. They use the same new architecture and design as the Core i9-7900X, though they are using a larger die. This die has up to 18-cores available on it, and it is utilized for the 12-, 14-, and 16-core HEDT processors as well. It uses the same mesh interconnect, has the same AVX-512 support, changes the cache weighting to the same ratio, includes the improved Turbo Max Boost 3.0 feature, and includes SpeedShift support.

If you want a refresher on what these technologies and features do, I have included that here, just below this paragraph. If you don’t need that, and instead want to jump straight into the new stuff including benchmarks and overclocking, I get it. Just click right here.

AVX-512

Although the underlying architecture of the Skylake-X processors is the same as the mainstream consumer Skylake line, which we knew as the Core i7-6000 series, there are some important changes thanks to the Xeon heritage of these parts. First, Intel has tried to impart the value of AVX-512 on us, each and every time we discuss this platform, and its ability to drastically improve the performance of applications that are recompiled and engineered to take advantage of it. Due to timing constraints today, and with a lack of real-world software that can utilize it, we are going to hold off on the more detailed AVX-512 discussion for another day.

View Full Size

Caching Hierarchy and Performance

We do know that the cache hierarchy of the Skylake-X processors has changed:

Skylake-X processors will also rebalance the cache hierarchy compared to previous generations, rebalancing to more exclusive per-core cache at the expensive of shared LLC. While Broadwell-E had 256KB of private L3 cache per core, and 2.5 MB per core of shared, Skylake-X moves to 1MB of private cache per core and 1.375MB per core of shared.

View Full Size

This shift in cache division will increase the hit rate on the lowest latency memory requests, though we do expect inter-core latency to increase slightly as a result. Intel obviously has made this decision based on workload profiling so I am curious to see how it impacts our testing in the coming weeks.

After more talks with Intel and our own testing, it’s clear that the changes made to the mesh architecture (below) and cache divisions have an impact on latencies and performance in some applications. Take a look at our cache latency results below:

Mesh Architecture Interconnect

I wrote about this new revelation that is part of both the Skylake-X HEDT consumer processors and the Xeon Scalable product this week, but it’s worth including the details here as well.

One of the most significant changes to the new processor design comes in the form of a new mesh interconnect architecture that handles the communications between the on-chip logical areas.

Since the days of Nehalem-EX, Intel has utilized a ring-bus architecture for processor design. The ring bus operated in a bi-directional, sequential method that cycled through various stops. At each stop, the control logic would determine if data was to be the collected to deposited with that module. These ring bus stops are located at memory controllers, CPU cores / caches, the PCI Express interface, memory controllers, LLCs, etc. This ring bus was fairly simple and easily expandable by simply adding more stops on the ring bus itself.

However, over several generations, the ring bus has become quite large and unwieldy. Compare the ring bus from Nehalem above, to the one for last year’s Xeon E5 v5 platform.

The spike in core counts and other modules caused a ballooning of the ring that eventually turned into multiple rings, complicating the design. As you increase the stops on the ring bus you also increase the physical latency of the messaging and data transfer, for which Intel compensated by increasing bandwidth and clock speed of this interface. The expense of that is power and efficiency.

For an on-die interconnect to remain relevant, it needs to be flexible in bandwidth scaling, reduce latency, and remain energy efficient. With 28-core Xeon processors imminent, and new IO capabilities coming along with it, the time for the ring bus in this space is over.

Starting with the HEDT and Xeon products released this year, Intel will be using a new on-chip design called a mesh that Intel promises will offer higher bandwidth, lower latency, and improved power efficiency. As the name implies, the mesh architecture is one in which each node relays messages through the network between source and destination. Though I cannot share many of the details on performance characteristics just yet, Intel did share the following diagram.

View Full Size

As Intel indicates in its blog on the mesh announcements, this generic diagram “shows a representation of the mesh architecture where cores, on-chip cache banks, memory controllers, and I/O controllers are organized in rows and columns, with wires and switches connecting them at each intersection to allow for turns. By providing a more direct path than the prior ring architectures and many more pathways to eliminate bottlenecks, the mesh can operate at a lower frequency and voltage and can still deliver very high bandwidth and low latency. This results in improved performance and greater energy efficiency similar to a well-designed highway system that lets traffic flow at the optimal speed without congestion.”

The bi-directional mesh design allows a many-core design to offer lower node-to-node latency than the ring architecture could provide, and, by adjusting the width of the interface, Intel can control bandwidth (and, by relation, frequency). Intel tells us that this can offer lower average latency without increasing power. Though it wasn’t specifically mentioned in this blog, the assumption is that because nothing is free, this has a slight die size cost to implement the more granular mesh network.

Using a mesh architecture offers a couple of capabilities and also requires a few changes to the cache design. By dividing up the IO interfaces (think multiple PCI Express banks, or memory channels), Intel can provide better average access times to each core by intelligently spacing the location of those modules. Intel will also be breaking up the LLC into different segments which will share a “stop” on the network with a processor core. Rather than the previous design of the ring bus where the entirety of the LLC was accessed through a single stop, the LLC will perform as a divided system. However, Intel assures us that performance variability is not a concern:

Negligible latency differences in accessing different cache banks allows software to treat the distributed cache banks as one large unified last level cache. As a result, application developers do not have to worry about variable latency in accessing different cache banks, nor do they need to optimize or recompile code to get a significant performance boosts out of their applications.

There is a lot to dissect when it comes to this new mesh architecture for Xeon Scalable and Core i9 processors, including its overall effect on the LLC cache performance and how it might affect system memory or PCI Express performance. In theory, the integration of a mesh network-style interface could drastically improve the average latency in all cases and increase maximum memory bandwidth by giving more cores access to the memory bus sooner. But, it is also possible this increases maximum latency in some fringe cases.

Turbo Boost Max Technology 3.0

With the release of the Broadwell-E platform, Intel introduced Turbo Boost Max Technology 3.0 that allowed a single core on those CPUs to run at higher clock speeds than the others, effectively improving single-threaded performance. With Skylake-X, Intel has improved the technology to utilize the TWO best cores, rather than just one.

View Full Size

This allows the 8-core and higher count processors from this launch to run at higher frequencies when only one or two cores are being utilized. In the two products that we have clock speeds for, that is a 200 MHz advantage over standard Turbo Boost technology. Intel hopes that this improvement in the technology gives them another advantage in any gaming or lightly threaded workload over the AMD Ryzen and upcoming Threadripper processors.

SpeedShift on HEDT

For the first time, the HEDT platform will get SpeedShift technology. This feature has been present since the launch of Skylake on the consumer notebook line, was updated with Kaby Lake, and now finds its way to the high performance platforms. The basis of the technology allows the clock rates of the CPU to get higher, and do so faster, in order to improve the responsiveness of the system for short, bursty workloads. It accomplishes this by taking over much of the control of power states from the operating system and leaves that decision making on the CPU itself.

View Full Size

Zoomed

Comparing the Core i9-7900X to the Core 7-6950X (that does not have SpeedShift) and the Core i7-7700K (Kaby Lake) shows the differences in implementation. The 7900X reaches its peak clock speed in 40ms while the Broadwell-E processor from last year takes over 250ms to reach its highest clock state. That’s a significant difference and should give users better performance on application loads and other short workloads. Note the difference on the 7700K though: the consumer part and Kaby Lake design is even more aggressively targeting instantaneous clock rates.

X299 Platform Still Going Strong

Though nothing has changed on our test bed for this review, it’s worth noting that there has been continuous updating around the X299 chipset and motherboards that use it. ASUS has released a handful of newer BIOS for the X299-Deluxe that we used on the 7900X story and for today’s review, that have improved stability, made overclocking easier, and even improved our storage performance (more on that at a later time).

View Full Size

A big thanks to our friends at Corsair for hooking us up with a collection of new RM1000x power supplies, Vengeance LPX 32GB 3200 MHz memory kits, and Neutron XTi 480 GB SSDs to upgrade all of our CPU testing platforms! They have helped streamline our testing process great.

View Full Size


September 25, 2017 | 03:43 AM - Posted by Puusari (not verified)

Great performance alright, but its a little evident that you try to paint an okish picture for Intel on the value front. They are asking double the price for how much..like 30% more performance. Emphasis on the cost benifit is lacking imo. Intel 4790 owner here.

September 25, 2017 | 05:03 AM - Posted by Martin (not verified)

...or 70% more for +15% performance.

7940X sppears to be the most interesting one of the Intel bunch so far. At $1300, it's "only" 30% more expensive than 1950x and so far it looks like it might match the performance.

September 27, 2017 | 02:29 AM - Posted by throAU (not verified)

In some markets, particularly those who buy such parts, performance per dollar is less relevant than outright performance.

e.g., some oil and gas model work a colleague has been involved in, the company was willing to buy a heap of workstations worth 30-50k to make the simulations run as fast as possible, as improving efficiency could be worth hundreds of thousands or millions of dollars per day.

in that context, saving a grand on a CPU but giving up ANY of your performance is a bad choice.

Sure,those markets are few and far between, but they do exist, and for those people, intel is a better choice in this segment.

For the vast majority though, who aren't dealing with those sorts of efficiency improvements and ROI improvements its a different story.

But for a lot of THOSE people, the HEDT platform isn't really required or worth the money anyway.

Its certainly a consideration worth mentioning anyhow.

September 27, 2017 | 02:36 AM - Posted by throAU (not verified)

Oh, and i write that as an AMD fan by the way.

I'd never buy one of those processors, and I'm deciding between Ryzen and ThreadRipper for my next build at the moment.

But I'm also an experienced IT veteran of some 20 years, and have been around long enough and seen enough to know that sometimes its not worth trying to force square pegs into round holes.

You use the most appropriate tool for the job, and in some cases, it is WORTH spending double, triple or many more times the price if the ROI is worth it.

September 25, 2017 | 03:54 AM - Posted by Skree (not verified)

Who cores?

September 25, 2017 | 04:05 AM - Posted by BobbyVal (not verified)

AMD panic mode. Price drop imminent!

September 25, 2017 | 05:19 AM - Posted by biohazard918

umm why? They already have a crushing advantage in performance per dollar. What would be the point when thread ripper is already very competitive for anyone paying for their own hardware.

September 25, 2017 | 05:26 AM - Posted by quest4glory

Pretty much this, I don't see very many people paying an extra grand for 40% more performance, overclocking it and then adding something like 2 480mm radiators + whatever else they can spare just to try to keep the processor cool.

September 25, 2017 | 08:09 AM - Posted by biohazard918

Eh cooling is much less of an issue then power delivery on the mobo if I had to guess. Look at how much heat amd dumped though a 120mm rad with there dual gpu cards. The numbers I am hearing are plus 500 watts which is in the same range as a 295x2. Just need a good block and pump. Now the real question is will your mobo catch on fire.

September 25, 2017 | 11:13 PM - Posted by quest4glory

Not really. GN has temp data for things like a 30 minute Blender run on an overclocked 7960X at 85C, with the 7980XE sometimes jumping up to 90. That is with a 360mm radiator. They had to delid and replace with liquid metal just to decrease the temps 10-15 degrees. Still too hot for my taste and most serious enthusiasts, I'd wager.

September 25, 2017 | 03:04 PM - Posted by Streetguru

The most they might do is release a 32 core Threadripper chip, won't be quite as good as the EYPC equivalent on Memory/PCI-e, but keeps their crown in this market segment.

September 25, 2017 | 05:25 AM - Posted by PcShillSuxDick (not verified)

Edited because the author was stupid.

September 25, 2017 | 05:53 AM - Posted by quest4glory

Way to keep it classy.

September 25, 2017 | 08:30 AM - Posted by CB

Yeah, Ryan's totally a shill. I mean, who would want to look at benchmarks and make an informed decision?

And, shills always point out how to other product wins handily in performance per dollar. /sarcasm

Nice review Ryan. Some of us actually enjoy reading the articles and benchmarks. I, for one, know that these products are out of my price range and use case but still love reading about the absolute bleeding edge of tech.

September 25, 2017 | 05:59 PM - Posted by DemXsEnSuchFiles (not verified)

All the online websites are in the compamnies's pockets for ad revenuse and review samples! So where have you been living lately as that's not new news industry wide. You have to go and read many reviews across many online news/review sites to get the proper view of reality. It has always been thus in Crony Capitalist America, or Scamerica as I like to call it.

So some websites spin one way and other websites spin the other and more research is reqired to get closer to the truth! Read as many reviews form as many sources as possible and even other poster's benchmarks on reddit/Anandtech forums/other sites. It's all X-Files with the benchmarking/review process anyways and the Truth Is Out There to find.

September 25, 2017 | 05:48 AM - Posted by Odizzido2 (not verified)

That's pretty damn expensive for what it offers.

September 25, 2017 | 05:53 AM - Posted by quest4glory

Some people are more than willing to pay the Intel tax. Same goes for Apple, luxury automobile makers, etc. etc. etc.

September 25, 2017 | 05:54 AM - Posted by Prodeous_at_work (not verified)

Honestly I'm happy about this release.

A. it clearly shows Intel finally felt threat from AMD, and put a Xeon in the Desktop environment.

B. AMD's TR is not that far back at 1/2 the price. and even power wise it is not that bad.

C. This makes me get the TR 1950 even more... Well done Intel :)

D. If money was no object, that new Intel chip would be to drool for. though if money was No Object, I would get a quad Xeon or Epyc setup :)

September 25, 2017 | 07:07 AM - Posted by FQ (not verified)

Exactly, I don't have issue with money but I'm not going to spend my money for no reason. Especially at my age (getting closer to 40). Getting the most expensive setup is stupid. I have no use for it and can use that money to invest instead. Even the the threadripper is way over what I need to use. I'm still debating if I should get a Ryzen as my Phenom II 1090T is serving me well for the last 7 years.

September 25, 2017 | 10:04 AM - Posted by Benjamins (not verified)

You should see TR4 NVME Riad results
https://www.youtube.com/watch?v=5cH9orZ9Qtg

September 25, 2017 | 07:33 AM - Posted by Anonymous69854255 (not verified)

AVX-512 - Is it effective for video software?
Encoded?
Movies ?
Real time conversion?
What video software works with AVX-512 ?

September 25, 2017 | 10:41 PM - Posted by James

I don't think AVX-512 is a very good idea. It should be able to do 16 32-bit operations per clock per unit, but even a small GPU can do hundreds, if not thousands of operations in a clock. The CPU clock is higher, but not by that much, and it may have to downclock the units due to power consumption. The GPU will probably be significantly more power efficient as far as performance per watt. If you have an application that could make use of 512-bit vector units, then the developer of that application should be porting it to run on a GPU. Hopefully servers that don't need much of any FP at all will not burn extra power due to the AVX units.

September 26, 2017 | 09:47 AM - Posted by ThreeCardIntelMonte (not verified)

Yes that's why Intel's Bean Counters are doing that 1 AVX 512 unit on some and 2 AVX 512 units on others and users having something else to have to pay extra($$$$$) for Intel to enable with "features" like RAID keys and other such product segementing schemes to milk those cows even more for profit. Those drops in Intel's sky high CPU SKU pricing will have to be made up in CPU feature reductions and elsewhere across Intel's CPU/MB platforms to try and claw back some of those high margins than Intel runs on.

It's Just more of Intel's 3 Card Monte where Intel trys its best to shuffle higher costs onto the MB/other CPU parts with features that are disabled that can be re-enabled for a price! Watch the Features, Watch those Features, can you guess where they have gone, lower pricing here and higher cost there to re-enable for a price($$$$$)! One Intel hand gives the other takes away, so watch those cards and not your pockets as thoes extra set of Intel claws reach deep down into your wallet and your pockets to snag every last bit of change!

September 25, 2017 | 09:21 AM - Posted by Benjamins (not verified)

Are we going to see x399 NVME raid tests. Der8auer has a video out about it.

September 25, 2017 | 10:27 AM - Posted by Mr.Book

Thanks for the review Ryan. On a related note, I would like to have seen platform costs included in the review.

Cheers.

September 25, 2017 | 11:03 AM - Posted by Manda (not verified)

Well, people being able to buy these kind of "enthusiast" rig isn't by getting wrong decision often. And that 1000$ difference looks like one.

September 25, 2017 | 11:29 AM - Posted by JustDoTheMathOnEpycSingleSocket (not verified)

Now let's look at the cost of the Epyc 7401P($1075) at 24 cores/48 threads and compare that to the cost of the Threadripper 1950X and Intel's HEDT SKUs. So for any Blender rendering workloads and other workloads AMD's 24 core/48 thread Epyc 7401P($1075) only costs $76 dollars more than the Threadripper 1950X($999) and a lot less than Intel's top end HEDT offerings. Then there is the the Epyc platfotm's Standard across all Epyc single and dual socket SP3 motherbpard SKUs support for 128 PCIe lanes and 8 memory channels with the single socket Epyc motherboards running in the $450-$625 range. Which is not too bad considering that the Epyc/SP3 motherboards usually offer dual 10Gb ethernet support and are fully Tested/Certified and validated for ECC memory with that ECC usage supported in the warrenties of both the Epyc/SP3 motherboard parts.

So the Epyc SP3 motherboards all support 128 PCIe lanes and 8 memory channels and actually are run through a standard series of rigorous Certification/Validation testing for ECC memory and the Epyc/SP3 MB Parts get the 3+ year warrinties and extended product availability and support that comes standard with the REAL Workstation/server Grade parts.

So on a feature for feature basis the Epyc 7401P and SP3 motherboard options have twice the PCIe lanes and twice the memory channels as any Threadripper CPU/X399 MB SKUs. And the Epyc/SP3 motherboard's(128 PCIe lanes/8 Memory Channels) costs less(even at $625 for the Gigabyte SP3 MB) on a MB-price/PCIe lane and MB-price/memory channel pricing/feature metric than even a $349 Threadripper X399 motherboard with its only 64 PCIe lanes and 4 memory channel support.

If you look at AMD's single scoket Epyc/SP3 MB options and their standard featre sets and the Epyc 7401P/Other "P" single socket processor price/featre offerings then that comes out better than the Theradripper/X399 MB platform on a feature for feature basis and way better than any Intel top end HEDT offerings as well. AMD's Epyc/SP3 MB options are the reverse of Intel's Xeon/Xeon MB options that cost much more on a feature for feature basis than any AMD/Intel consumer/HEDT variants with the Epyc/SP3 "P"/single socket SP3/MB actually a better feature for featire deal than AMD's consumer/HEDT Threadripper for the Real Epyc Branded Workstation/Server parts.

September 25, 2017 | 04:12 PM - Posted by James

So, TL;DR

September 25, 2017 | 11:54 AM - Posted by vodkakarhu (not verified)

the big guesting left in my mind is who is this cpu for?
draws too much power to be usefull in task where you make money pre wat. dosent have pcie lanes. expensive for streamers if cheaper cpu can do the same task. no ecc memory support at that price.
so what left youtubers that want to make videos bit faster?

September 25, 2017 | 04:17 PM - Posted by James

It is for the marketing department. Reviewers review it and it clearly gets higher bars on a bunch of thos test. That is the main purpose that they released it. They don't really have to sell any. An 18 core part is probably very low volume, even for Xeon parts. Price wise it may even be better to get a dual socket board and much cheaper processors even if you are staying with Intel. It will be almost non-existent in the consumer space.

September 27, 2017 | 02:41 AM - Posted by throAU (not verified)

Oil and gas models / simulation.

September 25, 2017 | 12:56 PM - Posted by cracklingice (not verified)

TDP is the amount of the heat generated by the processor and is not the amount of electricity consumed. Think of TDP more as the amount of power that isn't used in performing the work, but instead gets wasted generating heat.

September 25, 2017 | 02:58 PM - Posted by Streetguru

Every watt that goes into a CPU is converted to heat, same for the VRM.

September 25, 2017 | 01:09 PM - Posted by beb (not verified)

no ECC support ? and 2k$ ! I would just go for an AMD Epyc for 2.1k and I would get 32c/64t + 128xPCIE.

September 25, 2017 | 01:10 PM - Posted by beb (not verified)

+8ch memory!

September 25, 2017 | 04:31 PM - Posted by EvenMorePlusesOnEpyc (not verified)

+ The Epyc CPU/Socket SP3 Motherboards are all certified/tested/vaildated for ECC memory! Not So for any consumer Threadripper/X399 or Consumer Intel/HEDT/MB CPU/motherboard SKUs. And the Epyc/SP3 MB SKUs being fully certified/tested/vaildated for ECC memory uasge with the Epyc CPU's/Socket SP3 MB's ECC ability guaranteed in the parts' warranty.

+ The Epyc/SP3 motherboards often have dual 10Gb ethernet support and remote/headless server support/management chips and 3+ year CPU/MB warrienties.

+ The Epyc CPU/SP3 MB SKUs also come with extended periods of product support for Firmware/drivers/other support and extended product availability support. So the Epyc/SP3 MB parts will be offered for sale over an extended time period compared to the consumer branded parts.

+ The Epyc Server/Workstation CPUs/Socket SP3 have extra features like multi-GPU(Radeon Pro WX/Radeon Instict) peer to peer support on the Epyc/Socket SP3 grade motherboards for that and CPU/GPU Hardware/Firmware/OS virturalization features above and beyond what is available on any consumer CPU/Motherbiard platforms.

September 25, 2017 | 01:22 PM - Posted by shot (not verified)

No thread to thread latency tests?

September 25, 2017 | 04:28 PM - Posted by James

It should be pretty much flat for all cores for the Intel parts. It will be higher than the latency on a 4 core Intel chip (or a 4 core AMD CCX), and they are burning a lot of power to provide that uniform latency. With Zen, you get NUMA like memory access, but that can be optimized for, so it isn't an issue. The data movement then stays more in a 4 core cluster. The benchmarks and some of the apps here may not be NUMA (or at least AMD's specific implementation) optimized. I wouldn't put as much weight on the synthetic banchmarks as I would for a normal desktop part. The Pro apps should be made to take advantage of NUMA so Threadripper may perform differently in NUMA or interleaved mode. I didn't see what mode Ryan ran these test in.

September 25, 2017 | 04:51 PM - Posted by EvenMorePlusesOnEpyc (not verified)

Threadripper is not the best deal for Real Workstation usage, get over that TR/"Workstation" obsession! Epyc is the better deal over any Consumer Grade CPU/MB parts from AMD or Intel!

AMD is not Intel and The Epyc CPUs and Socket SP3 motherboards options especially the 1P Epyc CPU/single scoket SP3 MB SKUs are a better deal than any Theradripper "Workstation" wannabe SKUs!

AMD's Single Socket Epyc/SP3 platform Rules the price/performance metrics and Graphics users have a better chioce that's actually the professional choice. And That ECC memory support is fully tested/certified/validated for any Epyc/SP3 MB parts!

September 25, 2017 | 10:54 PM - Posted by James

I didn't say anything about Threadripper for workstations. People probably are buying it as gaming/encoding machines though. That isn't a bad use for it. Not everyone cares about ECC and other workstation features. Also a Threadripper 1900x is only 550$. Not everyone will want to drop 1000$ or more on the top Threadripper. Also, it is unclear how much Epyc boards will cost. TR boards seem to be about 350$ with 4 memory channels routed. Epyc will have all 8 memory channels plus a lot more pci-e. Also, you need to populate all 8 channels for full bandwidth. It will be significantly more expensive than a TR board and 4 memory sticks. It also will probably not be an ATX board. They will probably be E-ATX. Even though low end Epyc may be close to the top TR in price, the platform cost will not be the same. You should stop trying to push Epyc over TR when you don't know the actual cost.

September 26, 2017 | 11:22 AM - Posted by ThreeCardIntelMonte (not verified)

"you need to populate all 8 channels for full bandwidth"

But you can save money by getting the smaller capacity DIMMs and if you do want ECC memory the smaller capicity ECC DIMMs are much more affordable. So with 8 memory channels there is the option of populating across the 8 memory channels on Epyc while still being able to have a large total memory capicity of 32GB with for example: 8, 4GB ECC DIMMs(much lower cost) and leave each of the 8 channels' other memory slots emmpty. With TR's only 4 memory channels you will have to populate both the memory channels slots and that can cost memory/latency speeds/efficiency once you start going higher capacity with the DIMMs and using both the channel's DIMM slots.

There is regestered ECC DIMMs(Real costly) and Unregistered ECC DIMMs less costly, but if you start to populate both the channel's 2 DIMM slots then you are going to need Registered ECC DIMMs, while with only one DIMM slot per channel populated you can go with the lower cost unregistered ECC memory.

With 8 memory channels you get twice the effective bandwidth or you can NUMA mode up into 4 NUMA nodes, each with it's own dual channel primary/near memory store. It all depends on what workloads work best with what mode.
So On Epyc systems that Virtuilization Of CPU/GPU IP in the Zeppelin Die is all fully enabled and on some Epyc/SP3 Motherboards there is also support for AMD GPU to AMD GPU peer to peer networking among multi-GPUs.

So that's also an extra benefit, and who Knows what kinds of GPU Peer to Peer over that Infinity Fabric Protocol may be enabled for any Vega micro-arch beased GPUs with some future MB/Firmware updates on any Epyc SP3 based MB platforms. Epyc CPUs Speak Infinity Fabric protocol and so do Vega micro-arch based GPUs. And Infinity Fabric on Vega is similar to Nvidia's NVLink with the Infinity Fabric being a level above PCIe for those similar to Nvidia's NVLink sorts of direct GPU to GPU interfacing and CPU to GPU Interfacing.

There is a lot of IP in Vega that will be enabled once the software/driver/firmeware is all fully validated for that more Direct Sorts of Infinity Fabric sorts of Interfacing.

Sure Threadripper can be used for some encoding workloads but that Epyc(24 core/48 Thread) 7401P's price per core($44.80) is even less than the 1900X's(8 cores/16 Threads) $68.75 per core cost. And the Epyc SP3 motherboards cost($625 for that single socket Gigabyte Epyc/SP3 MB) has a per PCIe lane cost(Epyc SP3 MB's cost/#PCIe lanes, 128 PCIe lanes) of $4.88 Per PCIe lane, with the TR($350) MB-Price/#PCIe lanes(64) costing $5.47 pre PCIe lane. And Epyc's SP3 MB(Gigabyte of $625/#Memory channels(8) is $78.13 per memory channel while that TR/x399 MB $350/#Memory channels(4) for a per memory channel cost of $87.50 per memory channel.

So Epyc's cost/feature is less even though the overall cost of the Epyc 7401P/SP3 MB costs more than that TR $350MB/$550-1900X combo, and I did not even consder the Gigabyte Epyc/SP3 MB's dual 10Gb ethernet/other features supported that are Certified/Fully validated features on Epyc and not on the Threadripper/X399 platform.

September 25, 2017 | 02:28 PM - Posted by JohnGR

This platform is for the stupid and those who want the best. For the stupid who will throw hundreds of dollars on an expensive 2066 motherboard and get a quad core, thinking they will upgrade "cheap" in the future(having payed in front for the more expensive motherboard, that's why I call them "stupid") and those who just want the best, either because they have dollars to burn, or because they are professionals and time is money for them. Everyone between will be better with a Ryzen 7 or a Threadripper.

September 25, 2017 | 06:32 PM - Posted by James

If you are willing to spend 2000 dollars for the best performance, then you might want to wait and see what Epyc can do for the applications you care about.

September 27, 2017 | 10:32 AM - Posted by EpycCurbStompsIntelandTRInPricePerFeatureValue (not verified)

The Epyc 7401P(24 core/48 threads, at $1075) costs only $76 dollars more than the Threadripper 1950X(16 core/32 thread, at $999). And there are a least 2 single socket Epyc/SP3 motheboard's listed for sale now(1 Tyan at around $450, and One Gigabyte for $610-$625) so if you take the number of memory channels offered on the Epyc/SP3 MB's(128 PCIe lanes, 8 memory channels), then on a total MB/cost per feature metric the Epyc/SP3 motherboards come out costing less on any MB-Price/Per-PCIe lane and MB-Price/Per-Memory-Channel metric, even on a $625(Gigabyte Epyc/SP3 MB) price compared to a $350(Threadripper/X399 MB) price.

And the Gigibyte Epyc/SP3 motherboard at even $625 is offering dual 10Gb Ethernet and single 1Gb Ethernet allong with other workstation/server Grade features like fully certified/validated ECC tested support for ECC memory.

So it's better to look at those Price/Feature metrics and do a proper cost/benefit analysis because the Epyc single socket platform offerings are an even better feature for feature deal than any Threadripper/X399 platform offeringes for the Real Workstation/Server Epic/SP3 MB feature sets at a better price/feature metric.

September 25, 2017 | 02:58 PM - Posted by Streetguru

What's up with the 12/16 Threadripper chips having the same power consumption?

And why are the 10 core parts so far apart in power consumption?

September 25, 2017 | 06:30 PM - Posted by James

The lower the core count, the higher the base clock is for Threadripper.

September 25, 2017 | 03:46 PM - Posted by agello24 (not verified)

make sure before you purchase the i9-7980XE, that you lube up well and have intel use protection. cause they are about to fuck some buyers raw!

September 26, 2017 | 02:09 PM - Posted by malakudi

Something is wrong here. Ars Technica could not overclock i9-7960X over 4100MHz @ all 16 cores without hitting thermal throttling and you are able to overclock i9-7980XE at 4300MHz @ all 18 cores? They even used 360mm water cooling and still could not avoid thermal throttling above 4100 MHz, you have used 240mm one. So what is going on here? If pcper just has a "golden" chip, we cannot make conclusions about overclocking capability.

The article mentions that there were no stability issues at 4300, but it doesn't mention if thermal throttling occured on longer running tests, making the overclock result irrelevant.

September 30, 2017 | 11:11 AM - Posted by NumberCruncher (not verified)

Ah, did the author intentionally leave out the fact that AMD is exposing a whopping 64 PCIe lanes when evaluating these processors???

Clearly INTEL is the one in full panic mode here - like I've never seen them...

October 1, 2017 | 06:17 AM - Posted by Anonymouse (not verified)

Intel 18C SL-X -$2000

AMD 32C EPYC 7551P -$2100

Want a *real* workstation ? You don't want Intel then. Because single socket EPYC buries everything they've got. But the 'press' never mentions it. They just keep comparing $2000 Intel chips to $1000 Threadripper.

Bunch of shills. You're right, workstation buyers *don't* care about cost so much, which is exactly why they will spend the extra $100 on an EPYC, and leave this expensive, hot, zero value for money 'kidde 18C' junk from Intel in the trash, where it belongs.

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

By submitting this form, you accept the Mollom privacy policy.