Intel Fills In the Gaps on Core X-Series Processor Specs

Subject: Processors | August 7, 2017 - 01:34 PM |
Tagged: X-Series, processor, Intel, cpu, Core i9, core i7

Intel launched the first half of its X-Series processor lineup earlier this year, releasing up to the 10-core i9-7900X. But with the upcoming release of AMD's 16-core Threadripper 1950X, the real interest among enthusiasts are the specs of Intel's high core count X-Series parts.

View Full Size

After previously teasing partial specs for these parts, Intel today finally unveiled the complete details, starting out with the i9-7920X (12 cores/24 threads) with a 2.9GHz base and up to 4.4GHz boost clock and topping out with the i9-7980XE (18 cores/36 threads) with a 2.6GHz base and 4.4GHz max boost clock. Check the table below for the complete specifications:

  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 22 MB 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

From a pure core-count perspective, the Threadripper 1950X goes up against Intel's i9-7960X, but with a $700 difference in price. With Intel CPUs holding and IPC advantage over AMD, however, it's likely that the i9-7920X, and perhaps even the 7900X, will best Threadripper in certain gaming and productivity workloads.

Also interesting in Intel's announcement today are the base clocks of the 12-core i9-7920X (2.9GHz) and the 14-core i9-7940X (3.1GHz). Intel pushed the TDP of the 7940X to 165W, allowing it to increase the base clock over its 12-core counterpart. This suggests that Intel expects the 14-core 7940X, at a price-point of $1399, to be a popular choice in terms of price-to-performance.

Finally, Intel's release today reveals that all of the upcoming X-Series parts will have 44 PCIe lanes, compared to the 64 lanes AMD is offering on all Threadripper parts. There was some debate in the office this morning about how Intel's 44 lanes should cover most configurations for the foreseeable future, but this still remains one clear advantage for AMD's platform.

Intel's 4- to 10-core processors are already on the market. Intel says that the 12-core 7920X will launch August 28th, while the 14- to 18-core parts will launch about a month later, on September 25th.

Source: Intel

August 7, 2017 | 02:04 PM - Posted by Anonymously Anonymous (not verified)

So, Intel released a "1920x" as well?

"... After previously teasing partial specs for these parts, Intel today finally unveiled the complete details, starting out with the i9-1920X (12 cores/24 threads) with a 2.9GHz base and up to 4.4GHz boost clock ... "

August 7, 2017 | 04:44 PM - Posted by djotter

Yeah that i9-1920X threw me for a minute too. Meant to be i9-7920X I would think.

August 7, 2017 | 10:13 PM - Posted by Ryan Shrout

Sorry, we meant Core i9-7920X, obviously. Thanks!

August 7, 2017 | 04:04 PM - Posted by HEXiT (not verified)

i think there being a lil optomistic with there 140w projections. then again there probs measuring it differently to other manufacturers...

August 7, 2017 | 04:51 PM - Posted by Mr.Gold (not verified)

Intel
140w : 12 core * 2.9ghz = 34.8
165w : 14 core * 3.1ghz = 43.4

If you scale by power, both CPU are near identical.

AMD
180w : 16 * 3.4ghz = 54.4
(If you scale to 140w, the score is 42.3)

Intel 14+nm seem to be very power hungry

August 7, 2017 | 08:49 PM - Posted by James

I am still suspecting that the new cache interconnect is consuming a lot of power. It does provide a unified L3 cache at relatively low latency, but it will be moving a lot of data on relatively long interconnect to do it. Interconnect burns a lot of power. I don't know if there is a good way to test this though. I expect AMD to have a power consumption advantage since they have very short interconnect in a 4 core cluster. Massively parallel software should already be optimized for NUMA, or NUMA-like, architectures.

August 7, 2017 | 04:43 PM - Posted by Mr.Gold (not verified)

180w for 8 core.... This is what you expect the 1900x to be PCPer ?

Why dont you write N/A like all the other slot that are not disclosed?

ARGH! you are turning into Techreport...

August 7, 2017 | 05:52 PM - Posted by biohazard918

Rated tdp does not equal real world power consumption. There have been many instances of a tdp of x being applied to a line of cpus and the lower end models using substantially less power then high end parts with nominally the same tpd.

August 7, 2017 | 06:26 PM - Posted by Mr.Gold (not verified)

I agree. its been proven and its even more of an issue with "APU"s.
(performance vs TDP)

The thing that I see now is theoretical performance hitting the TDP wall.

#1 on the list is the X AVX512 support. It does improve IPC dramatically, but it also raise TDP the same. so the CPU find itself working at its lowest clock. reducing its benefit.

August 7, 2017 | 04:47 PM - Posted by Mr.Gold (not verified)

The TR core does have XFR... it seem by 200mhz
Not sure how applicable turbo boost 2.0 is on Zen ?

This is why the stock 1950xz score over 3000 in Cinebench. (3.6ghz clock on all 16 cores. 3.4ghz base + 200mhz xfr)

August 7, 2017 | 06:29 PM - Posted by ID10T (not verified)

Lol at people who spent $1600 on a 7940x

August 7, 2017 | 06:29 PM - Posted by ID10T (not verified)

Edit

7960x

August 7, 2017 | 07:10 PM - Posted by Clmentoz (not verified)

The 1950X(16/32) has a 3.4 GHz base clock over its 2 Zeppelin dies/cores while the i9-7960X(16/32) has on its single large monolithic die a base clock of ony 2.8 GHz. So the TR 1950X has the higher sustained all core clocking ability and there needs to be focused testing done for sustained all core boost with various cooling solutions on all of these SKUs.

I also think that AMD may have some Radeon Chill like software for the TR CPU SKUs to maybe clock boost up one of Threadripper's 2 Zeppelin dies more agressively than the other's die, owing to those two dies being thermally and physically separated under that large TR IHS/lid. Intel will have to contend with the thermals on its single large monolithic die designs in a more constrained manner as all the heat generated by any overclocked cores will all soak into the same monolothic die's silicon. AMD can effectively treat TR's 2 Zeppelin dies like singlular 8 core dies as far as CPU clock/thermal regimens are concerned.

And we will see how far that AMD/overclockers can take advantage of TR's unique, isolated thermally/physically, Modular die construction for better sustained overclocking on one of TR's 2 dies as opposed to the other and maybe keep 8 cores running at a higher sustained boost relative to the other 8 cores on the other die.

AMD does need to provide some middleware to manage TR's thermals with some form of gaming profiles to allow users to set the game's affinity to one Zeppelin die while the other Zeppelin die gets assigened the OS/Bloat, with the gaming die boosted higher and the OS/Bloat die held at lower clocks. There can even be the option give to TR/1950X users the ability to clock/group via CCX unit custom groupings and get the same relative higher sustained clock speeds acorss 8 of TR/1950X's 16 cores as the TR/1920X gets with only 8 pf its cores enabled across 2 dies is rated for(3.8 GHz).

It all dependes on the any game's needed core/thread resources and how that can be managed on Threadripper to an advantage.

AMD should have some gaming settings/profiles that let the overclocking community experiment with creative themal management via setting a game's profile to spread work across TR's 2 Zeppelin dies in different ways and see how that works out for the individual game's performance, for games that may not need all 16 TR CPU cores to run their gaming workloads. There can be some latency advantages also for gaming profiles with respect to CPU core affinity managed and thread assignment/reassignment managed on TR by the CCX, die, or for combinations of CCXs across both of TR's 2 modular Zeppelin dies.

So TR is not like very much like any single monolithic die design as far as thermals are concerned and Games and Gaming/Overclocking is going to be more interesting across Threadripper's 2 modular dies.

August 8, 2017 | 02:22 AM - Posted by Anonymouss.s.s.s (not verified)

I think all you are saying about use diffferent modes and frecuencies when enabling cores / threads distributed at the ccxs over the dies over the mcm is already done using senseMI.

I remember a slide of a PPT or some stuff in this way and a description on how the amount of cores was rising (and its phisical location inside the chip based on take advantages of Infinity Fabric cache L3 and the physical distance along the diagonal of the acrtivel dies inside the MCM) from only 1 to the maxcore of the sku.

The were different options in some core counts to get the max frecuency to the lower number of core / threads (games?) in the same ccx/die or run this amount of cores (productivity) selecting only one core of each ccx opposited physically at the two dies and opposited ccx too.

Seems a clever use of the massive mcm size and takes advantage of the spaced dies needed by the infinity fabric architecture and the 2 chip solution in a 1 chip idea of the Threadripper.

It's very near people can put its hands on the TRs and tweak them, i hope see a wide performance OC results to get great (best?) performance at less core or max core or get a more balanced very good performance at less core and max core, but nor the best por sure. Each customer may take the SKU really fits best with the core count / frecuency performance ratio regarded by the software is going to use.

August 7, 2017 | 08:23 PM - Posted by Scott Baker (not verified)

Does skylake-x have all of the skylake problems or did Intel fix them?

August 7, 2017 | 08:25 PM - Posted by Scott Baker (not verified)

Could you guys add an option to be emailed if someone replies to our posts?

August 8, 2017 | 09:05 PM - Posted by Tim Verry

This is a feature I want to see as well, but I'm not sure how possible/likely it is to see that on the current version of the site.

August 7, 2017 | 08:44 PM - Posted by RectangularNotRound (not verified)

Big and rectangular, looks as big as TR's IHS for the metal contacting metal(contact area via thermal compound) on these waterblocks. Hope to see more like this becoming available in the US. Going to need some big tubes of thermal compound.

"The blocks without the RGB lighting are available for around 58 USD (388 CNY), while the RGB-supported blocks are available for 63 USD (428 CNY). Threadripper water blocks are currently only listed in China."(1)

(1)

"BYKSKI for THREADRIPPER"

videocardz c o m/newz/bykski-showcases-first-waterblock-for-ryzen-threadripper

August 7, 2017 | 08:45 PM - Posted by RectangularNotRound (not verified)

G-Damn Spam filter blocking legit websites!

August 8, 2017 | 01:44 AM - Posted by YourName (not verified)

I'd say on the entire platform proposition that AMD has the better offer.

August 8, 2017 | 02:25 AM - Posted by Cyric (not verified)

Now lets see how real those turbo clocks are. 4.4Ghz from a base of 2.9 will be unable to be sustained for long. If the opposite where the case I would expect to see 3.4-3.8 base.

I believe that those frequencies are just to hit AMD's clocks and nothing more. I expect inability for Intel to maintain the advertised clocks in sustained load (appropriate to each TBoost scenario)

August 8, 2017 | 04:42 AM - Posted by Spunjji

I'm with you on that. Given how wildly optimistic the TDP rating of their 10-core CPU is and its behaviour in reality, I have absolutely no faith whatsoever that the 18-core model will hit that boost clock in anything other than specific benchmarks.

What it may do is allow Intel to score some relative victories in gaming performance vs. Threadripper at everybody's favourite 1080p resolution.

Why anybody spending that much on a CPU would game at 1080p, however, is utterly beyond me.

August 8, 2017 | 02:45 AM - Posted by Dormient (not verified)

It will be hilarious to see the temp and then the throttling on 1980x... Smart ones will buy AMD this gen. Too many advantages compared to intel:
1. Price/performance ratio
2. Better TIM quality
3. Soldered IHS
4. Temperature
5. Throttling issue
6. Power consumption
7. PCIe Lanes count

August 8, 2017 | 02:46 AM - Posted by Dormient (not verified)

Edit: 7980x. Somehow I keep typing 1980x anywhere...

August 8, 2017 | 05:26 AM - Posted by Anonymous_too (not verified)

All core turbo frequencies from Anandtech via PCGamer

http://images.anandtech.com/doci/11698/turbos.png

NB! These are **NOT** guaranteed frequencies, when all cores are stressed 100% continuously. Real world frequencies will be thermally / power limited.

August 8, 2017 | 09:17 AM - Posted by Benjamins (not verified)

According to engadget the 7960X gets 3200 in Cinebench, so for $700 more you can go from 3080 from the 1950X to 3200. This also means that the 7940X will be slower at $400 more.

August 8, 2017 | 10:11 AM - Posted by okijfowier (not verified)

Like how intel can manage 16/32 on small for factor were AMD essentially slapped together 2 8 core CPUs.

August 8, 2017 | 12:32 PM - Posted by DaveSaysThisMAN (not verified)

Not slapped together, Man! That's a Modular/Scalable design with the smaller 8 core(2 CCX-Unit) Zeppelin die/wafer yields at above 80% coming off those fab lines. With AMD creating the Ryzen(7,5,3), Threadripper, and the Epyc SKUs from the very same Zeppelin dies scaled up in a modular fashon via the Infinity Fabric for some vastly more affordable CPU SKUs.

And even Intel's New Mesh Interface has latency trade-offs in order to scale to higher core counts in even those big Intel monolithic die designs. AMD has those 8 core Zeppelin dies more thermally/physically seperated on the TR/Epyc MCM so that means that some overclocked cores on one Zeppelin die will not have the cores' thermal heat load direcly migrate via silicon transfer to any other seperate dies' cores. Intel will be more limited in dealing with highely overclocked cores as all the cores on Intel's SKUs share the same monolithic block of silicon.

AMD can treat for thermal conductance purposes each Zeppelin die as a seperate thermal domain as the Zeppelin dies are actually seperate dies physically as far as thermal conductance is concerned, with the Infinity Fabric tying those seperate Zeppelin dies together as one logical processor with up to 32 cores across 4 Zeppelin dies.

AMD has gone all Modular, Man!

August 8, 2017 | 12:42 PM - Posted by DaveSaysThisMAN (not verified)

Oh Man, Here comes the HEAT! Let's split up and get away from the HEAT before we get thremally busted, Man!

August 10, 2017 | 12:39 PM - Posted by Alansmithee (not verified)

The PCIe lanes in the comparison chart are wrong. AMD specs the gross number of lanes while Intel specs the net number of lanes after DMI. In other words, either Intel has 48 and TR has 64, or Intel has 44 and TR has 60. It's incorrect to say Intel has 44 and TR has 64.

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.