Intel Launches 16 and 18-Core Core i9 Desktop Chips To Take On AMD Threadripper (hothardware.com)

MojoKid writes: Intel has officially launched its Skylake-X processor offering in response to AMD's Ryzen Threadripper series of desktop CPUs. The new Core i9-7980XE and Core i9-7960X are 18 and 16-core configurations respectively, with 2.6GHz and 2.8GHz base clocks and 4.4GHz max boost clocks. Both chips support Intel HyperThreading, with 36 threads of processing for the 7980XE and 32 for the 7960X, while both also have 44 lanes of PCI Express connectivity and support for DDR4-2666MHz memory. Both chips also utilize Intel's X299 chipset platform and are LGA 2066 socket compatible. The Core i9-7980XE has 24.75MB of shared L3 cache, 1MB of L2 cache per core, and a TDP of 165W. The Core i9-7960X's details are essentially same, though two processor cores and the cache associated with them have been lopped off. The Core i9-7960X has a couple of advantages, however, in that its base clock is 200MHz higher than the flagship Core i9-7980XE and it has higher all-core frequency boost to 3.6GHz, while the 7908XE tops out at 3.4GHz on all cores. The new chips are multi-threaded beasts in the benchmarks, posting the highest scores seen to date in heavily threaded workloads. They also offer strong single-threaded performance that outpaces AMD's Ryzen processors. Power consumption is surprisingly good as well and only marginally higher than the 10-core Core i9-7900X. However, at $1999 for the Core i9-7980XE and $1699 for the Core i9-7960X, as usual with Intel high-end chips, they're certainly not cheap.

Comments

Too little too late

[xxxJonBoyxxx]

Sorry Intel, the new AMD procs offer great performance for the money. No reason to go Intel for at least a generation.

Re:Too little too late

[AmiMoJo]

There are other advantages to AMD parts too. More PCIe lanes, better chipsets especially if you want to do virtualization with IOMMU pass-through, longer expected lifespan of the socket and motherboard, and a less terrible secret backdoor system.

Re: This sounds perfect for Rust!

[nsuccorso]

Imagine how fast this Rust fanbot will run, as long as itâ(TM)s written in Rust! With this new processor family, the Rust bot will post up to 4 million odes to Rust per Slashdot story, per second! Youâ(TM)ll see no Rust gathering on this processor! Rust!!

Hooray!

[fuzzyfuzzyfungus]

So, it's a Xeon that can't do ECC. Seems totally worth it.

Re:Hooray!

[fuzzyfuzzyfungus]

I'm not privy to Intel's cryptic market segmentation schemes; but I'd imagine that it is because they now have competition. As long as AMD was basically irrelevant; "High end" could pretty much mean what Intel wanted it to: either the point on the i7 price performance curve where 'price' really starts to overshadow 'performance' or the end of the Xeon range with low core counts somewhat limited cache and total system RAM support; at their preference.

Now that AMD has some actually interesting parts again; Intel has less freedom to just call things "High end". Because AMD still lags on single threaded performance, they do have an "is a really fast i7 for extra money" option(the i7-7740x, only 4 cores and limited RAM support; but very, very, aggressive clock speed) and the slightly baffling i57640x(pay a nontrivial premium for an LGA2066 motherboard why exactly?); but because AMD is dishing out the core counts, PCIe lanes, and RAM capacities; they also have the i9s; which are painfully expensive by desktop standards; but look like awfully tempting budget Xeons unless something can be done about that.

ECC makes a pretty good 'something'. Lasering off virtulization support would go quite badly(making them effectively useless to anyone who spins up even the occasional VM, which is a fair number of people who buy $1000+ CPUs; but probably also being a massive bargain for people who don't plan to virtualize their workstation or server workloads, which is still a lot of units). Disabling AMT wouldn't be effective enough: gamer/enthusiast types wouldn't care; but neither would a lot of workstation or server customers(either all your computers are in one place, so fancy remote access tools aren't interesting; or you can just add an AST2400 or something if Intel tries to charge too much). Gimping core counts or PCIe lanes isn't an option because AMD; which pretty much leaves ECC.

Re:Hooray!

[yorgasor]

Except AMD Threadripper offers ECC support as well. In the olden days, Intel was chimping on the consumer and low server end, trying to keep the Xeons from beating the Itaniums, and that was what led AMD to jump in with native 64bit support and dual core chips and start eating Intel's lunch.

Re:Hooray!

[Anonymous Coward]

The advantage of ECC RAM is NOT about "oh, the system has been up for 2 years without a glitch, see, ECC RAM was a waste of time".

The whole point is that you you can see that there were no errors.

Just finished a 2 month compute task? No memory errors? Great, the code worked and any abnormal results are due to coding mistakes. Otherwise, you might be tempted to run the whole thing again on the off-chance it might have been a memory error.

I've had rigs up for > 5 years without a single ECC error, but I do not for one second believe ECC to be a waste of money.

Re:Hooray!

[Lord Crc]

The advantage of ECC RAM is NOT about "oh, the system has been up for 2 years without a glitch, see, ECC RAM was a waste of time".

The whole point is that you you can see that there were no errors.

Indeed. I just had to reformat an Intel NUC as the filesystem had been completely trashed. Firefox had been a bit unstable on it, but well, that didn't trigger any warnings on my end. I also got one or two weird errors when running apt, but I was running unstable so again I didn't think to much of it...

Then after a couple of months it stopped booting. Turned out the filesystem was bonkers, due to silent corruption.

Ran memtest86, and one small but sequential range of addresses failed for every test.

165W

[ChunderDownunder]

That's way hotter than the worst Pentium IV.

"Power consumption is surprisingly good". I wouldn't like to see the power bill at the end of the month and what sort of passive cooling is used to achieve a quiet workspace?

I can see why the review website is called Hot Hardware.

Re:165W

[Mad Merlin]

The Pentium 4 was also only a single core (or dual core if you count the short lived Pentium D line, which was based on Pentium 4 cores glued together) with vastly worse IPC. These have up to 18 cores and will chew through multithreaded workloads something like 50x faster than the fastest Pentium 4, while using comfortably less than double the power at full tilt (the i9-7980XE uses more than it's TDP at full load, around 195W, but recall that the Pentium 4 had a 115W TDP). Also, the idle and lightly loaded power usage on the i9-7980XE is dramatically lower than any Pentium 4. Unless you're running the CPU in part of a render farm that works 24/7 or something, it's going to be drastically quieter and use less power overall than any Pentium 4, not to mention absolutely obliterate it on performance.

Having said that, I'd still opt for a Threadripper over one of these if I was building a computer today. The extra raw performance from the i9-7980XE over the 1950X (between low single digit %s to around 40%, depending on the test) is nice, but when you consider the weaker X299 platform (fewer PCIe lanes, no ECC) and the toothpaste under the IHS in the Intel (meaning it's impossible to properly cool it without delidding and potentially destroying the CPU, not to mention the runaway power usage even with a mild overclock) for double the cost, Threadripper seems like the obvious pick.

Re:165W

[drinkypoo]

Don't forget the spectacular expense of the X299 platform. As usual, you'll wind up spending an extra hundred bucks on a decent motherboard for Intel as opposed to AMD.

I'm still running an FX-8350 on a G1 Gaming, and it's spectacular for what it is. Rock-solid, awesome price:performance for when I built it. A new AMD system would be literally twice as fast, though.

Threadripper has more pci-e and $700 less

[Joe_Dragon]

Threadripper has more pci-e and $700 less

Re: Threadripper has more pci-e and $700 less

[HumanWiki]

What idiotic nonsense. I know plenty of people that own cats up in to the 7 figure range and drive themselves because they enjoy the experience of driving an exotic, super or hypercar.

44 lanes!

[darkain]

44 lanes of PCIe? That is only four more lanes than my first generation Xeon E5 workstation that I'm using right now, which is only a quad-core chip. Granted though, this machine is pretty maxed out with that tho, even with just one GPU, thanks to PCIe SSD, dedicated sound card, and 10gbe networking.

Re:44 lanes!

[AmiMoJo]

PCIe lane starvation is a big problem now, and Intel is failing to solve it.

Now we have NVMe SSDs all wanting 4 lanes, and USB 3.1 needs a couple of lanes per port to give maximum bandwidth, standard desktop CPUs don't have enough for today's needs, let along room for expansion. If you want that system to last 5+ years you want some free lanes to add expansion cards later.

Threadripper finally delivers plenty of PCIe lanes. And ECC RAM support. And a software TPM. And it's a lot cheaper. And it runs cooler.

Re:This sounds perfect for Rust!

[Anonymous Coward]

I think we will really see Rust shine

Isn't this an oxymoron?

I'm not trolling. Just punning.

Re:Any faster?

[UnknownSoldier]

Depends on what you are doing. Is single-threaded? Multi-threaded? Needs ECC, Memory bound? Computer bound? etc.

For gaming? Nope. An i7-7700K is still faster then the iCore 9. LOL.

For rendering? Yep 2:07 vs 7:19

However, it is important to point out:

* Intel CPU's are the fastest CPU's around but you literally PAY through the nose 2x- 5x for a measly ~10% increase in performance.
* AMD's Ryzen and Threadripper CPUs have WAY better bang/buck. i.e. In Blender 2.78c AMD's Theadripper 1950X ($999) is faster then the Xeon E5 2699 v4 ($4,115). LOL.

Conclusion:

Is the i9-7960X worth a whopping $700 more then Threadripper 1950X?

For many people, the answer is no. That is money that could be spent into a better GPU or more RAM/SSD.

What CPU is "best" depends on your workload.

AMD - Job well done

[Lady Galadriel]

Competition is a good thing. Without AMD, and all the people who helped keep them alive, Intel would have sat on things like this.

In the long run, Intel parts are likely going to be better. They have more money, more chip desigers, and un-fortunantly more customers, (some that will pay outragous prices).

ECC

[nojayuk]

It's about time high-end and even commodity PCs moved to supporting ECC and using it as common practice. When total RAM fit in a PC was 1 or 2GB and data rates were a few hundred MB per second at best a bit error rate of 1 per trillion reads or writes was acceptable. Now that common motherboards can accept 64GB and more and RAM access speeds have also escalated the chances of a problematic bit error occurring in code or data have shot up, especially as the RAM's die mask sizes have decreased.

As an aside I've used commodity mobos in the past that accepted and would run ECC RAM but only as regular memory, the chipset didn't implement any kind of error-correcting capability. My workhorse machines for bit-bashing all have ECC properly implemented on server-grade hardware.

Re: Wow. $$$

[Zero__Kelvin]

More like 32 or 36 blowjobs simultaneously... on film.