Friday, October 4th 2019
Intel 10th Gen Core X "Cascade Lake-X" Pricing and Specs Detailed
Ahead of their October 7th product launch and November availability, we have confirmation of the specifications and pricing of Intel's 10th generation Core X "Cascade Lake-X" HEDT processors in the LGA2066 package. These chips feature compatibility with existing socket LGA2066 motherboards with a UEFI BIOS update, although several motherboard manufacturers are launching new products with some of the latest connectivity options, such as 2.5 GbE wired Ethernet, and 802.11ax Wi-Fi 6 WLAN.
The 10th generation Core X HEDT processor family is based on the new 14 nm++ "Cascade Lake" silicon, which comes with hardware fixes against several classes side-channel vulnerabilities, and introduces an updated instruction-set that includes more AVX-512 instructions, and the new DLBoost instruction. DLBoost leverages new fixed-function hardware on silicon to accelerate AI deep-learning neural-set building and training by up to 5 times. Intel's first wave of 10th gen Core X lineup is rather slim, with just four processor models. The company did away with the Core i7 brand extension, as core-counts in the mainstream desktop segment have already reached 8-core. The lineup now begins at 10-core/20-thread, with the chip's full 48-lane PCI-Express and 4-channel DDR4 interfaces enabled across the board. All models feature the "XE" brand extension, and feature unlocked base-clock multipliers.The Core i9-10900XE is your gateway to the series. This 10-core/20-thread chip comes with a fascinating price-tag of just USD $590, a significant drop from the $999 price for the previous-generation 10-core chip, the i9-9900X. It's clocked higher, with 3.70 GHz nominal, 4.50 GHz Turbo Boost 2.0, 4.70 GHz Turbo Boost Max 3.0 and 4.30 GHz all-core Turbo. The chip is endowed with 1 MB of dedicated L2 cache per core, and 19.25 MB of shared L3 cache.
The Core i9-10920XE is a $689 12-core/24-thread chip priced under AMD's upcoming flagship AM4 model, the Ryzen 9 3950X. It's marginally faster than its predecessor, the i9-9920X, with 3.50 GHz base clocks (same), 4.60 GHz Turbo Boost 2.0, 4.80 GHz Turbo Boost Max 3.0, and 4.30 GHz all-core turbo. Interestingly, the increase in core-count doesn't bring additional L3 cache, you get the same 19.25 MB.
The next step in this series is the $784 Core i9-10940XE, a 14-core/28-thread processor clocked at 3.30 GHz, with 4.60 GHz Turbo Boost 2.0, 4.80 GHz Turbo Boost Max 3.0, and 4.10 GHz all-core turbo. Yet again, you get just 19.25 MB of shared L3 cache. Interestingly, Intel did not plan a 16-core/32-thread model in this series, you jump straight to the flagship.
Leading the pack is the Core i9-10980XE, an 18-core/36-thread processor priced at a mouth-watering $979, which is less than half that of the previous-generation Core i9-9980XE. It ticks at 3.00 GHz, with 4.60 GHz Turbo Boost 2.0, 4.80 GHz Turbo Boost Max 3.0, and 3.80 GHz all-core turbo. You get a larger 24.75 MB of shared L3 cache. All four chips have their TDP rated at 165 W.
The 10th generation Core X HEDT processor family is based on the new 14 nm++ "Cascade Lake" silicon, which comes with hardware fixes against several classes side-channel vulnerabilities, and introduces an updated instruction-set that includes more AVX-512 instructions, and the new DLBoost instruction. DLBoost leverages new fixed-function hardware on silicon to accelerate AI deep-learning neural-set building and training by up to 5 times. Intel's first wave of 10th gen Core X lineup is rather slim, with just four processor models. The company did away with the Core i7 brand extension, as core-counts in the mainstream desktop segment have already reached 8-core. The lineup now begins at 10-core/20-thread, with the chip's full 48-lane PCI-Express and 4-channel DDR4 interfaces enabled across the board. All models feature the "XE" brand extension, and feature unlocked base-clock multipliers.The Core i9-10900XE is your gateway to the series. This 10-core/20-thread chip comes with a fascinating price-tag of just USD $590, a significant drop from the $999 price for the previous-generation 10-core chip, the i9-9900X. It's clocked higher, with 3.70 GHz nominal, 4.50 GHz Turbo Boost 2.0, 4.70 GHz Turbo Boost Max 3.0 and 4.30 GHz all-core Turbo. The chip is endowed with 1 MB of dedicated L2 cache per core, and 19.25 MB of shared L3 cache.
The Core i9-10920XE is a $689 12-core/24-thread chip priced under AMD's upcoming flagship AM4 model, the Ryzen 9 3950X. It's marginally faster than its predecessor, the i9-9920X, with 3.50 GHz base clocks (same), 4.60 GHz Turbo Boost 2.0, 4.80 GHz Turbo Boost Max 3.0, and 4.30 GHz all-core turbo. Interestingly, the increase in core-count doesn't bring additional L3 cache, you get the same 19.25 MB.
The next step in this series is the $784 Core i9-10940XE, a 14-core/28-thread processor clocked at 3.30 GHz, with 4.60 GHz Turbo Boost 2.0, 4.80 GHz Turbo Boost Max 3.0, and 4.10 GHz all-core turbo. Yet again, you get just 19.25 MB of shared L3 cache. Interestingly, Intel did not plan a 16-core/32-thread model in this series, you jump straight to the flagship.
Leading the pack is the Core i9-10980XE, an 18-core/36-thread processor priced at a mouth-watering $979, which is less than half that of the previous-generation Core i9-9980XE. It ticks at 3.00 GHz, with 4.60 GHz Turbo Boost 2.0, 4.80 GHz Turbo Boost Max 3.0, and 3.80 GHz all-core turbo. You get a larger 24.75 MB of shared L3 cache. All four chips have their TDP rated at 165 W.
124 Comments on Intel 10th Gen Core X "Cascade Lake-X" Pricing and Specs Detailed
Healthy competition is a good thing. Please, everyone, don't start foaming at the mouth because AMD is not the ONLY choice. That's how you make a new monopoly.
Using old inefficient 14nm when their competitor is on 7nm and has higher IPC is also a bad look. Imagine how hot and power hungry Intel's hypothetical 14nm equivalent of the 24-core Threadripper would be. That's right it would literally catch fire!
www.tomshardware.com/news/amd-ryzen-9-3950x-overclocking-watercooling-4.3-ghz,40555.html
As for MSDT, Intel is trying to push 14nm to tide them over to 10nm, but Coffee Lake is verging on joining the ranks of Prescott and Fermi from the amount of heat and power draw it suffers. As for 10nm, we all know of Intel's fab problems. It's one thing to hang by a 14nm thread until they can bring out a Conroe-level breakthrough, but do they have anything nearly that revolutionary?
Betting on Ryzen's weaknesses such as cache speeds and DRAM latency can only last so long. Yes, AMD will be saddled with it until they move on from Ryzen, but as the 3000's prefetchers' improvements have shown, they can do just fine without the theoretical DRAM performance. Look what Ryzen 3000 has done; they've already beat Coffee Lake on IPC.
I've always been an Intel guy, and would like to return to Intel someday because I believe they create the most solid firmware and platforms. I really hope I can see that day again.
But yeah, they have tons of work. Their turbo is inferior to AMD's XFR and other tech, power draw is only going up, chips are too complex... And they even had to adjust ring bus to mesh and now use both out of necessity on different product lines. Meanwhile, Zen is the same, top to bottom, completely scalable and even those CCX's help seed the entire product stack. That's my take and prediction as well. Intel is trying to find a way to not lose face and skip as much of 10nm as possible. Roadmaps were already adjusted, but we'll see another round of that. If you see what their current crop of 10nm can do, its nothing earth shattering in any way shape or form. And their other announced architectures... when a CPU engineer no longer talks about raw performance but rather about IGPs and optimizing for 'Creators' and all, you know they haven't got a thing to offer.
If you just enjoy having 16 cores without the need to use it for production, by all means go for it.
TR3 is the real competition to Intel's HEDT feature wise, which it will for sure slaughter Intel's lineup if AMD price them right. 24C48T for $999 would be nice.
But as you are saying, some applications may benefit from even more, but this would require much larger dies. If so, I think there are better uses for die space. For Skylake-SP/X, the HCC die was ~485 mm², I assume the Cascade Lake-SP/X die is comparable.
So this is smaller than the TU104 die used in RTX 2070 Super etc., but this is of course not an apple-to-apple comparison.
Skylake-SP/X was plagued with being produced on a lower volume 14nm+ production line, which has contributed to the high price and low availability. Cascade Lake-SP/X was according to roadmaps supposed to launch one year ago, but I assume it's the 10nm problems causing laptop and desktop parts to eat up much of the 14nm++ capacity that has caused the delay. That would sure make an attractive product.
But always keep in mind that workstations are built for a specific purpose in mind, so if the buyer are going to have a heavy AVX load or similar, the real world performance could just as easily tilt the other way.
Would be nice if Zen2 had AVX512 support
Source: AnandTech
Source: STH
Shamelessly copy ? Ever seen Intel Core 2 Exteme QX6850 from 2007, Kentsfield and even in 2010 with Clarksdae. Guess not lol, And hit Anandtech piece on this same processor and see what AMD called Intel as fake LOL during Phenom days.
AMD Ryzen CPUs use MCM because their WSA and R&D budget and how cheap it is (Intel also did the MCM due to same reason) vs Monolithic Nonpureplay Foundry like Intel. Foveros is coming and also EMIB came into KBL G CPU as well. This is not new thing lmao and how HBM failed hard and fell flat for Consumer (Ofc Vega/GCN needed it to scale and we know RVII EOL)
Intel stagnated due to no competition and they milked the whole world, Still do (CSL X Server a.k.a Cooper Lake glued server processor) and their 10nm went too aggressive and also BK as CEO failed. And their Knights Landing also failed but that socket lives !
Zen2 has DRAM latency but their high Cache and Intel IPC catchup helped them with their brilliant Ryzen 3000 series. And 14nm or not it beats 7nm still with what ? A 4 year old microarch, Zen+ was Haswell and what happened with 8C/16T 2700 vs a 7700K (Massacre happened). Turbo is inferior ? What ? XFR is completely pushing the edge of thay stupid Low power 7nm TSMC node (Made for Apple since they arr primary for CapEx funding and ARM) thank god Ngreedia went with HPP 7nm Samsung node and not suffer from same crap off inferior 7nm node. So yeah back to Turbo, Intel PL2 set to max. There you go Max clocks on all cores without any Junk of PPT, PBO, XFR2 marketing drama (Watch GN on this if you did not).
Intel processors also are made to be scalabale if you are not under a rock LCC, HCC (HEDT), XCC (Xeon) until SKL X because they switched HEDT to XCC due to AMD closing in as their 10nm failures, since their uArch is coupled with node. So yeah the uArch scales here while Cores are not scaled like AMD.
AMD recycles their cores off this MCM due to WSA as it helps keeping budget off CCD at 7nm and I/O chip of decoupling Northbridge at 14nm GoFlo (EPYC Rome, where as 12nm for Ryzen Matisse)
And finally Intel is still raking profits due to DC arena where adoption rate is slower when you bring in a new competetor esp given how NUMA Ryzen 2000 to non NUMA Ryzen 3000 uArchs came. While Monolithic Intel still didn't had to change the programming much.
Iris Plus IGP has advantage over existing AMD APUs consider it getting tougher only not simple when you count Intel's massive R&D. And finally Apple AirPods profit is higher than AMD at gross $5Billion.
To conclude consumer wins when competition arrives like this (9980XE ST and Gaming was good and comparable to 9900K, 10K CSL X will improve at the cost of power but if people buy what to lose, more lanes more options better pricing all win) and shakes up the incumbent. So stop frothing over corporates :)
We can safely assume TSMC designed their 7nm for mobile SoCs (over 90% of TSMC's sales).
So who knows? If AMD could influence TSMC in any way (by capital or as a major client) maybe these CPUs would boost higher. And some applications favor low latency (ideally: ringbus). CPUs have different architectures and they excel in different tasks. And every architectural choice has a cost.
Let's cherish the fact that CPUs differ in something other than core count and frequency. :)
There's really no reason why Intel would make CPUs more like Zen (since work perfectly well). And there's no reason for AMD to make CPUs exactly like *Lake, because their only advantage would be price. This TDP is as real as it gets. It's just a number. And Intel CPUs will behave accordingly by default.
Looking at Intel's earlier LGA2066 stuff, there's a good chance 10 and 12-core CPUs are within those 165W at full blast (sans AVX-512). AMD and Intel plan their lineup differently.
In the AMD world EPYC is solely a server lineup. Ryzen lineup (including TR) spans both consumer and pro use.
So for an 8-core or 24-core workstation you're expected to buy a Ryzen 7 or Ryzen TR accordingly.
Intel is not marketing HEDT towards professional use (not for production systems anyway). There's no official ECC support. And they aren't put in big OEM workstations.
Intel has a large Xeon lineup that is meant to cover all production scenarios.
On the other hand, Intel doesn't have a separate "consumer" and "pro" variants - many CPUs are vPro-eligible. And they offer IGP in multiple CPUs - including Xeons.
A new wave of speculative security holes in Intel incoming...
Hence, this is a very good example of how important AVX-512 is.