Intel's HEDT Conundrum

The High-End Desktop (HEDT) market has perhaps moved faster than any other segment in the PC industry, just two years ago Intel launched the 6950X, a 10 core/20 thread CPU with a base frequency of 3.0GHz and 3.5GHz boost frequency on the X99 platform for an eye watering $1700.

Just a year later with Skylake X, Intel released the 7900X, a 10 core/20 thread CPU with a base frequency of 3.30GHz and a boost frequency of 4.30GHz on the X299 platform for a more modest $999. 

Well here we are just over a year later, with AMD just hours ago launching the 2990WX, a 32 core/64 thread CPU with a base frequency of 3.0GHz, boost frequency of 4.20GHz on the X399/TR4 platform for $1799.

The core war is in full effect, and even though Intel could offer more models at varying price points, with higher clock speeds, higher IPC, higher memory speeds, lower latencies and just in-general better all-round products; they aren't, at least not yet.

I don't know why specifically Intel is unable to see the obvious, as someone who follows the HEDT market quite closely, it seems so blazingly simple how they can win back this market. In this piece, I'd like to explain what I think Intel should do to better engage with the HEDT market, because while small, it's a very diverse market that needs to be kept happy, as HEDT users are very influential in the DIY PC-ecosphere.

Before we dive in, let's take a brief look back at Intel's history in the HEDT market, post Broadwell-E; as before then AMD had nothing competing in this space. 

If memory serves correctly, around Q3 2016 we started getting rumours of a successor to Intel's X99 platform, original rumours stated that X299 would offer support for Optane storage, CPU-RAID and would offer up-to 10 cores, the exact same core count as the Broadwell-E 6950X, but, based on the Skylake-X architecture and the improved 14nm+ node.

A few months passed, and we started to get more information and leaks regarding AMD's upcoming Ryzen CPUs based on the newly designed Zen architecture, which would offer up-to 8 cores and 16 threads for the mainstream market, performance numbers were looking quite good. Around this time we also started getting reports that X299 would offer up-to 12 cores, a slight increase over the 10 core Broadwell-E part. 

At Computex 2017 Intel officially announce X299 and Skylake-X, with core counts ranging from 4, all the way to 18 cores, although the HCC variants, 12 - 18 cores had undefined clock speeds, TDPs, PCIE lane count and even memory support, indicating they were added last minute. While we'll likely never know for sure what Intel's original plan with X299 was, it seems to me they originally only planned on 10 cores being the highest model, with the news of Ryzen, they upped it to 12 cores, and when word of Threadripper started leaking, they upped the core count again to 18, encompassing the fully enabled High Core Count (HCC) die. 

A previously leaked roadmap (leaked in December 2017) had implied that Skylake-X would be replaced with Cascade Lake-X in Q3 2018, while not much is known about Cascade Lake, it was presumed it would be based on Intel's 14nm++ node, with higher clocks and possibly a soldered IHS to better aid cooling and overclocking, as one of the biggest complaints with Kaby Lake-X and Skylake-X was the use of thermal paste, instead of traditional indium solder (which was used on previous Intel HEDT lineups)

- The December Roadmap (courtesy of BenchLife and WCCFTech)

In April 2018, Kyle Bennett from HardOCP released an article stating that in Q3/Q4 2018, we're likely only going to see soldered Skylake-X refresh CPUs, with no other architectural changes, these CPUs will offer higher clock speeds and higher TDPs.

At Computex 2018 (June) Intel announced a 28 core HEDT part coming in Q4 2018 (presumably another knee-jerk reaction to the (at the time) rumoured 32 core TR Gen2 part), no other SKUs, pricing or platforms were announced, just that in Q4 2018, Intel would be launching a 28 core part. Myself and many others in the media assumed this would be Cascade Lake, based on the 14nm++ node. This part was shown using Intel's enterprise grade LGA3647 socket with six memory channels.

However just days later, it was reported by PC Watch that Cascade Lake was delayed until sometime in 2019, and instead Intel will be refreshing X299 (Basin Falls) with up-to 22 core / 44 thread part.

So, what is going on, is the new 28 core part in Q4 going to be Skylake-X or Cascade Lake-X? Why is Intel releasing a 22 core for X299, presumably that would be the XCC die, so why stop there, why not release the fully fledged 28 core die, why (presumably) limit it to the LGA3647 socket only? Did Intel create a die in-between the HCC and XCC die solely for this refresh?

There are more questions than answers, so what is going on?

Unfortunately there are still more questions than answers, however a newer leaked roadmap from last week may help shed some light on Intel's fragmented and outright confusing strategy.

Here's the roadmap (courtesy of WCCFTech)

So, things to note.

- Cascade Lake-X has been pushed back from Q4 2018 to Q3 2019, with a Skylake-X/Basin Falls refresh on the X299 chipset landing in Q4 2018.

- When Cascade Lake-X does land, it will be on the Glacier Falls platform -- no TDP or core counts announced.

- Intel has added an 'A' series lineup above the 'X' series of CPUs, indicating there will be two HEDT lineups from Intel.

- The 28 core HEDT part will be landing in Q4 2018, but it appears it will be based on Skylake-X (14nm+) and not Cascade Lake-X (14nm++) -- it's also peculiar how only 28 core is mentioned

Now, many may point to the fact this roadmap could be subject to change, or completely fake. However myself and a friend both published this on Twitter and had them DMCA'd. Indicating that these roadmaps are correct.

Anyway, as you can see, this is a complete mess that lacks any sort of direction or enthusiasm. I strongly believe Intel need to stop reacting to what AMD is doing and instead start being pro-active with their product lines and launches.

Anyway, onto what I think Intel should do to make their 'X' and 'A' series HEDT lineups a success.

It's so simple, it can be bullet pointed.

The 'X' series Basin Falls Refresh

Core Counts: 10 cores ($599) | 12 cores ($699) | 14 cores ($799) | 16 cores ($899) | 18 cores ($999) avg ~$57 per core

There is little point including 6 and 8 core options anymore, the i7-8700K (6c/12t) and upcoming i9-9900K (8c/16t) will negate those offerings.

Every SKU has the full 44 PCIE Gen3 lanes and Quad Channel Memory Support

Soldered IHS, higher TDPs/Clock speeds while maintaining full compatibility with existing X299 motherboards.

The 'A' series HEDT platform

Core Counts: 20 cores ($1399) | 22 cores ($1599) | 24 cores ($1699) | 26 cores ($1799) | 28 cores ($1999) avg ~$70 per core

While you could include lower end Low Core Count (LCC) and High Core Count (HCC) options and just give them more memory channels and PCIE lanes, it would further fragment the platform and cause unnecessary disparity between the X and A series lineups, having a 16 core with 44 PCIE lanes/Quad Channel on LGA2066 and another 16 core with 48 PCIE lanes/Hex Channel on LGA3647 -- as an example.

Every SKU has the full 48 PCIE Gen3 lanes and Hex Channel Memory Support.

Soldered IHS.

Being that this platform would be based on Intel's LGA3647 socket, new motherboards would be required, and higher TDPs/Clock Speeds could be explored.

Intel could go even cheaper if they wanted, it's entirely up-to them. However, don't you think this would make for a simplified, logical product stack that would compete well against AMD's 2nd Gen TR CPUs? I certainly do, not only is it simple, 5 models for each product lineup, but there are logical differences between the X and A series platforms.

The 'X' series platform encompasses LCC and HCC parts with Quad Channel support and 44 PCIE lanes, while the 'A' series would encompass only the XCC die parts, bringing the full 48 PCIE lanes and Hex Channel support. They could even add ECC support to the 'A' lineup if they want, or keep it disabled if they're worried it would damage Xeon sales.

The 'A' series parts would compete nicely against AMD's WX Threadripper parts (2970WX and 2990WX) while the 'X' series would compete against AMD's X Threadripper parts (2920X and 2950X), not only would Intel have an IPC advantage, clock speed advantage, latency advantage, but they'd also be able to offer more models at varying price points, and as I said before, the HEDT market is diverse - such a simplified product stack would be welcomed by all I feel.

Anyway, that's my semi-rant/opinion piece over, I don't have much faith Intel will follow anything close to what I've laid out here, but if I was managing Intel's HEDT product lineup, this is what I would proudly push.

Thank you for reading.

My Twitter