Thoughts on Linus' i9, X299 video?

[joshyboy82]

He makes great points, but will it have any effect? Does he have enough clout in the industry? That's why I'm asking it here, I listened to all of his points, and they seemed completely valid, especially the "Make the best chip you can, for the best price you can sell it, not preempt AMD's Threadripper with your own, so you don't get caught off-guard." Since Tom's also has, what I would perceive to be comparable influence (I could be way off base, Tom's employee's have moved to high level jobs in the manufacture industry) I'm curious if Tom's views Linus as off-base, or right on target, succinctly said by Linus.

 

[TechyInAZ]

They are very valid, I've been saying the same thing about Intel for the past 5 years.

It would take a LOT to change Intel, AMD is the only one that has enough firepower to actually make Intel get off it's chair. The community could too, but in all honestly I'm not sure how much.

 

[rhysiam]

He's right on and I don't think many people would find much, if anything, about what he said particularly controversial.

Kaby Lake X is utterly nonsensical.
Introducing the HCC silicon onto the HEDT platform is an interesting move. But the fact is that Intel is going to have to scramble to validate it.

The only thing I question is that Linus says Intel are deliberately waiting on the details for the HCC chips (14-18 core variants) because they want to see what AMD are going to do. I don't think that's right at all personally. The HCC silicone was original intended to be a Xeon only part (as has always been the case in the past), and now they're scrambling to get it released in a consumer variant. We don't have details - I suspect - because Intel is still working them out.

Will they be able to get Turbo Boost 3 working on it?
How high can they clock a 14-18 core part without completely losing control of TDP?
In the past Turbo Boost on HCC server parts was never important because anyone buying a HCC CPU has highly threaded workloads, so whether or not they could boost well over 4Ghz was irrelevant... that's not the case on a HEDT.
Last - but definitely not least - how is the dual-ring bus going to affect core-to-core latency? People are rightly nervous about how Threadripper is going to perform with 4 CCXs spread over 2 dies. But Intel's HCC silicone has similar core-to-core bandwidth and latency issues.

The bottom line is Intel could have been upping core counts for years, but they choose to keep shrinking die sizes and pocketing the higher margins. For comparison, die size estimates put the $1700 6950X at just slightly larger than the i5 2300 which launched for ~$187 in Q1 2011. That's the only reason AMD can actually compete at all, because Intel have been deliberately strolling for years. People will say that's good business for Intel - and maybe they're right, I don't know - but it shows just how important competition is to push progress.

 

[Seanie280672]

He makes some very good and valid points, I had to laugh..........Kaby Lake X only 16 PCI-e Lanes and only dual channel memory support on a quad board, so basically a high end board $300-$600 with multiple PCI-e x16 slots and up to 5x M.2 slots can actually only support......ummm 1 Graphics card and no nvme drives unless you can get one to run of the chipset, oh and dual channel ram, well done Intel, a round of applause for you muppets

Anyone else who wants to watch the video, heres the link: [video="https://www.youtube.com/watch?v=TWFzWRoVNnE"]https://www.youtube.com/watch?v=TWFzWRoVNnE[/video]

Mod Edit: Added video tag to link

 

[joshyboy82]

He makes some very good and valid points, I had to laugh..........Kaby Lake X only 16 PCI-e Lanes and only dual channel memory support on a quad board, so basically a high end board $300-$600 with multiple PCI-e x16 slots and up to 5x M.2 slots can actually only support......ummm 1 Graphics card and no nvme drives unless you can get one to run of the chipset, oh and dual channel ram, well done Intel, a round of applause for you muppets

Anyone else who wants to watch the video, heres the link: https://www.youtube.com/watch?v=TWFzWRoVNnE

Thank you for posting the video, I forgot. I guess I could edit it in. I won't though.

 

[joshyboy82]

He's right on and I don't think many people would find much, if anything, about what he said particularly controversial.

Kaby Lake X is utterly nonsensical.
Introducing the HCC silicon onto the HEDT platform is an interesting move. But the fact is that Intel is going to have to scramble to validate it.

The only thing I question is that Linus says Intel are deliberately waiting on the details for the HCC chips (14-18 core variants) because they want to see what AMD are going to do. I don't think that's right at all personally. The HCC silicone was original intended to be a Xeon only part (as has always been the case in the past), and now they're scrambling to get it released in a consumer variant. We don't have details - I suspect - because Intel is still working them out.

Will they be able to get Turbo Boost 3 working on it?
How high can they clock a 14-18 core part without completely losing control of TDP?
In the past Turbo Boost on HCC server parts was never important because anyone buying a HCC CPU has highly threaded workloads, so whether or not they could boost well over 4Ghz was irrelevant... that's not the case on a HEDT.
Last - but definitely not least - how is the dual-ring bus going to affect core-to-core latency? People are rightly nervous about how Threadripper is going to perform with 4 CCXs spread over 2 dies. But Intel's HCC silicone has similar core-to-core bandwidth and latency issues.

The bottom line is Intel could have been upping core counts for years, but they choose to keep shrinking die sizes and pocketing the higher margins. For comparison, die size estimates put the $1700 6950X at just slightly larger than the i5 2300 which launched for ~$187 in Q1 2011. That's the only reason AMD can actually compete at all, because Intel have been deliberately strolling for years. People will say that's good business for Intel - and maybe they're right, I don't know - but it shows just how important competition is to push progress.

Sorry, missclicked. Anyway, We all know Intel has been competing with itself. It releases an upgraded...whatever, so that people would buy it's new chips. But Intel, weirdly is reacting to Threadripper. If it knows it's competitor, Intel knows that it will outperform it, and for a higher price, but they seem concerned, and with the combination of different classes of high end chips merging together, plus it seems they dropped their Xeons down to a consumer chip, but nerfed the the RAID support to prevent smaller company IT guys from building a server, cheaper.

 

[rhysiam]

Sorry, missclicked. Anyway, We all know Intel has been competing with itself. It releases an upgraded...whatever, so that people would buy it's new chips. But Intel, weirdly is reacting to Threadripper. If it knows it's competitor, Intel knows that it will outperform it, and for a higher price, but they seem concerned, and with the combination of different classes of high end chips merging together, plus it seems they dropped their Xeons down to a consumer chip, but nerfed the the RAID support to prevent smaller company IT guys from building a server, cheaper.

I actually hadn't heard about that RAID nerf before watching that video. It'll be interesting to watch that for sure.

The whole market segmentation thing is interesting. I personally don't really mind that ECC requires the Xeon platform, somehow that seems like a reasonable/fair segmentation to me. But the PCIe lane nerf feels really petty on anything other than the entry level CPUs in the platform. And the RAID lock is outright ridiculous - if true. I'm not entirely sure why ECC feels fair and the others don't, they're all artificial limitations, but that's my perspective.

 

[Supernova1138]

Sorry, missclicked. Anyway, We all know Intel has been competing with itself. It releases an upgraded...whatever, so that people would buy it's new chips. But Intel, weirdly is reacting to Threadripper. If it knows it's competitor, Intel knows that it will outperform it, and for a higher price, but they seem concerned, and with the combination of different classes of high end chips merging together, plus it seems they dropped their Xeons down to a consumer chip, but nerfed the the RAID support to prevent smaller company IT guys from building a server, cheaper.

I actually hadn't heard about that RAID nerf before watching that video. It'll be interesting to watch that for sure.

The whole market segmentation thing is interesting. I personally don't really mind that ECC requires the Xeon platform, somehow that seems like a reasonable/fair segmentation to me. But the PCIe lane nerf feels really petty on anything other than the entry level CPUs in the platform. And the RAID lock is outright ridiculous - if true. I'm not entirely sure why ECC feels fair and the others don't, they're all artificial limitations, but that's my perspective.

Odds are it's because if you really need to have ECC, you're probably buying a Xeon chip anyway, ECC has no real use for consumers and most of the people who do need it probably also benefit from Xeon's higher core count and the ability to run multiple CPUs. PCI-E lanes and RAID are things that can actually be beneficial to the consumers, so placing artificial limitations on those unless you buy a really high end chip or extra hardware can be more grating.