Bondrewd
Veteran
But there's no CNL-H, only Y and U.
As for CFL-H, I am not sure.
Probably no.
As for CFL-H, I am not sure.
Probably no.
Intel has lost or let go a fair amount of its graphics resources in recent times, so there were some signs that it was not intending to push its graphics performance in the same range as AMD's APUs. Historically, it's also generally the case that Intel has done fine without that level of performance.
Its EDRAM-paired variants and even this custom deal also point to Intel not being interested in running in AMD's performance range while charging AMD's paltry prices.
AMD's announced APU products are based on Raven Ridge, and are not in the same league price-wise or performance-wise.
One interpretation from this deal is that neither AMD or Intel consider an APU with HBM and GPU resources on this scale practical. Intel, however, has the desire and integration capability to mash together a discrete and some variant of its mobile line.
AMD's x86 group likely had a similar choice at one point. So far, I have not seen an indication that there's anything like the linking capability in Raven Ridge, and the die shots do not show an IF link that would provide a similar connection to what Intel is saying they've implemented. The dedicated control link used by Ryzen isn't an obvious thing to look for, however. An xGMI-type option could also be possible without being visible in a die shot, but the first promise of that isn't in this space or mentioned for the near term.
Isn't CFL-H actually CNL-H? https://www.intel.com/content/www/u...ocessors-and-chipsets/platform-codenames.html
Bondrewd
Veteran
There's no CNL-H and we know nothing of CFL-H.
For now I'll assume it uses KBL-H.
Not sure one should be comfortable to accept the loss of a good margin sector and replacing that with a lower margin GPU deal that importantly also solidifies Intel CPU division for mobile and possibly some other segments and feel OK as Zen core is gaining share elsewhere.
The CPU division is where the best revenue and margins exist, AMD needs to expand that business as much as possible.
If this deal only affects Apple well the fallout may be less but the announcement on the AMD side came from the Radeon gaming division VP Scott Herkelman, specifically he mentions AAA games and gamers, content creators, expanded set of enthusiasts.
This is a great market to develop a complete CPU-GPU solution that is solely AMD, the margins in such segments are great.
But anyway this deal IMO solidifies Intel at the expense of any potential for AMD CPU team in near future or next generation launch gaining a foothold now in these type of segments, leaving the CPU division with the lower margins higher volume competing segments.
And we know that is not usually as profitable.
I guess we will know more in the months to follow the scale of any such deals between Intel and the Radeon division and whether it has any meaningful impact on AMD business.
Last edited:
Some of the articles say that this is an Intel CPU+IGP paired with GCN+HBM.
I would have expected that the Raven Ridge designers had more lead time than Intel, but current information doesn't include the capability. Given the touting of AMD's new modular infrastructure, this should have been even easier to implement. I'd interpret this as the option being actively decided against.
RTG may supply the GPUs, but I wouldn't think it would have veto power over a more extensible Raven Ridge if the designers wanted it.
If there was a willingness to implement the fabric in Vega already, it seems like the next incremental step would have been to add PHY for it, although we don't know if plans changed with Vega 10/Greenland or what's going on with Vega 11.
Apple or somebody else that feels like handling the complexity specific to a hybrid setup and disparate D3D12 feature levels/doesn't care about DX12. It does feel like someone is motivated enough to go through with this specific combination.
The notable GPU to CPU relationship that isn't clearly covered by something like a Raven Ridge system with Vega 11 is the announced shared power management. If Vega 11 has HBM, and either Ryzen or Raven Ridge save area with their on-die southbridge, a substantial portion of the marketing features like area savings are shared.
It strikes me as counterproductive if RTG were allowed to circumvent AMD's overall product interests without the rest of the company being able to weigh in, but I have no visibility on any internal shenanigans.
Semi-custom at least early on was characterized as needing some hundreds of millions to a billion dollars in expected revenue at a minimum before AMD would bother. I'm curious if that still holds.
As an AMD shareholder, I would ecstatic because in addition to “just” revenue with each additional semi-custom client AMD gains another major stakeholder in its financial viability. Yes, there is potential competition with internal product but bird in hand/two in the bush and all that, plus it’s hard to put the price on the qualitative benefits of having your arch rival rely on you for a solution.
If I was still an Intel shareholder, I would be dismayed because this represents yet another failure of leadership, with the company seemingly stumbling from one execution failure/dubious financial decision to another.
1) Completly mishandled process management and...
2) Lost the critical manufacturing leadership
3) Got mired in stagnation/lack of innovation malaise after tearfully promising “never again” following Pentium 4 fiasco
4) Enabled a resurgence in market and mindshare for left-for-dead rival
5) Tharashed and flailed, marketing-wise in response to Ryzen/Thredripper like a drunk teenager caught urinating in a public park
6) Now is buying products from AMD, putting much-needed cash in their R&D coffers that will be used down the road to develop products targeted at Intel
7) But hey, while the core market was a smoldering dumpsterfire, spent $18 billion dollars on a fledgling self-driving company in the process of losing its highest-profile client. Truly the best use of resources given the circumstances!
Bondrewd
Veteran
It's twice to thrice as hilarious considering BK's background.
Let's not forget them openly FUDing at AMDs multi-die approach while promising to deliver exactly it for 10nm products.
Honestly retarded acquisitions happened even back in the Otellini days (member McAfee?).
I mean Otellini liked anti-competitive practices, but at least his Intel delivered good products.
Bondrewd
Veteran
That's PCH, duh.
300 series chipsets besides Z370 are also Cannonlake platform.
The Register article pretty much sums up my thoughts on this (when you read it to the end).
Intel are not friendly at all to AMD and their CPU Division, yet we have this alternative situation with the Radeon group.
http://www.theregister.co.uk/2017/11/06/amd_intel_laptop_chip/
Anyway, to give a comparison on the Radeon deal, the CPU division might as well work with Nvidia to compete against a very dominant Intel (CPU perspective) in the HPC-AI/scientific/analytics space because AMD GPU team is still off from having a competing product there for a while yet due to the complete ecosystem requirements.
It could enable the CPU division to sell more strongly (same way IBM is working with Nvidia) in the most lucrative market, but it would cripple the AMD GPU divisions aspirations in that market - not saying they should just using it as some context.
Like I said there is a balance between what is good for the division and the company as a whole, but the scope of any such deal between Intel and Radeon Group needs to be seen months down the line from now and I could be over-reacting (I am a fan of the CPU division and what they have done).
Intel are not friendly at all to AMD and their CPU Division, yet we have this alternative situation with the Radeon group.
http://www.theregister.co.uk/2017/11/06/amd_intel_laptop_chip/
Regarding business impact, I should say it can be both positive and negative.
Anyway, to give a comparison on the Radeon deal, the CPU division might as well work with Nvidia to compete against a very dominant Intel (CPU perspective) in the HPC-AI/scientific/analytics space because AMD GPU team is still off from having a competing product there for a while yet due to the complete ecosystem requirements.
It could enable the CPU division to sell more strongly (same way IBM is working with Nvidia) in the most lucrative market, but it would cripple the AMD GPU divisions aspirations in that market - not saying they should just using it as some context.
Like I said there is a balance between what is good for the division and the company as a whole, but the scope of any such deal between Intel and Radeon Group needs to be seen months down the line from now and I could be over-reacting (I am a fan of the CPU division and what they have done).
Bondrewd
Veteran
Goddammit, this is a semi-custom chip, RTG only did the design.
They don't need to, GPU users will be begging for EPYC themselves anyway.
I wonder whether this project started before Ryzen CPU cores were final. Nowadays you have to ask whether it would be simply easier for AMD to integrate Ryzen cores with Vega GPU... Which they did with Raven Ridge. They also now have HBM2 GPU products in Vega 56/64. Why didn't AMD simply do a Raven Ridge APU with scaled up GPU + HBM2?
The only thing that would make sense is that Apple ordered this chip for the next Macbooks. They already use Intel CPUs and AMD GPUs. Integration saves cost, space and power. Intel still has slightly bit better single thread performance and process advantage, so it makes sense that this is built on top of their tech. This would also explain the secrecy around the project and no news about the launch products. Apple doesn't like talking about their products before the launch.
Apple moving Macbooks to their custom GPUs and CPUs is a threat to both Intel and AMD. This is the only sane reason why these companies would collaborate.
The only thing that would make sense is that Apple ordered this chip for the next Macbooks. They already use Intel CPUs and AMD GPUs. Integration saves cost, space and power. Intel still has slightly bit better single thread performance and process advantage, so it makes sense that this is built on top of their tech. This would also explain the secrecy around the project and no news about the launch products. Apple doesn't like talking about their products before the launch.
Apple moving Macbooks to their custom GPUs and CPUs is a threat to both Intel and AMD. This is the only sane reason why these companies would collaborate.
Bondrewd
Veteran
Interposers are not as cheap as EMIB; besides AMD most definetly has no time to serve such niche markets.
Also the SKU reeks of Apple, and Apple is not going to use AMD CPUs until these have native TB3.
Absolutely.
Actually it's a threat to Intel only. Intel asked for semi-custom design.
Apple wont ARM the iMac or iMac Pro or Mac Pro where most Radeons go.
You save a bit on the interposer silicon, but EMIB assembly can't be easier, so I would expect yield to be lower.
The primary driver for EMIB is that it allows much larger packages. Imagine an Epyc style processor with EMIB inter-die links; much more bandwidth at lower power consumption.
Cheers
Bondrewd
Veteran
Yet not fully connected, 2 hop diameter inside the package (well, at least to my understading).
Should be good for GPUs, questionable for CPUs.
Yup, this is one more reason to bury the old zombie idea "discrete GPUs are going to be dead in 5 years". If that would be the way, why would a solution like this exist (and even be developed by two companies who normally compete with each other)?