AMD, Analysis, Breaking, Business, CPU, Enterprise, Exclusive, Hardware, Intel, Opinion, VR World

Intel Can’t Hide Strategical Mistakes, AMD to Pounce?

If you would align your wristwatch to the sound of Intel’s tick tock, you would miss your important meeting. The launch of a new generation of processors represents a symbolic beginning of a new cycle and for Intel, “Kaby Lake” launch was everything but normal. First launching the Ultra-low Voltage parts (the U series), then the Desktop line-up and finally the mainstream notebook designs, with troublesome rumors about its enterprise, Xeon-branded counterpart.

Kaby Lake is perhaps too focused on the processor power task rather than processor performance. A lullaby. The same story for the last six years continues. Admittedly, for the past 18 months there were whispers that something big is going at AMD. A few months later the first RYZEN specimens are leaking into the wild. Ryzen is good and Intel after a decade in hibernation has a reason to wake up.

In 2014, Intel unveiled the unlocked Pentium G3258, and with Kaby Lake the company finally brought an unlocked i3, the “i3-7350K”, as well as Pentium processors with HyperThreading technology. The pace of that “progress” was likely to come by sluggish listening to the market, unrelated to Ryzen processors. While AMD will probably show its complete line-up during the rumored special NDA event on February 16th, 2017 and likely unveil it to the audience on February 28th, the price and performance are more or less known. AMD insisted on a “hard” launch for RYZEN processors. Confirming that is the relatively recent agreements between AMD and GlobalFoundries regarding the wafer sourcing and preparation for the 7nm technology node, which now includes not just TSMC, but Samsung as well. We would not be surprised if AMD included Intel in the amended Wafer Supply Agreement. After all, the biggest non-surprise of the year will be Intel “Lake” silicon with a special low-clocked Radeon RX 540 die joined in what many see as unholy matrimony. Shows a thoughtful approach that should cover and solve yield issues, otherwise standard Achilles heel of AMD.

It seems that the thing has matured and we yet just need to wait a few weeks. Probably inspired by RYZEN performance, Intel apparently leaked several “innovations” for the middle and upper segment. Xeon Gold line of processors should combat 16, 24 and 32-core workstation processors based on “Naples” and it’s 16-core younger brother. AMD releasing four completely new dies in a single year (4C and 8C for desktop/mobile, 16C/32C for workstation/server)? Two new Sockets? They really planned a comeback that just might be a knockout too.

Kaby Lake-X is the same as Skylake-X: LGA2066 Required

New i5 and i7 models are being leaked; Core i5 7640K and i7 7740K. Core i5 7640K is a quad-core model with a 4.0GHz base clock and Turbo mode of 4.2GHz, with TDP rated at 112W. Core i7 7740K is quad-core processor with Hyper-Threading and a base frequency of 4.3GHz, Turbo being set at 4.5GHz for all four cores (!). TDP of this i7 model is also 112W so it is posbile that just like i3-7350K, i5 7640K will also have Hyper-Threading.

Intel's Basin Falls Platform:Socket LGA2066 for Kaby Lake-X and Skylake-X Processors

Now, this is where it gets interesting. Both Core i5 7640K and i7 7740K are codenamed Kaby Lake-X, while the existing i7-7700K and i5-7600K are codenamed Kaby Lake-S. The new X parts come with integrated graphics disabled, and a high-performance thermal interface material (TIM) under the integrated heatspreaders (IHS). It took long time but, overclockers, welcome! Sadly, there’s always rain on someone’s parade and this time we’re talking about an Indonesian Monsoon. Kaby Lake X is not the answer to AMD RYZEN, but rather a processor line which already leaked. The Socket in question is brand new and unfortunately, it is not a six-channel memory controller known from Socket LGA3657, which is used on Xeon E7 for 4P and 8P motherboards.

Should AMD launch its 16C/32T or 32C/64T processor and the new 8-channel memory controller into the workstation and high-end desktop space (HEDT), the bandwidth results are all but guaranteed to walk all over anything Intel can offer today: Intel Skylake-X should reach 85.3 GB/s, while the AMD Naples should easily beat that with 170.7 GB/s. It gets worse though, as there are some leaked tests that show 8-channel memory controller supporting DDR4-3000: 192GB/s is the figure one AMD GPU can only dream off: of course, we’re talking about Radeon RX 460 (112 GB/s).  And that’s a GDDR5 powered GPU. AMD processors for Socket SP3 and SP4 are starting to look tasty indeed.

These numbers could shake Intel the most; Xeon family. Intel is preparing Xeon Gold line of processors targeting the pro-consumer market of creative productivity applications. Xeon Gold line will be based on the 14nm “Skylake-EP” silicon. The first model that leaked is the “Xeon Gold 6150,” featuring 18 cores with HyperThreading. The core clock speed is 2.70 GHz and highest Turbo Boost clock is set at 3.70 GHz. It comes with 1MB of L2 cache per core and 24.75 MB of shared L3 cache (read: physical silicon has 22 cores, 4 disabled for yield purposes). Memory interface is (again) quad-channel DDR4 on a Socket LGA2066 (Purley is Xeon’s twin brother of Basin Lake for HEDT).

Intel’s $7 Billion for 7nm in Arizona: Striking a Balance?

Fab 42 is Intel’s project for what is called to be the world’s most advanced semiconductor plant – something we doubt due to lack of EUV process, which is what Samsung will bring to the table for their own 7nm process. Altough Intel missed wave after wave, focusing on loosing money in making chips for mobile devices, Fab 42 should show that Intel is “chipzilla” once more, and demands to be treated with respect.

Just a few years ago, Intel announced it is delayin the Fab 42, because it projected that “due to lack of demand for the company’s chips” it would not be posible to fully utilize the Fab 42 capacity. One dumping of mobile strategy and firing ton of people later, Intel announced that it will invest more than $7 billion to finally complete the Fab 42 semiconductor foundry. It is interesting that on that occasion, Intel’s CEO Brian Krzanich sent off an e-mail to employees mentioning Moore’s Law several times.

Addition to the entire AMD / Intel story is the fact that in a year when Razen is coming to market, Intel might license AMD’s own GPU technology for integration into its future processors. In a story which leaked as a part of the whole Kyle Bennett vs. Raja Koduri blow out, which followed Kyle Bennett vs. Roy Taylor blow out, one of major secrets leaked – that RTG was the prime runner for getting the Intel GPU deal, after Intel fired over 1,000 employees from its graphics division. Actually, the whole graphics division allegedly got the boot. Given that RTG lost the Nintendo deal to NVIDIA, they had engineering resources to trying to nab Intel. Also, a long standing rumor that Raja Koduri wanted to leave AMD for Intel, and got RTG as a “consolation prize” reared its ugly head as well. Be that as it may, no matter what happened behind the closed doors at AMD / RTG – it worked since they stock skyrocketed over the past 18 months. In fact, if you had $100,000 for AMD and $100,000 for NVIDIA shares in July of 2015, today you would have… $803,589 and $592,257. Yes, you would be a millionaire. On the other hand, $100,000 spent in INTC stock in July 2015 would yield a grand total of… $122,443. As anemic as Kaby Lake performance is compared to Skylake.

By teaming with AMD i.e. RTG on the GPU side of things, there should be less money wasted and Intel will have to raise the bar which is good for all the users. What is the future? There are signs that Intel has made the chip with on-die silicon photonics technology, which just might be one of the most important technologies of our generation. Then again, we saw the first silicon photonics demo back in 2007, and even Thunderbolt originally was an optical interconnect. However and if, commercialization is still three to five years away from us.

  • Pet Mey


  • Ashley Gann

    “strategical” 🙁

    • whispering malarkey

      It’s a common mistake. Many misspell “strategery”.

  • Paul Hamilton

    Rumored NDA event… What? I keep up with AMD news and I have heard of nothing of this and searching around still can’t find anything. Source for this?

  • Oh, you have to take chances, strategies sometimes fail, they work in another company or place. Knowing the time to use is the secret of success

  • Howards_cat_Tarzan

    To be honest, Ryzen address just a tiny part of the CPU market. The ASP of a PC is under 400 dollars = 0 Ryzen. The market PC over 1000 dollar is 90% Apple = 0 Ryzen. Servers are Intels domain.
    AMD did the thing I have advocated (with Steve Jobs) since 2009: Remove the iGPU and give us more cores and cache.
    Todays Intel home CPUs are 50-80% iGPU. Intel could remove that TODAY and release a 4×2 CPU with 256 eDRAM. The cost manufacture this CPU is the the same as the crappy iGPU versions.
    The non competition for X86 since Apple moved to Intel (yes. Not only AMD disappeared but also PPC. Until Apple switched both Intel and PPC bumped the CPU every quarter like clockwork. When Apple switched: 1 bump/year for Intel). If you look at the die size and transistor count for the quad core 2 since 2006 and factor in node shrinks: We should have customer 10-12 core Intels for the same price. It cost the same for Intel to manufacture.

    The reason why Intel don’t do this is the server market. Intel bleeds that market with CPUs costing 4000-9000 dollars. If Intel released 10-12 cores CPUs for 300 dollars that would hurt the server market.

    X86 is insanely priced. The problem is that customers and press defends it. Look at a BOM price for iPhone7 for example. “Apple A10 BOM 25 dollars”. Apple A10 is LARGER than intels 4 core/8 threads CPUs but people don’t talk about a “25 dollar BOM” for Intel. Todays Intels are under 130mm2 large. The cost under 25 dollars to manufacture and we think its normal to pay 300 dollar + 100 dollar motherboard tax.

    What AMD done with Ryzen is nothing spectacular. Moving from 28nm to 14nm FinFET should give 50% more performance. Add to that AMD removed iGPU. Great!

    Intel must keep AMD alive. Without AMD as a X86 license the government would do the right thing and force Intel to license out X86 like ARM does with ARM. With X86 licensed out we would see the same 50% + each year that ARM have had since 2007.

    Intel have 90% profit margins on CPUs. A huge part of this is that Intel have survived by being 2 nodes ahead of the competition. Intel needs to be ahead since CISC have a 30% size tax. The same performance on X86 needs a 30% larger die than RISC/ARM/PPC. Thanks to Sammy/Apple (Apple prepaying for fab lines) we have seen a rapid progress in node shrinks non-intel. 5 years: 40-32-28-20-16-14-“10″nm. Intels advantage is gone.

    Please kill off X86. Its not the fastest. Its not the cheapest. Its just good enough and have Windows.