Intel: 8th generation processor with integrated AMD graphics chip and HBM2 memory on the EMIB bus
A recent statement from Intel thundered like a bolt from the blue. The company has officially announced that it is working on a new generation of processors in which high-performance x86 cores will be combined with an AMD Radeon graphics chip into a single processor. This was made possible thanks to Intel's EMIB (Embedded Multi-Die Interconnect Bridge) bus, which allows you to mount chips with different technological processes on a single substrate, and provides a high speed of data exchange between them. And so that no one seemed to be little, Intel announced that the memory will also have high-bandwidth HBM2 memory.
For more than a year, Intel has often spoken about the new EMIB technology, the main idea of which is the ability to assemble several different silicon arrays into a single processor, and to ensure high speed of data exchange between them. And all this for a price much lower than when using a silicon interpozer. At Intel's Manufacturing Day at the beginning of 2017, Intel presented a slide showing the capabilities of the new technology: a single processor including x86 cores made using the same technology, a graphics chip using the second technology and, for example, IO, memory, and a wireless communication chip, also made on various technologies. In this way, EMIB turns the processor into a similarity to the LEGO constructor.
Devices with EMIB technology first appeared on the market with the release of the programmable logic circuit (FPGA) of Intel Altera, on which EMIB was used to connect the main FPGA chip with transceivers. After this successful experience, the main goal was to add blocks of RAM to the same “package”. Such a “mix” of different matrices in one chip would allow the end user to configure the chip for any unique tasks. The advantages of EMIB were quite obvious - without the disadvantages of the MCP and the high cost of interposer, the technology goes far beyond the limitations of the usual lithographic process. All of the above talked about the imminent appearance of EMIB in PC central processing units, and now we are seeing on the market high-end server hardware 900 mm2 chips, including several matrices manufactured by various technological processes.
After the announcement of EMIB, Intel's Manufacturing Day and Hot Chips, there was a lot of talk about how exactly Intel is going to bring the new technology to users of personal computers. The main issue was the presence of Intel's own integrated graphics chip. And although Intel and NVIDIA had a joint cross-licensing agreement signed in 2011 - the agreement expired on April 1, 2017, no mention of its extension was voiced by either of the parties. From time to time there were speculations about possible cooperation with AMD, which, despite the competition in the central processor market, looked like a more interesting partner. However, neither Intel nor AMD were in a hurry to disclose any information about the agreement. Historically, Intel has refused to comment on such issues in advance. And although some sources managed to publish tests of new equipment on SiSoft (reporting “information leakage”), the first announcement from Intel was made in early November 2017.
Intel's official announcement in the announcement of the new chip contains several details that should be considered more closely.
Intel uses a neutral phrase “product that will become part of the family,” from which it is unclear what is at stake in applying technology in the entire line of eighth-generation processors, or, roughly speaking, about a single processor model. Currently, the Core-H processor line is represented by a 45-watt Kaby Lake with Intel's integrated GT2 graphics.
It will be interesting to know if the core part of the Core-H will be replaced by a new chip from AMD, or will the Core-H chip be redesigned, on which only the processor cores will remain, or both graphics processors will be able to work independently.
The introduction of HBM2 in the new product does not look complicated - Intel has successfully integrated HBM2 into its products based on Altera EMIB, so in this part everything should go without problems.
The next interesting point is "AMD's customized discrete graphics chip from RTG." This means that none of the existing AMD products is ready for EMIB integration, but AMD is ready to create a custom design of the existing chip for placing it on an Intel silicon matrix.
One of the problems when running multiple chips in one package is managing the data transfer and power consumption of the chips. Recently, AMD solved this problem on its server processors and inside its APUs by using an internal Infinity Fabric connection (which replaces the HyperTransport bus), which seems to remain outside the joint project. The statement states that the “assembly” chip “shares a power framework”, I would also like to consider this statement deeper. Alternatively, Intel can use separate power for the CPU and GPU using the built-in voltage regulator (as, for example, it was in the Broadwell microarchitecture), or do something similar to AMD using a single power supply circuit with digital LDOs (Low-dropout regulator) - application of this technology was reported to Ryzen Mobile AMD just a couple of weeks ago.
It seems that Intel is ready to make several announcements over the next few months in this project, and the CES (Consumer Electronics Show) will be very soon, in January 2018.
Now, although this is a step back, let's consider what all this means and what market Intel is targeting. AMD recently announced (and will release with its closest products) a Ryzen Mobile mobile platform, including a quad-core Zen and up to 10 CU of Vega graphics chip. The statements from Intel and AMD do not specify which graphics core they use (is it possible to use the last-generation chips for competitive purposes?), But at the same time they claim that they will use Core-H series processors that consume 45 watts. AMD currently does not have products in roughly the same range and focused on the development of Ryzen Mobile for thin and ultra-light notebooks. If AMD introduces Ryzen Mobile to more powerful devices, such a new product will be a direct competitor to the new joint development.
A look at the image provided by Intel when announcing a product actually adds a couple more questions. Here we see the Intel chip on the right, the AMD graphics chip in the middle and the HBM2 chip next to the GPU. The Intel chip is quite far from the AMD chip, which suggests that these two devices are not connected via the EMIB bus if the layout is accurate. But the proximity of a large GPU chip to what looks like a HBM2 stack suggests that they are just connected via EMIB (judging by how closely the chips are located in Altera products)
Apparently, EMIB is used, but it does not seem to be used to connect ALL chips. It is particularly interesting that neither Intel nor AMD offered any additional briefings on such a loud announcement and, as a result, we have a lot of unanswered questions.
And one more thought for last. Apple makes extensive use of Intel 45W processors for iMacs. Now Intel is offering AMD graphics (the preferred third-party graphics chip maker for Apple) in a segment that previously only existed based on Intel's Crystalwell / eDRAM products, which means that Apple’s widespread use of the chip could be the next step in the evolution of this product.
Having rested a couple of hours from reflection, we found several new ideas. First, judging by the wording and the Intel video, it is safe to say that EMIB is used only between the GPU and HBM2. The distance between the CPU and the GPU is too large for EMIB, so, most likely, they are simply connected by a new implementation of PCIe. Remote chip placement can also reduce power dissipation.
The agreement between AMD and Intel is that Intel buys chips from AMD, and AMD provides driver support, as was done for consoles. There is no cross-licensing: Intel simply provides AMD IP to provide EMIB connectivity to the GPU, but this IP is valid only for products that AMD sells to Intel (similar to the semi-official agreement on which the collaboration will be based).
Since Intel buys AMD chips, it is reasonable to assume that they will buy more than one configuration, in accordance with how Intel wants to organize the product stack. Intel can combine a smaller 10 CU GPU design with a dual core, and a more powerful 20+ CU GPU with a quad-core mobile processor. It seems that some benchmark sources suggest that there are at least two variants of Polaris-like configurations, possibly up to 24 CU (computing units) in the top model. We will obviously wait before confirming these statements, since Polaris was not originally created for the HBM2 memory. In theory, working with HBM2 memory requires a GPU developed specifically for HBM2 - here the key word is “date management”. However, if the GPU is working with memory "normal", you will probably need to use the HBM2 controller.
Ideally, AMD could sell Intel its Polaris designs, for example, two generations ahead. With the recent financial success of AMD, they can afford it, or Intel can offer a high price for the latest developments. However, none of the companies commented on the agreement between the two corporations, limiting themselves to press releases.
In discussions with Peter Bright from Ars Technica, we came to the conclusion that Intel’s GPU will still have its own integrated graphics and the system will operate in the switching mode of graphic matrices. This is easy to implement if the CPU and GPU are connected via PCIe, since all the mechanisms are in place. Thanks to the integrated Intel graphics processor, video playback will be performed on an Intel chip, and then sent to the display controller - this will allow you to temporarily turn off the power of the AMD GPU and HBM2, saving energy. If the GPU and HBM2 were included constantly, we would expect a reduction in the battery life of future devices.
It was separately discussed whether the new product is aimed primarily at Apple. This is possible, given that Apple is lagging behind Intel, which is implementing eDRAM on its Crystalwell processors, and the latest generation of Crystalwell seems to be used (with rare exceptions) in the Apple iMacs. As mentioned above, Intel stated that they have several partners interested in the new product, and that we should expect more information about the device in the first quarter of 2017. With such a statement about the device, it is reasonable to assume that there are various OEMs waiting to start working with hardware.
Among the many devices reviewed by us, this turned out to be one of the most mysterious. Intel reported on Core-H processors with chips that consume 35 watts and 45 watts. At the same time, they reported that the new product would not be a direct competitor to Ryzen Mobile. However, in the presented demo video, it is clear that the best use for this design is thin and light notebooks, such as 2-in-1 and ultra-portables. Does this mean that Intel will switch to 15W devices? Well, if Intel buys several chip configurations from AMD, then the binding of dual-core i5 to 10 CU graphics is more than plausible. And, if AMD sells Intel the older Polaris design, then it, at least, reserves this advantage.
As advertising. On the eve of the winter holidays promotions become even more relevant! Hurry up to take advantage of the New Year offer and get a 25% discount on the first payment when ordering for 3 or 6 months!
These are not just virtual servers! This is a VPS (KVM) with dedicated drives, which can be no worse than dedicated servers, and in most cases - better! We made VPS (KVM) with dedicated drives in the Netherlands and the USA (configurations from VPS (KVM) - E5-2650v4 (6 Cores) / 10GB DDR4 / 240GB SSD or 4TB HDD / 1Gbps 10TB available at a uniquely low price - from $ 29 / month , options are available with RAID1 and RAID10) , do not miss the chance to place an order for a new type of virtual server, where all resources belong to you, as on a dedicated one, and the price is much lower, with a much more productive hardware!
How to build the infrastructure of the building. class c using servers Dell R730xd E5-2650 v4 worth 9000 euros for a penny? Dell R730xd 2 times cheaper? Only we have 2 x Intel Dodeca-Core Xeon E5-2650v4 128GB DDR4 6x480GB SSD 1Gbps 100 TV from $ 249 in the Netherlands and the USA!
For more than a year, Intel has often spoken about the new EMIB technology, the main idea of which is the ability to assemble several different silicon arrays into a single processor, and to ensure high speed of data exchange between them. And all this for a price much lower than when using a silicon interpozer. At Intel's Manufacturing Day at the beginning of 2017, Intel presented a slide showing the capabilities of the new technology: a single processor including x86 cores made using the same technology, a graphics chip using the second technology and, for example, IO, memory, and a wireless communication chip, also made on various technologies. In this way, EMIB turns the processor into a similarity to the LEGO constructor.
Devices with EMIB technology first appeared on the market with the release of the programmable logic circuit (FPGA) of Intel Altera, on which EMIB was used to connect the main FPGA chip with transceivers. After this successful experience, the main goal was to add blocks of RAM to the same “package”. Such a “mix” of different matrices in one chip would allow the end user to configure the chip for any unique tasks. The advantages of EMIB were quite obvious - without the disadvantages of the MCP and the high cost of interposer, the technology goes far beyond the limitations of the usual lithographic process. All of the above talked about the imminent appearance of EMIB in PC central processing units, and now we are seeing on the market high-end server hardware 900 mm2 chips, including several matrices manufactured by various technological processes.
After the announcement of EMIB, Intel's Manufacturing Day and Hot Chips, there was a lot of talk about how exactly Intel is going to bring the new technology to users of personal computers. The main issue was the presence of Intel's own integrated graphics chip. And although Intel and NVIDIA had a joint cross-licensing agreement signed in 2011 - the agreement expired on April 1, 2017, no mention of its extension was voiced by either of the parties. From time to time there were speculations about possible cooperation with AMD, which, despite the competition in the central processor market, looked like a more interesting partner. However, neither Intel nor AMD were in a hurry to disclose any information about the agreement. Historically, Intel has refused to comment on such issues in advance. And although some sources managed to publish tests of new equipment on SiSoft (reporting “information leakage”), the first announcement from Intel was made in early November 2017.
Intel's official announcement in the announcement of the new chip contains several details that should be considered more closely.
The new product, which will be part of our 8th generation Intel Core family, combines our high-performance Intel Core H series processor, second-generation high-speed memory (HBM2) and a discrete graphics chip from third-party AMD Radeon Technologies Group * - all in a single package of processors.
Intel uses a neutral phrase “product that will become part of the family,” from which it is unclear what is at stake in applying technology in the entire line of eighth-generation processors, or, roughly speaking, about a single processor model. Currently, the Core-H processor line is represented by a 45-watt Kaby Lake with Intel's integrated GT2 graphics.
It will be interesting to know if the core part of the Core-H will be replaced by a new chip from AMD, or will the Core-H chip be redesigned, on which only the processor cores will remain, or both graphics processors will be able to work independently.
The introduction of HBM2 in the new product does not look complicated - Intel has successfully integrated HBM2 into its products based on Altera EMIB, so in this part everything should go without problems.
The next interesting point is "AMD's customized discrete graphics chip from RTG." This means that none of the existing AMD products is ready for EMIB integration, but AMD is ready to create a custom design of the existing chip for placing it on an Intel silicon matrix.
In close cooperation, we developed a new custom graphics chip, and this is an excellent example of how we can compete and work together, ultimately providing the innovations that consumers expect ... Thus, we have developed a resource allocation mechanism - this new interface allows the processor to work together. Intel discrete graphics chip and dedicated graphics memory. We have added unique software drivers and interfaces to this non-standard discrete graphics processor, which coordinates information among all elements of the platform.
One of the problems when running multiple chips in one package is managing the data transfer and power consumption of the chips. Recently, AMD solved this problem on its server processors and inside its APUs by using an internal Infinity Fabric connection (which replaces the HyperTransport bus), which seems to remain outside the joint project. The statement states that the “assembly” chip “shares a power framework”, I would also like to consider this statement deeper. Alternatively, Intel can use separate power for the CPU and GPU using the built-in voltage regulator (as, for example, it was in the Broadwell microarchitecture), or do something similar to AMD using a single power supply circuit with digital LDOs (Low-dropout regulator) - application of this technology was reported to Ryzen Mobile AMD just a couple of weeks ago.
Expect news in the first quarter of 2018, including new systems from major OEMs based on this exciting technology.
It seems that Intel is ready to make several announcements over the next few months in this project, and the CES (Consumer Electronics Show) will be very soon, in January 2018.
Now, although this is a step back, let's consider what all this means and what market Intel is targeting. AMD recently announced (and will release with its closest products) a Ryzen Mobile mobile platform, including a quad-core Zen and up to 10 CU of Vega graphics chip. The statements from Intel and AMD do not specify which graphics core they use (is it possible to use the last-generation chips for competitive purposes?), But at the same time they claim that they will use Core-H series processors that consume 45 watts. AMD currently does not have products in roughly the same range and focused on the development of Ryzen Mobile for thin and ultra-light notebooks. If AMD introduces Ryzen Mobile to more powerful devices, such a new product will be a direct competitor to the new joint development.
A look at the image provided by Intel when announcing a product actually adds a couple more questions. Here we see the Intel chip on the right, the AMD graphics chip in the middle and the HBM2 chip next to the GPU. The Intel chip is quite far from the AMD chip, which suggests that these two devices are not connected via the EMIB bus if the layout is accurate. But the proximity of a large GPU chip to what looks like a HBM2 stack suggests that they are just connected via EMIB (judging by how closely the chips are located in Altera products)
Apparently, EMIB is used, but it does not seem to be used to connect ALL chips. It is particularly interesting that neither Intel nor AMD offered any additional briefings on such a loud announcement and, as a result, we have a lot of unanswered questions.
And one more thought for last. Apple makes extensive use of Intel 45W processors for iMacs. Now Intel is offering AMD graphics (the preferred third-party graphics chip maker for Apple) in a segment that previously only existed based on Intel's Crystalwell / eDRAM products, which means that Apple’s widespread use of the chip could be the next step in the evolution of this product.
Additional comments
Having rested a couple of hours from reflection, we found several new ideas. First, judging by the wording and the Intel video, it is safe to say that EMIB is used only between the GPU and HBM2. The distance between the CPU and the GPU is too large for EMIB, so, most likely, they are simply connected by a new implementation of PCIe. Remote chip placement can also reduce power dissipation.
The agreement between AMD and Intel is that Intel buys chips from AMD, and AMD provides driver support, as was done for consoles. There is no cross-licensing: Intel simply provides AMD IP to provide EMIB connectivity to the GPU, but this IP is valid only for products that AMD sells to Intel (similar to the semi-official agreement on which the collaboration will be based).
Since Intel buys AMD chips, it is reasonable to assume that they will buy more than one configuration, in accordance with how Intel wants to organize the product stack. Intel can combine a smaller 10 CU GPU design with a dual core, and a more powerful 20+ CU GPU with a quad-core mobile processor. It seems that some benchmark sources suggest that there are at least two variants of Polaris-like configurations, possibly up to 24 CU (computing units) in the top model. We will obviously wait before confirming these statements, since Polaris was not originally created for the HBM2 memory. In theory, working with HBM2 memory requires a GPU developed specifically for HBM2 - here the key word is “date management”. However, if the GPU is working with memory "normal", you will probably need to use the HBM2 controller.
Ideally, AMD could sell Intel its Polaris designs, for example, two generations ahead. With the recent financial success of AMD, they can afford it, or Intel can offer a high price for the latest developments. However, none of the companies commented on the agreement between the two corporations, limiting themselves to press releases.
In discussions with Peter Bright from Ars Technica, we came to the conclusion that Intel’s GPU will still have its own integrated graphics and the system will operate in the switching mode of graphic matrices. This is easy to implement if the CPU and GPU are connected via PCIe, since all the mechanisms are in place. Thanks to the integrated Intel graphics processor, video playback will be performed on an Intel chip, and then sent to the display controller - this will allow you to temporarily turn off the power of the AMD GPU and HBM2, saving energy. If the GPU and HBM2 were included constantly, we would expect a reduction in the battery life of future devices.
It was separately discussed whether the new product is aimed primarily at Apple. This is possible, given that Apple is lagging behind Intel, which is implementing eDRAM on its Crystalwell processors, and the latest generation of Crystalwell seems to be used (with rare exceptions) in the Apple iMacs. As mentioned above, Intel stated that they have several partners interested in the new product, and that we should expect more information about the device in the first quarter of 2017. With such a statement about the device, it is reasonable to assume that there are various OEMs waiting to start working with hardware.
Among the many devices reviewed by us, this turned out to be one of the most mysterious. Intel reported on Core-H processors with chips that consume 35 watts and 45 watts. At the same time, they reported that the new product would not be a direct competitor to Ryzen Mobile. However, in the presented demo video, it is clear that the best use for this design is thin and light notebooks, such as 2-in-1 and ultra-portables. Does this mean that Intel will switch to 15W devices? Well, if Intel buys several chip configurations from AMD, then the binding of dual-core i5 to 10 CU graphics is more than plausible. And, if AMD sells Intel the older Polaris design, then it, at least, reserves this advantage.
As advertising. On the eve of the winter holidays promotions become even more relevant! Hurry up to take advantage of the New Year offer and get a 25% discount on the first payment when ordering for 3 or 6 months!
These are not just virtual servers! This is a VPS (KVM) with dedicated drives, which can be no worse than dedicated servers, and in most cases - better! We made VPS (KVM) with dedicated drives in the Netherlands and the USA (configurations from VPS (KVM) - E5-2650v4 (6 Cores) / 10GB DDR4 / 240GB SSD or 4TB HDD / 1Gbps 10TB available at a uniquely low price - from $ 29 / month , options are available with RAID1 and RAID10) , do not miss the chance to place an order for a new type of virtual server, where all resources belong to you, as on a dedicated one, and the price is much lower, with a much more productive hardware!
How to build the infrastructure of the building. class c using servers Dell R730xd E5-2650 v4 worth 9000 euros for a penny? Dell R730xd 2 times cheaper? Only we have 2 x Intel Dodeca-Core Xeon E5-2650v4 128GB DDR4 6x480GB SSD 1Gbps 100 TV from $ 249 in the Netherlands and the USA!
All Articles