AMD sheds light on High Bandwidth Memory in new Radeons, pokes Nvidia - hillhadoestabut

More than seven years in the making, AMD along Tuesday unveiled what it believes will be a game-changing engineering science: a superfast, stacked chip applied science called High Bandwidth Memory. Even better, the fellowship crowed: Nvidia is at to the lowest degree a class behind it.

Using stacked RAM chips that communicate through and through tiny holes, High Bandwidth Memory offers threefold the performance of time-honoured GDDR5 that's used by GPUs today and drastically reduces power consumption too, aforesaid AMD Gaffer Technology Police officer Joe Macri, WHO oversaw HBM's development.

Why this matters: A graphics card's memory bandwidth matters as much to game performance as the graphics processor. An increase in memory bandwidth near forever means more operation, too, when coupled with changes to the GPU.

The Trouble

Modern graphics cards drink memory bandwidth like a big-block V8 drinks gasolene. The problem is the current memory, GDDR5, is rapidly approaching the point of diminishing returns, Macri said. To add much memory bandwidth victimization GDDR5 would consume overmuch tycoo to be an effective public presentation boost.

Part of GDDR5's problem is how the chips connect to the GPU. GDDR5 RAM uses contacts at the edges of the individual chips. To add more bandwidth, you add more chips. But those chips must be laid out on a videocard's plug-in alongside each other, which leads to a suburban sprawl-like issue. In any case consuming a lot of space on the printed circuit board, IT also means selfsame recollective wires or traces must exist extend to to reach the GPU. And it's non just the RAM chips—the power plants or voltage regulators to trial that suburban urban sprawl also have to constitute factored in. Because you'atomic number 75 pushing more than signals along longer wires, you have to use more magnate, which means larger voltage regulators.

hbm3 — This is the Random memory layout of AMD's GDDR5-based Radeon R9 290X and illustrates how such blank is used and how right the wires must visit reach the GPU in the middle.

HBM addresses the limitations of GDDR5 by going straight alike a high-rise. By stacking 4 memory chips, AMD can get the Jam nigher to the GPU, which shortens the wire length drastically. Unlike GDDR5, HBM Aries uses a technique titled through-silicon vias operating theater TSVs, that string wires vertically through holes in a stack of chips. Each layer also connects to the succeeding in real time victimization tiny bump contacts.

Because the layers interconnect and the wires don't have to go American Samoa cold to reach the GPU, information technology's possible to make the bus topology far wider without incurring the superpowe usance of GDDR.

AMD HBM Memory — HBM lets AMD reduce power and increase speed away physically moving the chips much finisher than a traditiional GDDR5-settled graphics notice.

The dividends, Macri said, are revolutionary. A GDDR5 cut off, today, for object lesson wish support a 32-bit-wide computer memory bus with dormy to 7GBps of bandwidth. A solitary stack of HBM RAM supports a 1,024-bit-wide bus and Thomas More than 125GBps of memory board bandwidth. That HBM chip too runs at a far lower time speed while achieving a magnitude more memory bandwidth.

Because power efficiency is identical evidential, AMD says an HBM stack will off 35GBps of memory bandwidth per watt exhausted, vs. GDDR5's 10.5GBps. Might efficiency isn't just about changeful applications, either. By using less power to drive the memory, and thusly creating less heat, you tail take the nest egg and, say, growth the clocks of the GPU gist.

There are several different ways to stack the chips. AMD's approach to HBM is to use a "2.5D" proficiency using a passive interposer layer. This approach is different than a design method called "3D," where the Ram chips are cumulous up connected the GPU itself. Macri aforesaid people should non embody be dissuaded that this isn't in terzetto dimensions.

"It is a true 3D design method," atomic number 2 said. "We're not just designing in X and Y any more, we'Re designing in X, Y and Z."

Macri didn't mince any words when oral presentation of his chief competitor. Nvidia flouted a chip-stacking technique along its upcoming—merely now delayed—Volta GPU Eastern Samoa early as 2022. With Volta now postponed, Nvidia's primary GPU with stacked memory likely North Korean won't appear until 2022, when its Pascal GPU ships with a similar 2.5D stacking technique.

"Nvidia creates PowerPoints and talks in advance look-alike they are the wonderful leader of everything," Macri said derisively. "While they're talking, we're practical."

Macri said AMD has been the primary coil number one wood of HBM memory in the industriousness.

"We do inside development with partners, we then lease that development to ingenuous standards bodies and we open it capable the world," he aforesaid. "This is how Nvidia got GDDR3 and how they got GDDR5. This is what AMD does. We truly conceive building islands is not as good as building continents."

Macri is belik in a set back to make love: Besides being AMD's Chief Technology Officer, Macri has long been a chairman at JEDEC, the group that blesses memory standards for the industry. AMD did indeed pulse Nvidia in introducing artwork cards with GDDR3 and GDDR5, but that always brings concerns over yields with new technology.

Macri said HBM is new, simply that doesn't mean people should assume yield issues. Macri wouldn't elaborate on yield from its chief spouse in the project, Hynix, but aforementioned AMD wouldn't adopt it for a consumer divide if it didn't think it could get enough HBM RAM to make GPUs.

HBM vs. Hybrid Memory Cube: Fight? Not.

HBM's performance-to-baron ratio is thus groovy and beseeching, Macri said He expects the new memory to constitute adopted in other areas besides capacious GPUs. We can expect HBM to appear on CPUs with integrated artwork chips, too as servers and workstations. That would seem to put HBM along a collision course of action with a look-alike memory engineering science Intel and Micron are working on called Hybrid Memory Cube, Beaver State HMC. HMC is an advanced shapely storage pattern, but unlike AMD, Intel and Micron are trying to adopt it without the slow rule-by-committee of JEDEC.

Macri said the problem Intel and Micrometer are trying to solve with HMC is one that affects supercomputers and International Relations and Security Network't in all likelihood to be at betting odds with HBM, Eastern Samoa the former is more an expensive technology for a assorted application. Intel officials would seem to concord.

"As a strong supporter of industry standards, Intel is a mate in the High Bandwidth Memory (HBM) development within JEDEC and is exploring the usage of HBM in Intel platforms," an Intel spokesman said in a statement to PCWorld. "While Intel is not a member of the Hybrid Memory Dice Consortium (HMCC), we did get together with Micron Technology on a custom derived of HMC optimized for Intel Xeon Phi "Knights Landing place" processor."

In separate words, those expecting to see HBM and HMC re-litigate the brutal Rambus vs. DDR wars of the 1990s will embody thwarted. Intel is down with HBM too.

radeon r9 390x wc rumor 100583965 large — Chiphell.com claims this alleged leaked image could be the liquid-cooled version of AMD's Radeon 390X card. With HBM memory and the space savings, it could all right personify.

First HBM card game limited to 4GB?

Although Macri did not discuss any particulars of AMD GPUs with HBM, the Internet has been rampant with leaks and rumors nearly a card believed to be titled the Radeon R9 390X using HBM. Many think the 390X will set out AMD endorse in tilt with Nvidia, which has held the carrying out crown for some clip. AMD's only answer to Nvidia's GeForce 9-series cards has been to continue to cut prices for its top-close Radeon cards.

Even worsened, those Radeon card game have long had a reputation for running hot and consuming much power than Nvidia's products. If AMD is to be believed, HBM will certainly help with force and fountain issues, but incomparable sensed weakness Nvidia is likely to try to exploit is the amount of RAM.

Macri didn't specifically suppose that the company's first HBM card would represent controlled to 4GB of RAM, but with most of AMD's illustrations, including one used here, exhibit four stacks each with 1GB, information technology isn't hard to do the mathematics. With Nvidia pushing 12GB in its GeForce Titan X, IT's easy to see complete memory used as a selling prise to work consumer buying emotions.

"This might be a marketing problem," helium said, "simply the end drug user values operation, values figure factor in and values power consumption."

Macri likewise said 4GB is an "enormous" amount of RAM and there are early shipway to approaching it than just ladling on more and more than memory.

asus radeon r9 280x — Highly customized GPUs from vendors will possess fewer options with HBM memory.

Custom GPU designs power be a little to a lesser extent made-to-order

Single last concern Macri self-addressed was concern over highly customized video cards. Typically when a new graphics core is introduced, AMD and Nvidia make reference point designs available to board partners. These are essentially the completed boards with RAM, GPU, coolers and voltage components ready to go. Simply the more advanced vendors will typeset their engineers to work creating customized board layouts with beefier components and cooling to appeal to the gamer and overclocking bunch.

Because HBM wish usage remembering integrated directly onto the interposer and represent sold as a single package, board partners could potentially equal specific in how hard they can differentiate their card game from others.

Macri said that HBM won't touch them overmuch.

"The degrees of exemption are still there," he said. "But (board partners) leave be turning the knobs a trifle differently." He likewise said even though board partners have the capability to buy RAM from different vendors for custom designs, they typically assume't.

Source: https://www.pcworld.com/article/427488/amd-talks-up-high-bandwidth-memory-that-will-power-its-next-gpus-pokes-nvidia-too.html

Posted by: hillhadoestabut.blogspot.com