High Bandwidth Memory (HBM) is what will be used in the GPUs of the future, and at this point the concept hardly needs an introduction. The first generation of HBM will be used in AMD Fiji, a GPU that is expected to come in the next few months.
AMD uses HBM1, Nvidia HBM2
Nvidia will use 2nd generation HBM that will enable its Pascal GPU to get to a whopping 32GB on the highest end card, 2.7 times more than the already impressive 12GB used on Titan X cards.
The 4-HI HBM1 has a 1024-bit interface, can handle two prefetch operations per IO and has a maximum bandwidth of 128GB per second. The tRC is 48nm, with tCCD of 2ns (1tCK), and the VDD voltage of 1.2V. For example GDDR5 has 1.35 to 1.5V and a top bandwidth of 28GB/s throughput per chip.
In addition, 4-Hi HBM1 16Gb (2GB per chip) and 8-Hi HBM1 32Gb (8GB per chip) is possible and HBM2 will get double the bandwidth and density. With first generation HBM memory it is at least theoretically possible to design a card with 8GB to 16GB of memory, assuming that the company would be using 4 HBM chips on an interposer, resulting in 512GB/s bandwidth on a four-chip HBM card.
Boosting density and capacity
HBM uses Trough Silicon VIA (TSV) technology, as it is stacking the memory cores on top of each other on the same base die. It is amazing that SK Hynix manages to make these layered interconnections trough the silicon. It’s one thing to read research papers about it, but seeing the finished product is something else. As you can see, HBM chips will be significantly smaller than GDDR5 and DDR3 chips.
The HMB1 acts as a stack of eight 2Gb chips, resulting in 16Gb (2GB) per chip, while the second generation HBM doubles the density to 32Gb (4GB) per chip, with 4-Hi HBM2 modules or even 64Gb 8GB with 8-Hi HBM2 memory. This is how Nvidia will be able to get 32GB with four chips on Pascal.
So if you do the math you can get 512GB with four HBM1 chips and 1024GB/s with four HBM2 second-generation chips, which means AMD might have a second generation Fiji with HBM2 and Nvidia will use the same technology on Pascal, sometime in 2016.
In any case, HBM represents the biggest change on the memory front in years. Coupled with new FinFET nodes, 2016 promises to be a very eventful year for the GPU industry.
The performance of the GM200-powered GTX Titan X is just absolutely amazing. There’s nothing bad about the Titan X when it comes to performance, at all. At 1080p, 1440p and 4K there are improvements across the board compared to the GTX 980, and it even keeps up with two of them in SLI.
The two buzzwords for 2015 are Non-Volatile Memory Express (NVMe) and 256-bit 3D NAND. NVMe is a set of commands that unbinds NAND from the limitations of the Advanced Host Controller Interface. AHCI was introduced as the register-level interface for SATA. When SATA was introduced, flash in the densities we have today wasn’t on the horizon. Back then, hard drives were going to rule for decades. Of course, their mechanical nature capped performance, limiting the utility of deep queue depths. SATA capped native command queuing at 32 queues of one command (far more than was needed). NVMe increases this limit to 64,000 queues, and each queue can sustain up to 64,000 commands.
NAND flash is advancing, too. Improvements in manufacturing technology already enabled the first 3D V-NAND from Samsung. IMFT will follow with 3D flash by mid-2015, and it’s rumored that we’ll see 256Gb densities. Overnight, 1TB SSDs will turn into 2TB SSDs. The manufacturing costs should be equal once the dust settles, so your wallet won’t suffer when it comes time to step up capacity.
We should note that the video is not of an actual Apple engineer but rather a parody of sorts. The video is basically an editing of a Spanish video with English subtitles overlaid on it. We have to say that it is pretty funny and our favorite was how the reason there was only one port was because Jony Ive forgot to mill the rest of the ports before showing it to Tim Cook.
The video was uploaded by Armando Ferreira and has close to 2 million views at this time of writing. If you haven’t seen it, check out the video above to get your Friday laughs in.
Maybe it ages me to admit it, but I tend to think of gaming laptops as thick, heavy slabs that nobody except the most devoted would ever want to remove from their desks. It’s a surprise to me, then, when a product like Asus’ ROG G501 comes along. This all-aluminum, 15.6″ notebook is ultra-slim, at eight-tenths of an inch (or 20.6 mm) thick, yet it packs one of Nvidia’s latest mobile graphics chips, a GTX 960M with 4GB of RAM.
Selecting the right hard drive for your laptop or PC just became a bit easier thanks to Mushkin. The introduction of the ECO2 series of SSD’s brings the price per gigabyte down to well below .50 cents, and it does so with reliability and a trusted brand name. With read speeds up to 550MB/sec and write speeds up to 530MB/sec you are getting a very competitive SSD at an excellent price that is suitable for all of your storage needs. The ECO2 is being introduced in 120, 240, and 480GB capacities and while you would think that at 480GB the price tag would be similarly inflated the 480GB version retails for a mere $159.99 making it the best SSD deal we have seen.
Now, though, Sony has brought the cassette back from the dead by unveiling a tape that can hold a whopping 148 gigabytes per square inch. If you can’t do the math, that’s 185 terabytes of total data. We’ll wait as you toss your iPod into the trash.
After plenty of discussion and plenty of benchmarks, we’re still not entirely sure what to make of MSI’s GT80 Titan SLI.
Priced at a lofty £3,500 and designed to be the be-all-and-end-all of gaming laptops, the system is as extreme as the price tag suggets and largely lives up to its billing. Dual GTX 980M GPUs deliver a massive amount of gaming potential, the storage array is absurdly quick, the sound system is alarmingly loud, and heck, there’s even a mechanical keyboard for gamers who won’t accept anything less. For performance purists, no other laptop comes close.
MSI has succeeded at throwing everything but the kitchen sink into an 18.4in form factor, but the no-holds-barred approach does result in unenviable trade-offs. Size and weight are considerable, fan noise can frustrate, battery life is poor, and given the lavish choice of GPUs, the GT80 Titan SLI is crying out for a high-res display.
That all brings us to what’s perhaps the most pressing concern: who would spend £3,500 on a big and bulky laptop? The killjoys among us will argue that it’s cheaper to buy a more powerful desktop, and you’d still have enough money left over to pick up a nice Ultrabook or two. But hey, someone out there is going to want ludicrous gaming performance in a laptop form factor. For those select few, MSI’s GT80 Titan SLI is worth getting excited about.
Expect the new MacBook to tip the scales at 2 pounds, and it measures 24% thinner compared to the 11” model, being at a mere 13.1mm. An Intel Core M runs proceedings from within, accompanied by 8GB RAM and a 256GB SSD with Apple promising “all-day battery life”. Expect this new MacBook model to be available from April 10 onward at $1,299 a pop.
Some early benchmarks of devices at MWC show just how speedy the Galaxy S6’s storage can be. It just destroys every currently available phone. Maybe you won’t even mind the lack of a microSD card when the internal storage is this fast.
After spending a total of 200 hours researching and testing more than 20 Wi-Fi routers, plus analyzing reader comments and feedback, the $100 TP-Link Archer C7 (v2) is the router we recommend for most people right now. This dual-band, three-stream wireless-ac router usually costs between $80 and $100—the same price as many older, slower routers. But unlike those slower routers, the C7 supports the fastest connections of every major device you can buy today (or already own).
With the launch of Samsung’s Galaxy S6 and S6 Edge, and HTC’s One M9, one of the main question on everyone’s mind was which device will perform better. After all, both of them come with industry leading chipsets on board, namely the Snapdragon 810 and the Exynos 7420. Samsung caused quite a stir around the mobile world when it became rumored that the Korean manufacturer was choosing to drop Qualcomm’s jewel in crown with one of its own in house SoCs. And it had every reason to do so. After all, the Exynos 7420 is manufactured on Samsung’s very own 14nm FinFET manufacturing process, giving it not only significant performance, but also power enhancements. So if you’re wondering how the two stack up against each other, and against Apple’s A8 (found in the iPhone), take a look at the benchmarks below to find out.
Moore’s law which is the observation that transistor density doubles every year, later amended to every couple of years, has slowed down dramatically in the past few years. Scaling Silicon transistors down has become increasingly difficult and expensive and at around 7nm it will prove to be downright impossible.
Digital computing which is what the entire world has relied on for the past several decades is based on one basic concept, on or off. The zeroes and ones in binary simply indicate if a signal is present or not. The fundamental flaw with Silicon transistors is that at the 7nm point the transistors sit so close to each other that an effect called quantum tunneling occurs. This effect unfortunately means that the transistor cannot reliably be turned off and for the most part will stay on.
What this means is that your binary code is not binary any more because it’s all made up of ones and no zeroes. Which in turn breaks the fundamental rule of digital computing. So the physical limitations of Silicon are very real and in fact insurmountable. One alternative is to find a material which can physically scale down past Silicon or can achieve faster switching speeds. The other is to rely on something other than electricity to achieve/read on or off states such as light.
The first option is more straight forward. Which is why the semiconductor industry has been researching alternative materials that are not only capable of scaling down past Silicon but can also be manufactured using similar techniques. There are dozens of different Silicon alternatives out there. Unfortunately each one has one or more significant challenges ahead of it.
However there’s one very promising short-term Silicon alternative that will most likely supersede Silicon for a few years. It’s a III-V semiconductor based on two compounds and four different elements. Indium gallium arsenide ( InGaAs ) and indium phosphide (InP). Imec, a research center tasked with finding the next thing after silicon, has already managed to fabricate FinFET transistors using InGaAs and InP on a 300mm 22nm Silicon wafer a year and a half ago. Imec is funded by Intel, IBM, TSMC, Samsung, Hynix and every other major semiconductor player with a fab that you can think of.
Samsung Electronics Starts Mass Producing The Industry’s First 128GB Universal Flash Storage, Almost Certainly Galaxy S6-Bound
These differences result in performance improvements across the board for UFS 2.0. Compared to eMMC 5.0, they are 1.4x faster at sequential reading, 1.66x at sequential writing, 2.71x at random reading (19000 Input Output Per Second vs 7000 IOPS), and 1.07 at random writing. The difference is even more staggering compared to MicroSD cards: sequential read and write speeds are more than tripled, while random read and write IOPS are multiplied by factors larger than 10.
The promise is that with UFS 2.0, consumers will be able to run multiple applications in the background, download and upload big files, and play massive games or UHD videos simultaneously, without any compromise on performance. That explains why Samsung will likely forego its MicroSD slot in the Galaxy S6 in favor of an embedded UFS 2.0 solution.
While it just started mass producing the 14nm FinFET chipsets, and showcasing the 10nm FinFET semiconductor technology at the Solid State Circuits Conference (ISSCC) in San Francisco this week, Samsung has outdone itself again by confirming that the Korean company has all the backbone to create chipsets to be as small as 5nm.
According to Samsung’s Kinam Kim, who confirmed “There are no fundamental difficulties until 5nm’. Furthermore, the Korean company has begun finding ways to shrink things even further to an insane 3.25nm level.
But before 3.25nm, the question was already raised by many in the conference as to what material will Samsung use to fabricate these chipsets? Intel has hinted during the same conference that silicon is not a viable option for chipsets below 7nm. Apparently, Intel itself has plan to use Indium Gallium Arsenide (InGaAs) to make chipsets with transistor size of 7nm and below.
But Samsung remained mum, holding on to their trade secret.
What we know is Samsung has already planned in using the 14nm FinFET technology based Exynos 7 processor for all of its upcoming smartphones and tablets; for example the Galaxy S6, which will take full advantage of the company’s first 14nm chipsets.
Samsung had been heavily investing in its semiconductor business to try competing with Qualcomm and Intel.
Called SSD-3000M, the product has just as much storage capacity as the name suggests: three thousand megabytes, or three terabytes.
That is a lot of space. Even on the hard disk drive market that capacity would be considered on the high side, on the consumer front.
Granted, 4 TB drives have been seen in the news a lot more, and there are now even 6 TB and 8 TB units, but 3 TB is a high capacity regardless. For a 2.5-inch SSD to possess such an amount of free bytes is remarkable.
Samsung’s has developed a new line of eMMC solutions that promises to accelerate the storage performance of next-gen mobile devices. The fingernail-sized memory cards are basically self-contained SSDs, and according to Samsung VP of Memory Marketing Jim Elliot, they’re the first based on the latest eMMC 5.1 standard. The spec is so fresh it’s not even listed on Jedec’s website.
Elliot says eMMC 5.1 adds command queuing, a feature that’s been around in SATA drives forever. Instead of executing commands in the order they arrive, queueing collects multiple commands and executes them in the order that makes the most sense. Performance should improve as a result, especially in multitasking scenarios. This Marvell presentation (PDF) from last year’s Flash Memory Summit indicates that the eMMC 5.1 queue is 32 commands deep, just like SATA’s Native Command Queuing implementation.
As one might expect, the new drives have better performance ratings than the eMMC 5.0 units Samsung announced in 2013. Sequential reads are unchanged at 250 MB/s, but writes are up from 90 MB/s to 125 MB/s. Random I/O rates have risen more dramatically, from 7k/7k IOps for reads/writes to 11k/13k IOps.
Storage capacities are unchanged at 16GB, 32GB, and 64GB. That probably means 128GB mobile devices will continue to be few and far between.
Samsung is “in the process of preparing to ship [its] first eMMC 5.1 products to some smartphone and tablet OEMs,” according to Elliot. The drives will presumably be used in Samsung’s own products, as well.
VESA has released a new Embedded DisplayPort standard that supports higher resolutions and a new panel segmentation system. The eDP 1.4a spec builds upon version 1.4, which dates back to 2013, and it’s expected to appear in systems by next year.
As one might expect, the new standard comes with a higher data rate. The link speed has been boosted to 8.1Gbps, giving the four-lane interface 32.4Gbps of total bandwidth. Support for version 1.1 of VESA’s Display Stream Compression Standard has also been added. Taken together, these upgrades enable 8K notebook displays with 7680×4320 pixel arrays.
The reality is that NVIDIA’s Maxwell architecture and the GeForce GTX 980 pretty much prove we’re not going to get much more performance out of 28nm manufacturing. However, a pair of 980s do get within striking distance of our ultimate goal, and it’s easy to suggest we’re probably just one generation off of having a perfect 4K gaming experience with dual-GPU, and two generations off of single-GPU. 2015 is going to see both GPU vendors finally graduate from 28nm, and if I were to hazard a guess, I’d say the 4K performance problem will largely be solved before the year is out.
Enter 2013. Some group of engineers at MediaTek must have gotten their hands on an Adderall prescription or something because in just one year— one single year— they were suddenly producing 28nm quad-core processors. Although they were celebrating their 1 Ghz single-core chipset and swimming in the fast network speeds of 3G just a year earlier, they were now producing multi-core chips with HSPA+ support as if they were born for it. Suddenly, MediaTek was now starting to be seen as a very legitimate potential threat to other semiconductor manufacturers. The obscure Taiwanese company went from producing years old technology that was long forgotten in the West to producing SoCs that were exceeding all expectations. Over a period of just six months, their GPUs tripled in clock frequency. A month later, their CPUs went up 40% in clock frequency too. But that was just the start.
Eight-core chips? Who the ____ wants or even needs that?
Then, almost out of no where, MediaTek went from chasing old technology to literally breaking records and leading the way for innovation. It was something that was expected for some time, just as a novelty, but no one was delivering on it. All of a sudden the underdog, MediaTek, jumped up and took the lead. Just one month after improving their last round of quad-core chipsets they decided to really make themselves stand out as more than just a follower. Introducing… the MT6592. The 2 Ghz octo-core processor with the Mali-450 GPU clocked at 700 Mhz. The smartphone world threw their hands up in amusement, amazement, disdain or blatant disgust— depending on who was reading the news. MediaTek, however, was celebrating their engineering achievement.
- Ancient cannibals used chillies and spices to ‘marinate’ their victims
- Bionic lens promises SUPERHUMAN sight three times better than 20/20
- Large Hadron Collider breaks atom smashing record
- Is the thunder god vine the key to beating obesity? Ancient chinese remedy led to a 45% decrease in body weight in obese mice
- Collaborative rendering reduces bandwidth for streaming games
- Are gays ‘born that way’? Most Americans now say yes, but science says no
- Brainprint technology could let you unlock devices with thoughts
- Natural sounds can help improve office productivity and brain power
- World’s oldest tools predate Homo by 700,000 YEARS could rewrite human history
- Bionic Foot Powered By The Brain
- Watching 3D Movies Might Improve Brain Power
- An Introduction to HBM – High Bandwidth Memory
- Blogs RO
- Famous Quotes
- Food recipises
- Health – Brain
- IT Hardware