AMD’s Bet On ARM Does Is Not Working

Buried in the AMD results was a note which seemed to hint that AMD’s plan to flog ARM based server chips was not going very well.

More Details Uncovered On AMD’s ZEN Cores

Our well informed industry sources have shared a few more details about the AMD’s 2016 Zen cores and now it appears that the architecture won’t use the shared FPU like Bulldozer.

The new Zen uses a SMT Hyperthreading just like Intel. They can process two threads at once with a Hyperthreaded core. AMD has told a special few that they are dropping the “core pair” approach that was a foundation of Bulldozer. This means that there will not be a shared FPU anymore.

Zen will use a scheduling model that is similar to Intel’s and it will use competitive hardware and simulation to define any needed scheduling or NUMA changes.

Two cores will still share the L3 cache but not the FPU. This because in 14nm there is enough space for the FPU inside of the Zen core and this approach might be faster.

We mentioned this in late April where we released a few details about the 16 core, 32 thread Zen based processor with Greenland based graphics stream processor.

Zen will apparently be ISA compatible with Haswell/Broadwell style of compute and the existing software will be compatible without requiring any programming changes.

Zen also focuses on a various compiler optimisation including GCC with target of SPECint v6 based score at common compiler settings and Microsoft Visual studio with target of parity of supported ISA features with Intel.

Benchmarking and performance compiler LLVM targets SPECint v6 rate score at performance compiler settings.

We cannot predict any instruction per clock (IPC improvement) over Intel Skylake, but it helps that Intel replaced Skylake with another 14nm processor in later part of 2016. If Zen makes to the market in 2016 AMD might have a fighting chance to narrow the performance gap between Intel greatest offerings.

Courtesy-Fud

AMD Coherent Data Reaches 100 GBs

After a lot of asking around, we can give you some actual numbers about the AMD’s coherent fabric.

The inter-connecting technology already sounded very promising, but now we have the actual number. The HSA, Heterogeneous System Architecture MCM (Multi Chip Module) that AMD is working on can give you almost seven times faster score than the traditional PCIe interface.

Our industry sources have confirmed that with 4 GMI (Global Memory Interconnect) links AMD’s CPU and GPU can talk at 100GB/s. the traditional PCIe 16X provides 15GB/s at about 500 ns latency. Data Fabric eliminates PCIe latency too.

AMD will be using this technology with the next gen Multi Chip module that packs a Zeppelin CPU (most likely packed with a bunch of ZEN cores) and a Greenland GPU that of course comes with super fast HBM (High Bandwidth Memory). The Greenland and HBM can communicate at 500 GB/s and can provide highest performance GPU with 4+ teraflops.

This new MCM package based chip will also talk with DDR4 3200 memory at 100GB/s speed making it quite attractive for the HSA computation oriented customers.

Source

AMD Misses Again

Fabless chipmaker AMD has come up with a mixed set of results for the second quarter. The company managed to make as much cash as the cocaine nose jobs of Wall Street expected, but missed revenue expectations.

In fact its revenues were below the psychologically important billion figure at $942 million.

We knew it was going to be bad. Last week we were warned that the results would be flat. The actual figure was $942m, an 8.5 per cent sequential decline and a 34.6 per cent drop from the same period a year ago.

As you might expect, there are some measures of this not being AMD’s fault. The company is almost entirely dependent on PC sales. Not only have these fallen but don’t look like they are going to pick up for a while.

AMD’s Computing and Graphics division reported revenue of $379m, which was down 54.2 per cent, year-on-year. Its operating loss was $147m, compared to a $6m operating loss for last year’s quarter.

Lisa Su, AMD president and CEO, in a statement said that strong sequential revenue growth in AMD’s enterprise, embedded, and semi-custom segment and channel business was not enough to offset near-term problems in its PC processor business.  This was  due to lower than expected consumer demand that impacted sales to OEMs, she said.

“We continue to execute our long-term strategy while we navigate the current market environment. Our focus is on developing leadership computing and graphics products capable of driving profitable share growth across our target markets,” she added.

In the semi-custom segment, AMD makes chips for video game consoles such as the Nintendo Wii U, Microsoft Xbox One, and Sony PlayStation 4 consoles. That segment did reasonably well, up 13 percent from the previous quarter but down 8 percent from a year ago.

But AMD’s core business of processors and graphics chips fell 29 percent from the previous quarter and 54 percent from a year ago. AMD said it had decreased sales to manufacturers of laptop computers.

Figures like this strap a large target on AMD’s back with a sign saying “take me over” but AMD is not predicting total doom yet.

For the third quarter, AMD expects revenue to increase 6 percent, plus or minus 3 percent, sequentially, which is a fairly conservative outlook given the fact that Windows 10 is expected to push a few sales its way.

AMD supplies chips to the Nintendo Wii U, Microsoft Xbox One, and Sony PlayStation 4 consoles and these seem to be going rather well.

Source

Did AMD Commit Fraud?

AMD must face claims that it committed securities fraud by hiding problems with the bungled 2011 launch of Llano that eventually led to a $100 million write-down, a US court has decided.

According to Techeye US District Judge Yvonne Gonzales Rogers said plaintiffs had a case that AMD officials misled them by stating in the spring of 2011 and will have to face a full trial.

The lawsuit was over the Llano chip, which AMD had claimed was “the most impressive processor in history.”

AMD originally said that the product launch would happen in the fourth quarter of 2010, sales of the Llano were delayed because of problems at the company’s chip manufacturing plant.

The then Chief Financial Officer Thomas Seifert told analysts on an April 2011 conference call that problems with chip production for the Llano were in the past, and that the company would have ample product for a launch in the second quarter.

Press officers for AMD continued to insist that there were no problems with supply, concealing the fact that it was only shipping Llanos to top-tier computer manufacturers because it did not have enough chips.

By the time AMD ramped up Llano shipments in late 2011, no one wanted them any more, leading to an inventory glut.
AMD disclosed in October 2012 that it was writing down $100 million of Llano inventory as not shiftable.

Shares fell nearly 74 percent from a peak of $8.35 in March 2012 to a low of $2.18 in October 2012 when the market learned the extent of the problems with the Llano launch.

Source

Intel Gives Exascale A Boost

Intel’s exascale computing efforts have received a boost with the extension of the company’s research collaboration with the Barcelona Supercomputing Center.

Begun in 2011 and now extended to September 2017, the Intel-BSC work is currently looking at scalability issues with parallel applications.

Karl Solchenbach, Intel’s director, Innovation Pathfinding Architecture Group in Europe said it was important to improve scalability of threaded applications on many core nodes through the OmpSs programming model.

The collaboration has developed a methodology to measure these effects separately. “An automatic tool not only provides a detailed analysis of performance inhibitors, but also it allows a projection to a higher number of nodes,” says Solchenbach.

BSC has been making HPC tools and given Intel an instrumentation package (Extrae), a performance data browser (Paraver), and a simulator (Dimemas) to play with.

Charlie Wuischpard, VP & GM High Performance Computing at Intel said that the Barcelona work is pretty big scale for Chipzilla.

“A major part of what we’re proposing going forward is work on many core architecture. Our roadmap is to continue to add more and more cores all the time.”

“Our Knights Landing product that is coming out will have 60 or more cores running at a slightly slower clock speed but give you vastly better performance,” he said.

Source

Intel Releases 16GB Xeon Phi

Intel has announced five Xeon Phi accelerators including a high density add-in card while upping memory capacity to 16GB.

Intel has managed to get its Xeon Phi accelerator cards to power the Tianhe-2 cluster to the summit of the Top 500 list, however the firm isn’t waiting around to bring out new products. At the International Supercomputing show, Intel extended its Xeon Phi range with five new products, all of which have more than one TFLOPS double precision floating point performance, and the Xeon Phi 7120P and 7120X cards, which have 16GB of GDDR5 memory.

Intel’s Xeon Phi 7120P and 7120X cards have peak double precision floating point performance of over 1.2 TFLOPS, with 352GB/s bandwidth to the 16GB of GDDR5 memory. The firm also updated its more modest Xeon Phi 3100 series with the 3120P and 3120A cards, both with more than one TFLOPS of double precision floating point performance and 6GB of GDDR5 memory with bandwidth of 240GB/s.

Intel has also brought out the Xeon Phi 5120D, a high density card that uses mini PCI-Express slots. The firm said that the Xeon Phi 5120D card offers double precision floating point performance of more than one TFLOPS and 8GB of GDDR5 memory with bandwidth greater than 300GB/s.

That Intel is concentrating on double precision floating point performance with its Xeon Phi accelerators highlights the firm’s focus on research rather than graphics rendering or workstation tasks. However the firm’s ability to pack 16GB into its Xeon Phi 7100 series cards is arguably the most important development, as larger locally addressable memory means higher resolution simulations.

Intel clearly seems to believe that there is significant money to be made in the high performance PC market, and despite early reservations from industry observers the firm seems to be ramping up its Xeon Phi range at a rate that will start to give rival GPGPU accelerator designer Nvidia cause for concern.

Source

Haswell Refresh Coming Next Year

Intel has been executing its tick tock strategy flawlessly since January 2006 and now there is some indication that we might see the first slip in 8 years come 2014. Intel’s latest roadmap claims that in 12 months from now, in Q2 2014 Haswell will be replaced by a “Haswell refresh”.

Haswell is a tock, a 22nm new architecture and Broadwell is supposed to be based on Haswell fundamentals, but shrunk to 14nm like a proper “tock”. In case that the Haswell refresh is a tweaked 22nm core, this would mean that after 7 years of execution and billions of investments in cutting edge fabrication processes, Intel would have to slow things down.

It is not certain what would happen to 2015 Skylake, a new 14nm architecture, or the 10nm Skymont that is supposed to be the shrink, but in case Broadwell gets pushed back by a year there is a big possibility that the whole roadmap would slip a year.

When it gets ready the Haswell refresh (possibly a disguise name for Broadwell ed.) is replacing Core i7, Core i5, Core i3, Pentium and Celeron based Haswell chips, some sooner rather than later.

The chipset responsible for Haswell refresh is already branded as Z97 and H97 in desktop versions replacing the Z87 and H87 boards proving that the socket are likely to continue existing at least through 2014. It will be interesting to see the developments and if Broadwell is really delayed or this is just game of words on Intel’s part.

Source

Does Haswell Need A Separate GPU?

Nvidia actually has a person with the catchy title of Chief Blogger and this person managed to get “an interview” with Rene Haas a VP and GM of computing products that currently takes care of Geforce mobile among other things.

Rene was asked to explain “why Gamers still need a discrete GPU with Haswell” and the answer is as logical as why do you use a seatbelt. Rene expects that Intel will continue to suck in graphics (our words not his ed.) and that that most popular games won’t play well on Haswell at standard resolution.

It seems that history really does repeat itself, as Intel had big claims for both Ivy Bridge and Sandy Bridge had been struggling to run new games of their time. Any serious gamers know that the answer is a proper discrete GPU. Haswell won’t change that, claims Rene.

It looks like Nvidia will have at least as many design wins as with Ivy Bridge, and Ivy Bridge was the record number or design wins for Nvidia. Rene claims that with this refresh Nvidia will have as much as 95 per cent of the gaming notebook which is nothing short of spectacular.

Rene also attacks Intel boldly claiming that “Their (Intel) comparison is misleading on a number of fronts.” Commenting the fact that Intel claims that GT3e will be faster than Geforce GT 650M. Intel based its claims on synthetic benchmarks, something that can be optimised, while Nvidia prefers real games, and even if GT3 wins again Geforce GT650, the new Geforce GT 750 is much faster than its predecessor and will have double the performance of GT3e in games.

Rene reminds us that GT3e is only available in top quad core mobile cores such as Core i7 4880QM that usually find their place in $3,000 notebooks. Rene tells customers that getting a Core i5 of Core i3 notebook with a better discrete GPU is the right way to get better gaming performance, although the vast majority of consumers already know that.

We remember that the last time we sat down with Rene, he said that when Intel gets faster with Integrated, Nvidia will simply gets even better with its corresponding low-end products and offers something faster. The cat and mouse game never ends.

Source

Intel Releases More Celeron CPUs

Intel added three curious ultra-low voltage chips to its official price list and their official designations is strange.

All three are 22nm parts and their product numbers are N2805, N2810 and N2910, which seems to indicate that they are Atoms, but they are listed in the ULV Celeron M section, reports CPU World. The top SKU features four cores with no hyperthreading, which means that it is probably based on the new Valley View M core.

The N2805 is a dual-core clocked at 1.46GHz, with a single megabyte of cache. The N2810 is also a dual-core, but it’s clocked at 2GHz, while the N2910 is the previously mentioned quad-core, with 2MB of cache and a clock speed of 1.6GHz. All of them are priced at $132, which sounds like way too much for Atom branded parts.

With Temash and Kabini just around the corner, Intel needs to step up its game in the low-end low-voltage market fast, but at this point it seems that AMD be the first to market and it will enjoy at least a few months on top. Even when Intel launches its first 22nm Atoms, it won’t have an easy time matching AMD’s price or performance.

Source

Next Page »