TSMC jumps to save the Moore’s law

Fudzilla readers should be familiar with TSMC but for people finding this over the search engines, all your iPhones, most Qualcomm Snapdragon’s, …

Fudzilla readers will be familiar with TSMC but for people searching for this, every iPhone, most Qualcomm Snapdragon’s, AMD’s 7nm Ryzen, EPYC, Radeon as well as Geforce / Tesla / Quadro are made there. For the record, TSMC stands for Taiwan Semiconductor Manufacturing Company.

Godfrey Cheng, who recently joined as the head of global marketing, wrote a nice blog on his vision of Moore’s Law. There has been a lot of discussion about how the “Law” is dead and how doubling the number of transistors every two years id history.

TSMC has a solution for its continuation as Godfrey concludes that instead of chasing the frequency and increasing the single transistor performance, the solution is in higher density. Industry opponents to Moore’s Law claim that you cannot continue to shrink the transistors any further and that there will be physical limitations.

To refresh your memory, Intel was stuck at 14nm for a few generations and now finally managed to get to 10nm while companies like Apple and Qualcomm, who are at the leading-edge technology for its phone chips used 10nm in 2017 / 2018 phones and 7nm in 2018/2019 or current generation phones.

To continue putting billions of chips on a very small 2D area, it was logical that after 7nm comes 5nm and so on. Apple’s 7nm A12 chip inside of the iPhone Xs has 6.9 billion transistors on the 83.27 mm2 area.

TSMC solution for the continuation of Moore’s Law

The solution called N5P is a process developed by the company to bring customers highest possible density with the best performance. Today’s workloads revolve around CPU, GPU related tasks and are heavily influenced by the enormous need for computation for the existing and the future AI workloads.

One of the big obstacles is to feed the CPU/GPU or an integrated circuit or SoC with enough data. The parallel processor is only good if the interconnect and memory interface can keep all the pipes busy and fed.

TSMC sees a solution in massive interposer and advanced packaging techniques. A few recent examples familiar to readers involve AMD Vega, Nvidia Tesla, and Xilinx Versal chips, all having the HBM 2 memory on the interposer. It is no coincidence that all of the above products are manufactured at TSMC.

Having memory on the same package as the CPU/GPU speeds up computation as the pipelines are getting the data at much higher speeds compared to traditional memory. HBM 2 supports speeds up to 460 GB/s with some products and Radeon VII, or Vega 20 has a total bandwidth of 1028 GB/s manufactured on TSMC’s 7FF (Fin Fet) process. Nvidia’s V100 Volta-based Tesla compute AI card has a total bandwidth of 900 GB/s, and any GPU/AI company wants even more bandwidth than that.

Godfrey is teasing the N5P node, and TSMC will likely give us a bit more details at the Hotchips Conference starting Tuesday, August 20 in the keynote session named “What Will the Next Node Offer Us?” by Dr. Philip Wong, VP Corporate Research, TSMC.

TSMC’s first shot at 5nm called N5 process started risk production in March 2019 and compared to N7 7nm volume production the N5 offers 80 percent more density, 15 percent more speed and 30 percent less power. The new eLVT transistors can offer speed gains as high as 25 percent.

The new N5P that is expected to enter the risk production next year should squeeze an additional seven percent in speed and 15 percent in power compared to N5.

System-level innovation

Godfrey did show one big tease – a TSMC’s chip on wafer or substrate technology with a massive 2500 mm2 interposer. The interposer bonds two massive 600 mm2 processors combined with eight HBM memory chips in 75 mm2 packages.

Vega 7 is 331 mm2 while the Tesla V100 is one of the biggest chips ever built with 815 mm² size and 21 billion transistors at 12nm. Bearing that in mind, TSMC has a solution for two 600 mm cores that will be able to accommodate dramatically more transistors in the same area in N5P or even in N5 5nm process.

2500 mm2 interposer and two 600mm2 chips and 8 HBM chips

The 2500 mm2 interposer with so many memory and dual chips would do a lot of AI-related calculations, and we all know that there are a lot of calculations needed for the ADAS and self-driving, medicine, new material research, fuel efficiency and anything that needs optimization including band and stock transactions.

Godfrey headshot

Godfrey who we know from ATI, AMD and most recently as an EVP of the market at Synaptics, has the chance to help TSMC get more much-needed exposure and coverage, as frankly, this is currently the best manufacturing facility especially when it comes to cutting edge technology including 7nm and 5nm. The 5nm process is just around the corner, and as a performance enthusiast, we cannot wait to learn more.

Related Posts:

  • No Related Posts

AMD Zen 3 design finalised on 10% faster 7nm+ node

TSMC’s 7nm+ would be able to offer a performance increase inline with AMD’s previous Ryzen refresh of around 10% according to the pure-play …

AMD Zen 2 only just launched and we’re already talking up Zen 3… the cycle never ends. But that’s great for us PC hardware enthusiasts, and AMD has officially announced that the Zen 3 architecture design has been finalised on the 7nm+ process node ready for its expected 2020 launch.

The confirmation comes from AMD’s recent 2nd gen EPYC launch event (via TechPowerUp). The slides suggest there will be no more intermediary steps in its architecture nomenclature, as has previously been the case – such as the shift from Zen to Zen+. It would appear the shift from 7nm to 7nm+ has been deemed worthy of the Zen 3 title. But while that may imply its a more significant step up on Zen 2 than Zen+ was to Zen, there has never been any mention of Zen 2+, so it’s not like anything appears to have changed over at the red camp.

TSMC’s 7nm+ would be able to offer a performance increase inline with AMD’s previous Ryzen refresh of around 10% according to the pure-play foundry. TSMC’s refined 7nm node, N7+, will utilise extreme ultraviolet lithography, or EUV. Using a short 13.5nm wavelength, EUV is intended to speed up semiconductor manufacturing by reducing the necessary masking and processes required by longer wavelengths to create today’s minuscule processors.

A Zen 2 refresh on the new node seems the likely direction for team red, instead leaving the big changes to Zen 4 in 2021. But we can’t say for sure just yet. Over at AMD’s EPYC launch event the company also confirmed the Zen 4 architecture is in design stage, but, seeing as AMD’s confirmed ‘leapfrogging’ design teams in the past, that’s to be expected.

引き続き開発中 pic.twitter.com/LN1JfMt93w

— Yusuke Ohara/大原 雄介 (@YusukeOhara) August 7, 2019

Zen 4 will likely utilise the 6nm process node: a refined variant of the 7nm process node. Yep, process node naming is often gibberish. While TSMC is well on its way to 5nm, the 6nm node may make for a better fit for AMD as it follows all the same design rules as the 7nm node it currently employs with Zen 2.

AMD’s backwards compatibility pledge also ends next year, which will likely see Zen 3 the last to utilise the AM4 socket and Zen 4 the first of a new socket cycle. Maybe it’s time, the AM4 socket has been with us since 2016.

Related Posts:

  • No Related Posts

AMD’s Next-Gen Zen 5 Core Shows Up In Chief Architect’s Profile, Post-2020 Chip Architecture For …

If AMD was to utilize TSMC’s latest 5nm node, also dubbed as N5, the company may expect an uplift of 80% in transistor density, 15% in overall …

AMD has just released their 7nm based Zen 2 core architecture but it looks like the development of their post-2020 cores is already underway. While we know about Zen 3 and Zen 4, it looks like their far-future Zen 5 architecture has once again shown up online which more or less confirms it for future generations of Ryzen, Threadripper and EPYC processors.

AMD Zen 5 Next-Gen Core For Ryzen, Threadripper, EPYC CPUs Confirmed Through Chief Architect’s Profile

AMD’s Zen 5 is a post-2020 core architecture that will be featured on future generation processors in the Ryzen, Threadripper and EPYC family. We cannot confirm if AMD will be using the same CPU naming conventions in 2020 and beyond but just like now, all processor families would feature the new architecture.

Related AMD’s Powerful Ryzen 7 2700X and 2700 CPUs Are a Budget System Builder’s Dream Thanks to Prime Day 2019

The AMD Zen 5 core was confirmed to be in development by AMD a while ago in their slides during the Zen+ launch. It has once again been confirmed by David Suggs, who’s the chief architect for the AMD Zen 2 and Zen 5 microprocessor cores. AMD has teams within the CPU department who are working on different Zen cores in parallel. David, in particular, has been the Chief architect behind the AMD Zen 2 core which recently made its debut with the Ryzen 3000 processors and also, the upcoming Zen 5 cores.

We know that AMD’s Zen 2 is based on the 7nm architecture and the next in line, Zen 3, would be based on the 7nm+ process and is on track for launch in 2020. The AMD Zen 4 cores are currently in-design and scheduled for launch in 2021. By this time, AMD is expected to jump on the 5nm process node which would allow the processor cores to be up to 80% more denser than what 7nm currently allows.

If AMD was to utilize TSMC’s latest 5nm node, also dubbed as N5, the company may expect an uplift of 80% in transistor density, 15% in overall performance and a 45% reduction in die area with their next-generation Zen based Ryzen series. Following is what TSMC has to say of their 5nm process node:

N5 has entered risk production in the first quarter, and we expect customer tape-outs starting this quarter and volume production ramp in the first half of 2020. With 1.8X logic density and 15% speed gain and an ARM A72 core compared with 7-nanometer, we believe our N5 technology is the most competitive in the industry.

With the best density, performance, power and the best transistor technology, we expect most of our customers who are using 7-nanometer today will adopt 5-nanometer. With N5, we are expanding our customer product portfolio and increasing our addressable market. Thus, we are confident that 5-nanometer will also be a large and long-lasting node for TSMC.

via TSMC

The Zen 5 core which should launch in between 2022-2023 can be based on a further optimized 5nm process which would allow for significant performance uplifts. Since there are no more “+” within the Zen family (e.g. Zen+), we should expect noticeable upgrades that would further lift IPC along with improved chiplet interconnects. AMD’s Zen 5 core would mark another breakthrough for AMD by the time it launches while their competition would still be relying on their in-house 7nm node.

Related Posts:

  • No Related Posts

AMD Radeon VII: The Best GPU for Cryptocurrency Mining?

According to cryptocurrency news outlet VoskCoin, which as “benchmarked the top coins to mine in 2019,” if you want “the most powerful GPU for …

The recently released AMD Radeon VII graphics card appears to be the latest champion of GPU crypto mining, reportedly able to achieve a hash rate of 90 MH/s (90 million hashes per second) without any overclocking when used for mining Ethereum (ETH).

Although AMD announced the card in January, it was not until early April that it started becoming available in retail channels. The Radeon VII, which is around 30% faster than AMD’s previous flagship, the RX Vega 64, is manufactured using a 7nm process, has 16GB of HBM2 memory, and has a memory bandwidth of 1TB/s.

According to cryptocurrency news outlet VoskCoin, which as “benchmarked the top coins to mine in 2019,” if you want “the most powerful GPU for mining ever,” the AMD Radeon VII is the graphics card to buy.

Here are the results obtained by VoskCoin when using the card for mining various popular cryptocurrencies:

Radeon VII mining hashrates stock settings (stock clocks below)

1136 mV 1801 Core 1000 Mem 0 PL 100% Fan

Ethereum — Ethash — Phoenix Miner 4.2c 90 Mh/s @ 319 Watts

Monero — Cryptonight R — SRB Miner 1.8.2 ~2970 H/s @ 245 Watts

Zcoin — MTP — sgminer 0.1.1 3.2 Mh/s @ 320 Watts

Grin — Cuckoo29 — Gminer 1.37 5.1 G/s @ 220 Watts

Beam — Equihash 150_5 — Gminer 1.37 24 Sol/s @ 240 Watts

Ravencoin — X16R — WildRig ~25 Mh/s @ 320 Watts

ProgPoW 30 Mh/s @ 360 Watts

On Monday (April 29), Technology news outlet Wccftech reported that, in stock configuration, RX Vega 64 only manages 32 MH/s, and that the Titan V only manages 69 MH/s. This is pretty impressive since the “Radeon VII is immensely cheaper, now available for $680 – $700, whereas the Titan V is within the $3000 range, and the Radeon VII achieves a higher hash rate per watt.”

Featured Image Courtesy of AMD


Related Posts:

  • No Related Posts

AMD Radeon VII is the new king of Ethereum cryptocurrency mining

The Titan V was once considered one of the very best GPUs for cryptocurrency mining, able to mine Ethereum twice as fast as the RX Vega 64.

The AMD Radeon VII, AMD’s recent high-end graphics card, isn’t just great at pushing pixels while playing games, as it appears to be the new king of Ethereum cryptocurrency mining when it comes to performance.

According to VoskCoin, a member of the Bitcointalk forum, the Radeon VII is able to achieve a hash rate of 90MH/s without any tweaking. As Wccftech points out, this is almost three times the performance of the AMD RX Vega 64, and handily beats the 69MH/s hash rate of Nvidia’s powerful Titan V graphics card.

If you’re happy to do some tweaking to the GPU, then the AMD Radeon VII will perform even better, with a hash rate of between 90MH/s and 100MH/s.

These impressive results are in part thanks to the improved memory bandwidth of the new card, and the Radeon VII comes with a memory bandwidth of 1TB/s and 16GB of HBM2 (High Bandwidth Memory), whereas the older RX Vega 64 has a memory bandwidth of 484GB/s and 8GB of HBM2.

Perhaps most importantly of all, the Radeon VII is a more powe-efficient card, which brings running costs down when you’re using it to mine every hour of the day.

Clash of the Titans

The Titan V was once considered one of the very best GPUs for cryptocurrency mining, able to mine Ethereum twice as fast as the RX Vega 64.

The fact that the AMD Radeon VII breezes past Nvidia’s GPU when it comes to Ethereum mining is noteworthy because even though the Titan V is now a few years old, it still costs near $2,999 (around £2,200, AU$4,000), compared to the Radeon VII, which costs $699 (£699, around AU$980).

A much lower cost, for much better results, makes the Radeon VII a far better purchase for budding miners. While the popularity of cryptocurrency mining has died down of late, the release of this card, which offers such good results for such a reasonable price, could convince many people to fire up their old mining rigs again.

Check out VoskCoin’s video review of the Radeon VII’s mining capabilities below.

Related Posts:

  • No Related Posts