The AMD Execution Thread [2020]

Status
Not open for further replies.

pharma

Legend
AMD to usurp Apple as TSMC's largest 7nm customer
January 3, 2020
Supply chain sources speaking to Taiwan's Apple Daily newspaper (nothing to do with Apple Computer Inc.) have indicated that AMD will become TSMC's largest 7nm customer as we traverse 2020. Furthermore, the 7nm production lines will be ramped up from 110,000 to 140,000 wafer starts per month (WPM) in H2 2020.
...
The main event affecting the 7nm share is Apple's 'departure', and AMD's doubling of orders. Apple Computer is said to be getting its upcoming A14 SoCs produced on TSMC's new 5nm process. Its A12 and A13 SoCs were built at 7nm. Apple gave notice that its 2020 iPhones were to be built upon 5nm processor tech almost a year ago. Accompanying Apple, Bitmain and Canaan are also transitioning to 5nm this year.

https://hexus.net/tech/news/industry/138323-amd-usurp-apple-tsmcs-largest-7nm-customer/
 
2nd generation AMD EPYC processors will power weather forecasting supercomputer
AMD has announced a new deal with the European Centre for Medium-Range Weather Forecasts (ECMWF) which will see the CPU giant supply 2nd generation EPYC processors to power one of the most advanced meteorological supercomputers in the world, in the Atos BullSequana XH2000.

AMD had a successful year in 2019 with many organisations turning to AMD for EPYC processors to power their latest supercomputer systems. HP Cray and the UK government’s ARCHER2 supercomputers have already committed to using AMD EPYC Rome CPUs for the foreseeable future and now the ECMWF will use 2nd generation EPYC processors in its new BullSequana XH2000 system.

The new Atos BullSequana XH2000 featuring AMD EPYC 7742 processors will be operational at the ECMWF in 2021. Once the system is up and running it will allow the ECMWF to run medium-range weather predictions at a high resolution of around 10km, meaning it will be able to produce reliable and advanced predictions about the intensity of incoming extreme weather events ahead of time.
https://www.kitguru.net/components/...will-power-weather-forecasting-supercomputer/
 
AMD vs. Intel CPU Market Share Q4 2019: EPYC and Ryzen Growth Decelerate, Mobile Ryzen Up

https://www.tomshardware.com/news/a...c-and-ryzen-growth-decelerate-mobile-ryzen-up

Mercury Research, the premiere CPU market share analyst firm used by the likes of AMD and Intel, released its fourth quarter 2019 market share numbers today, highlighting the ninth consecutive quarter of share gains for AMD across the desktop, mobile, and data center lineups.

However, unit share gains were surprisingly muted for both the desktop and the data center segments. Meanwhile, the company made strong progress in the mobility market, which is an encouraging sign given that laptops comprise roughly two-thirds of the consumer silicon market.

AMD now has 18.3% unit share of the desktop CPU market. The company gained 0.3 percentage points compared to last quarter, which is softer than expected given that the company launched its 3000-series products for both the mainstream and HEDT segments in time for the holiday shopping season. Notably, the fourth quarter has comprised AMD's strongest growth during each of the last three years.

AMD dominated Intel in Amazon's most popular CPU rankings during Black Friday and Cyber Monday, often occupying the top ten spots. However, this reminds us that the retail market is much smaller than the OEM/SI segment where Intel has historically ruled the roost.
 
AMD is offering cache/memory coherency (essentially unified memory) between the CPU and GPU over Infinity Fabric 3, that's a big factor for AMD's winning 2 systems out of 3.
 
AMD is offering cache/memory coherency (essentially unified memory) between the CPU and GPU over Infinity Fabric 3, that's a big factor for AMD's winning 2 systems out of 3.
Not sure if Frontier has it, though, or does AMD support it already or has it been confirmed for Zen 3 platform? (yes, I know, they haven't actually confirmed which Zen Frontier uses, but it's late for Zen 2 Epyc and too early for Zen 4 Epyc)
 
Curious. It seems El Captain will be based on standard CPU and standard GPU parts, no custom parts like the Frontier system.
While the Frontier system that is being installed in 2021 and put into production in 2022 is based on custom Epyc CPU and custom Radeon Instinct GPU motors, the contract with Lawrence Livermore specifies that El Capitan will be built with standard Epyc CPU and standard Radeon Instinct GPU parts, according to Forrest Norrod, general manager of the Datacenter and Embedded Systems Group at AMD. To be precise, El Capitan will deploy the future “Genoa” Epyc chips based on the Zen4 cores, which Norrod said would demonstrate “single core and multicore leadership” in performance and that these would be linked to Radeon Instinct GPU accelerators by a future Infinity Fabric 3.0 interconnect, which we know will be based on a future PCI-Express I/O transport.
...
But I think it’s important to understand that that the primary goal of this system is large scale physics simulation and not deep learning.”
https://www.nextplatform.com/2020/03/04/lawrence-livermore-to-surpass-2-exaflops-with-amd-compute/
 

Funny how the Exascale systems were proposed to only use 10MW of power but now a 2x Exascale system will now use 40MW or 20MW per Exascale.

It will be interesting to see what Nvidia/Mellanox will have out in the same 2023 time frame.
 
Funny how the Exascale systems were proposed to only use 10MW of power but now a 2x Exascale system will now use 40MW or 20MW per Exascale.
El Cap is circa 30MW.
Also we traded power for timelines, that's in itself kinda a win.
It will be interesting to see what Nvidia/Mellanox will have out in the same 2023 time frame.
Nothing resembling a capable host CPU for sure.
Cray will also continue kicking ass for shits and giggles.
 
I suppose this is the appropriate place for this? AMD has been a victim of IP theft
https://www.amd.com/en/corporate/data-update

TorrentFreak has interviewed supposed thief: https://torrentfreak.com/amd-uses-dmca-to-mitigate-massive-gpu-source-code-leak-200325/
WCCFTech claims to have interviewed people who have seen the files and have experience on verilog: https://wccftech.com/amd-stolen-gpu-ip-exclusive/
Their main points are: verilog (HDL) files of specific functions within GPUs, no big blueprints of the chips or anything like such, compatible only with AMDs internal design language > useless to others, screenshots suggest rest of the files are more of the same
 
Report: TSMC CoWoS Production Line at Full Capacity as Demand Increases
https://www.tomshardware.com/news/tsmc-cowos-production-full-capacity-demand
https://www.digitimes.com/news/a20200409PD203.html
According to DigiTimes, AMD, Nvidia, HiSilicon, Xilinx and Broadcom have placed orders for the tech, with demand for high-performance computing chips, high bandwidth memory (HBM)-powered AI accelerators and ASICs during the past two weeks.

Examples of CoWoS packaged silicon are AMD's AMD's Vega VII graphics cards, as well as Nvidia's V100 cards, which have HBM on the same silicon interposer where the GPU is. With the GPU and memory so close together, memory bandwidth is significantly higher on these chips compared to those using GDDR6 memory located elsewhere on the graphics card's PCB. Additionally, the PCB becomes much smaller.

Last month, TSMC announced its new CoWoS tech that packs a whopping 1700mm-squared of silicon onto a single interposer. With this change, the interconnect bandwidth was also increased to a staggering 2.7 TBps, which is a 2.7 times increase over TSMC's 2016 technology.

PS.
NVIDIA And AMD Just Bought Up All Excess Capacity At TSMC For Next Generation GPUs And CPUs
https://wccftech.com/nvidia-amd-ramped-orders-tsmc-2020/
https://www.digitimes.com/news/a20200423PD202.html
 
Last edited:
Nothing yet on AMDs Q1 results?

Well: https://ir.amd.com/static-files/57e8b0f9-29b8-420d-9f4e-a3aed607f456

Revenue: 1.79b, +40% yoy, -16% qoq
Net income: 162M, up 146 yoy, down 8 qoq
Margin: 46 %, up 5 pp yoy, up 1 pp qoq

Picks:
Computing and graphics net revenue is up 73% yoy, but down 13% qoq. Surprisingly, Enterprise, Embedded and SemiCustom is down 21% /25 % and operating at a loss of 26M. And this despite what I would call AMDs strongest portfolio ever.
 
Computing and graphics net revenue is up 73% yoy, but down 13% qoq. Surprisingly, Enterprise, Embedded and SemiCustom is down 21% /25 % and operating at a loss of 26M. And this despite what I would call AMDs strongest portfolio ever.

Actually it was expected, not surprising, because current gen consoles are practically dead from AMDs point of view and next gen hasn't ramped up yet, they commented on it too:
As expected, semi-custom product revenue was negligible in the quarter, as Sony and Microsoft, both reduced inventory in advance of next-generation console launches. We expect semi-custom revenue to increase in the second quarter and be heavily weighted towards the second-half of the year, as we ramp production to support the holiday launches of the new PlayStation 5 and Xbox Series X consoles.

Enterprise, Embedded and Semi-Custom segment revenue was $348 million, down 21% from $441 million in the prior year, due to the expected decline in semi-custom sales, partially offset by strong data center growth.

https://seekingalpha.com/article/43...-results-earnings-call-transcript?part=single
 
Actually it was expected, not surprising, because current gen consoles are practically dead from AMDs point of view and next gen hasn't ramped up yet, they commented on it too:




https://seekingalpha.com/article/43...-results-earnings-call-transcript?part=single

I think CarstenS's point was that AMD's position in the server market is stronger than it's ever been. But this market has a ton of inertia, so I suppose AMD's Enterprise, Embedded and Semi-Custom business unit may still be dominated by console chips.
 
Status
Not open for further replies.
Back
Top