Hardware stub

HW News - Threadripper 3 TDP, Epyc Supercomputer w/ 748,000 Cores, Aldi Gaming PC

Posted on October 21, 2019

Inspired by megastore compatriot Walmart, it seems Aldi now wants to sell a gaming PC to you alongside your groceries. Assuredly similar in spec, this week's news round-up also talks about the Archer 2 Supercomputer, which is probably equivalent to a few hundred thousand Aldi gaming computers. The Archer 2 will leverage about 748,000 cores built atop the Epyc processor lineup from AMD. More mainstream desktop-oriented news includes Intel's i3 chips potentially becoming more similar to i7s going forward, and PCIe Gen6 looking toward 2021.

Threadripper Leaks

GN received independent confirmation of non-compatibility leaks regarding Threadripper 3, but has also received information on the pinout, socket, chipset, and TDP for Threadripper 3 parts, as reported to AMD’s partners in design documents.

Information is detailed below:

  • The new socket is sTRX4
  • It is not backwards compatible with Threadripper 1000 and 2000 motherboards, and those parts are not forward compatible with the new sTRX4 motherboards
  • Socket is mechanically identical to SP3, but there are changes to the pins
  • Castle Peak on sTRX4 socket. CPUs will be 4-ch and DDR4, 64 Gen4 lanes
  • 4094 pins LGA
  • TRX40 chipset for supporting IO. “500-series” chipsets
  • Sharkstooth is the reference design board for HEDT Threadripper
  • Reference boards went out a few months ago
  • Tctl 95C will exhibit thermal throttling on the CPU
  • Up to 280W TDP and down to about 180-190W TDP. tCase is defined between 65 and 80 degrees Celsius for the TDP formula, depending on CPU group (A, B, C), with HSF theta_ca between .12 and.14. We’re avoiding giving hard numbers as we aren’t sure what AMD plants to identify sources.

Separately, rumors online this week indicated a 3960X 24-core CPU.

Rumor: Intel’s i3 Chips Will Gain Hyperthreading

Intel has historically segmented its features based on price, and hyperthreading has been no exception. Just last year, Intel caught flack for removing hyperthreading from i7 SKUs like the i7-9700K. Now, as AMD continues to ratchet up the competitive pressure, rumors suggest that Intel will bring hyperthreading to its lower-end i3 SKUs, where it continues to be the most vulnerable to AMD.

As found by well-known hardware leaker TUM_APISAK on Twitter, there’s a new submission in the SiSoft Sandra database that shows an i3-10100 with a 3.6 GHz base clock. Most interestingly, it’s a 4C/8T part. The i3-101000 is presumably part of Intel’s upcoming Comet Lake, based on 14nm +++ silicon. If this holds true, then Intel will be selling a Core i3 part with the same configuration that used to be limited to flagship i7 SKUs just a few short years ago.

Past rumors have also suggested that Comet Lake will equip i9 SKUs with up to 10 cores and 20 threads, as well as the possibility of new motherboards and chipsets.

Source: https://twitter.com/TUM_APISAK/status/1182640081623891969

Intel’s 10nm Desktop Plans Aren’t Dead

A rumor began to run rampant earlier this week, orginigating from German hardware outlet HardwareLUXX. The site had claimed to have reliable information from “insider circles” that Intel was going to kill its 10nm desktop plans, and opt to wait until 2022 for its 7nm CPUs. The rumor seemingly pegged lower than expected frequencies from the 10nm node that were ill-suited for desktop parts.

However, Intel has expectedly refuted the rumor, stating that “We continue to make great progress on 10nm, and our current roadmap of 10nm products includes desktop."

What seems to have given the rumor traction is how long Intel has taken to get 10nm to a point of volume shipping -- and it’s still only shipping mobile parts in the form of Ice Lake. Then there's the aforementioned Comet Lake, which is yet another iteration of 14nm and another derivative of Skylake. There’s also the rumored Rocket Lake, which is supposedly set to debut in 2021, based on yet more 14nm silicon. The defunct Cannon Lake doesn’t paint a pretty picture for Intel’s 10nm desktop ambitions, either.  

Ultimately, Intel has been tightlipped about what form 10nm desktop parts will take, and when they might land. For now, we’ll have to take Intel at its word and see what happens. The statement to Tom’s Hardware was decidedly ambiguous, and leaves a lot of weasel room for what happens. We don’t know, for instance, if 10nm desktop support will include HEDT, mainstream, or NUC parts, or just one of them.

Source: https://www.tomshardware.com/news/intel-yes-there-will-be-10nm-desktop-cpus

PCIe 6.0 Looks Good For 2021

We talked about a PCIe Gen5 demo last week, but that was a week ago, so no one cares about PCIe Gen5 anymore.

PCIe 6.0 was originally announced last June, just after the PCI-SIG group finalized the PCI 5.0 specification. However, at the time, the specification for PCIe 6.0 was little more than a draft. Now, it’s already seen significant progress, as the PCI-SIG has issued an update stating that PCIe 6.0 has officially entered version 0.3.

The technical details -- such as bandwidth, throughput, technology, etc.-- haven't changed since the original announcement. Rather, the announcement seems to serve as a mile marker for development, ensuring that the specification looks good for a 2021 finalization, as the PCI-SIG wants to avoid the long gestation period PCIe endured between 3.0 and 4.0.  

Source: https://www.anandtech.com/show/14985/pcie-60-dev-reaches-v03-ontrack-for-a-full-specification-in-2021

TSMC’s 16nm Lead Time Increases

We recently reported that TSMC had tripled its production lead time for 7nm chips, due to wildly popular demand. For reasons that aren’t readily apparent, it seems TSMC may be extending lead time for 16nm as well.

Per Digitimes’ report, even TSMC’s less advanced nodes are seeing increased demand and limited capacity. Digitimes also notes that “TSMC has already seen its 16nm chip supply fall short of demand.” TSMC’s 16nm node has been in production since 2013, so a sudden stretch in production and lead time is interesting.

It’s possible that the fab is simply overloaded, or there may have been a slight disruption in production; earlier this year TSMC had to scrap thousands of chips due to contaminated photoresist chemicals, for instance. TSMC is also embroiled in a legal dispute with foundry peer GlobalFoundries, although that likely isn’t hindering chip production -- yet.  

Source: https://www.digitimes.com/news/a20191014VL201.html

Aldi Wants To Sell You A Gaming PC With Your Groceries

You read that headline right. Aldi, purveyor of affordable groceries, now has a gaming PC to sell. The Medion Gaming Desktop PC as it’s called, will start at around £ 949.99, or $1,222. Admittedly, Walmart set the bar pretty low, so there’s plenty of room for Aldi.

It seems as if Aldi is partnering with Medion Erazer to offer gaming PCs on its website, as the gaming PCs aren’t in stores. Medion Erazer is owned by Lenovo, and also sells PCs at places like Costco. The Medion Gaming PC will come with an i5-9400, 16GB of DRR4-2666 memory, a 1GB PCIe SSD, and a Geforce RTX 2060.

There’s no indication as to what motherboard is being used, and there’s also no information on the PSU. It does come with a 3-year warranty.

Source: https://www.aldi.co.uk/medion-gaming-desktop-pc/p/700959312887900

Archer2 Supercomputer Will Utilize 11,696 AMD Epyc CPUs

In another supercomputer win for AMD, Epyc CPUs will be powering the new Archer2 supercomputer for UK Research and Innovation (UKRI). Archer2 will be built by Cray, and will succeed the original Archer supercomputer.

Archer2 will consist of 5,848 Shasta compute nodes, with each node boasting two Epyc Rome 64C CPUs, for a total of 11,696 CPUs. That’s 748,544 CPU cores clocked at 2.2 GHz, and almost 1.5 million threads. Archer2 is being built in the same area that houses the current Archer supercomputer, meaning there will be a period of downtime for the transition. Archer is set to go down on February 18th, 2020. Archer2 should boot up on May 6th, 2020. Below are the Archer2 specifications, as quoted from HPC Wire.

  • A peak performance estimated at ~ 28 PFLOP/s
  • System Design:
  • 5,848 compute nodes, each with dual AMD Rome 64 core CPUs at 2.2GHz, for 748,544 cores in total and 1.57 PBytes of total system memory
  • 23x Shasta Mountain direct liquid cooled cabinets
  • 5 PBytes of Lustre work storage in 4 file systems
  • 1 PByte all-flash Lustre BurstBuffer file system
  • 1+1 PByte home file system in Disaster Recovery configuration using NetApp FAS8200
  • Cray next-generation Slingshot 100Gbps network in a diameter-three dragonfly topology, consisting of 46 compute groups, 1 I/O group and 1 Service group
  • Shasta River racks for management and post processing
  • Test and Development System (TDS) platform, to be installed in advance
  • Collaboration platform with 4 x compute nodes attached to 16 x Next Generation AMD GPUs
  • Software stack:
  • Cray Programming Environment including optimizing compilers and libraries for the AMD Rome CPU
  • Cray Linux Environment optimized for the AMD CPU blade based on SLES 15
  • Shasta Software Stack
  • SLURM work load manager
  • CrayPat as profiler
  • GDB4HPC as debugger

Source: https://www.hpcwire.com/off-the-wire/cray-to-supply-archer2-supercomputer-with-amd-epyc-processors/?utm_source=rss&utm_medium=rss&utm_campaign=cray-to-supply-archer2-supercomputer-with-amd-epyc-processors

Editorial: Eric Hamilton
Host, Additional Reporting: Steve Burke
Video: Josh Svoboda, Andrew Coleman