HW News - RISC-V Core Designs, NVIDIA & Reviewers, 8GHz Intel CPU OC
Posted on December 22, 2020
If you’ve missed the YouTube uploads lately, we’d recommend taking a look at the RX 6900 XT from AMD, as well as our tear-down of the Lian Li Galahad AIO. Of course, we’d be remiss not to mention that we took Cyberpunk 2077 through our GPU bench and CPU bench (and an updated CPU bench with 1.05).
Regarding news, we have some interesting commentary following hardware and component shortages and how they’re affecting segments outside of the DIY scene. Seagate has also gone the route of custom, specialized silicon to address future storage needs. We’re also seeing China’s biggest chipmaker, SMIC, added to the DoD blacklist.
There’s more to cover, of course, so find the article and video embed below.
00:56 | The Saga of Hardware Unboxed & Nvidia
This one will remain exclusive to the embedded video as it was off-the-cuff.
05:52 | Seagate Designs Custom RISC-V Cores
Seagate recently announced that it has new in-house designed RISC-V cores aimed at accelerating storage performance and needs. Seagate broke the news at the RISC-V Summit that took place recently.
The cores, which have yet to be named or identified in any meaningful way, will eventually land in a pair of processors: one that is performance optimized, and another that is area-optimized. These new cores will no doubt be destined for storage controllers, and according to Seagate, will allow for better security and performance.
Seagate also noted the need to better address data in transit, and data on the edge (see: edge networking), as well as bolstering overall real-time processing power. Thus, Seagate decided on custom silicon.
According to Seagate, the performance optimized core will allow for “up to triple the performance for real-time, critical HDD workloads versus current solutions.” Meanwhile, the area-optimized design is “optimized both for footprint and power savings.” Seagate says its custom RISV-C silicon will be crucial in driving its future products, but didn’t explicitly mention what products the new cores would land in first.
07:13 | GPU Drought Extends to Data Center Segment
It seems Nvidia’s data center segment may be experiencing some form of the Great GPU Drought of 2020. Although, Nvidia is likely going to great lengths to lessen the blow to one of its most lucrative markets.
In a conference call, transcribed by SeekingAlpha, Nvidia’s Ian Buck, VP of Accelerated Computing Business Unit, fielded a number of questions from Wells Fargo’s Aaron Rakers. The call itself was actually very interesting, as it relates to Nvidia’s server and AI business, so it’s worth a look.
However, within the context of supply shortages, we arrive at one somewhat vague, but telling statement from Buck on the supply and demand for Nvidia’s A100 accelerator.
“I mean, it's going to take several months to catch up with some of the demand. I think the -- what's exciting is the sort of the interest and growth in both training and inference. We delivered a record number of quarter on T4 [sic], for example. And with the -- every time we introduce a new architecture, it's a game changer, right? So A100 is 20x better perf than V100, and with that comes a new wave of demand and interest in our products,” said Buck.
“So it will take several months to catch up with the demand, but it is always an exciting time to have on the platform refreshes that we experienced when you get the injection of new hardware,” Buck continued.
The Ampere-based A100 is Nvidia’s latest workhorse data center GPU, succeeding the V100. Nvidia recently expanded the A100 catalogue with an 80GB variant. While the GeForce side of Ampere is built on Samsung’s 8N process, the professional/data center side of Ampere lives on TSMC’s 7nm. As ever, TSMC is booked with AMD -- for both its RX 6800 series and Ryzen 5000 -- as well as SoCs for the new Xbox and PlayStation consoles. Not to mention Apple’s A14 Bionic, albeit that’s on 5nm.
10:58 | Component Shortages Also Affecting System Integrators
A report out of PCWorld confirms that it’s not only consumers having a hard time with hardware shortages.
Friend of the site and PCWorld’s Gordon Mah Ung spoke with “half a dozen PC vendors” off-record regarding supply constraints affecting their business.
According to Gordon’s interviews, one common theme among the smaller vendors is that bigger OEMs -- like the aforementioned HP and Dell -- are getting first claims at inventory, which isn’t exactly surprising. “It feels like we’re feeding off scraps,” one such vendor told PCWorld.
Regarding GPUs specifically, it seems smaller vendors have also been relegated to paying above MSRP for any stock they may be able to get. “We don’t even negotiate prices on GPUs anymore,” a vendor told PCWorld. On the upside, Mah Ung noted that he was told by some vendors that they were managing to receive RTX 30-series cards on a semi-regular basis. PCWorld’s interviews also highlight that the RTX 30-series cards are the hardest to get, with the RTX 3080 and 3090 being of particular rarity.
With CPUs, vendors have noted that AMD’s Ryzen 5000-series are equally hard to get. So much so, that vendors have tried to offset the losses by shipping systems with older Ryzen 3000 parts, which in turn, has driven up demand for specific Ryzen 3000 SKUs. Intel CPUs, for the time being at least, seem more readily available. According to the vendors Mah Ung spoke with, some believe this is because Intel has stronger supply chains in place. This isn’t news for Intel -- the company has maintained a stronghold in OEM, SI, and laptop distribution that has been relatively uninterrupted by AMD Ryzen.
As for why shortages are so persistent, it seems to be a smattering of reasons. There was LLLL already strong demand for these products, and the pandemic has had a unique effect on consumer spending in the entertainment space. Furthermore, the pandemic, as well as the US-China dispute, has taken a toll on logistics such as container ships and air freight.
In our own reporting, we’ve learned that some manufacturers have millions of dollars worth of parts stuck in ports, but that they can’t get containers on ships to send them around the world. Air freight isn’t an option for many of these, as it blows-out the ability to make any profit.
The vendors told PCWorld that they’ve seen demand “up 300 percent to 400 percent for desktop gaming machines.” Moreover, demand isn’t expected to subside any time soon, and shortages are likely to remain in place through early 2021. In addition, as many vendors are dealing with above MSRP prices and rising shipping costs, it’s likely some of those costs will get passed to customers.
Be sure to check out PCWorld’s report in its entirety, linked below.
19:51 | SMIC Gets Blacklisted
Unsurprisingly, China-based chipmaker SMIC has been put on The Department of Defense’s entity list, essentially blacklisting the company from American business. SMIC, much like Huawei, has been a target of the current administration, over alleged ties to the Chinese military and feared espionage. The DoD laid out its plans to add SMIC, as well as other Chinese companies, to its list last month.
SMIC (Semiconductor Manufacturing International Corporation) is China’s largest chipmaker, and it is part state-owned and part publicly listed. SMIC operates globally, and its US customer base includes Broadcom, Qualcomm, and Texas Instruments. SMIC also holds support from notable US investment groups, such as Vanguard Group and BlackRock Institutional Trust Company.
SMIC is essentially the linchpin in China’s domestic chip production ambitions, as it attempts to wean itself off of American technology. In the current climate, SMIC continues to find itself unable to get access to critical chip IP and manufacturing equipment. SMIC recently announced its N+1 process, and while it didn’t offer a transistor geometry, it’s expected to be comparable to Samsung’s 8nm process.
Source: https://uk.reuters.com/article/uk-usa-china-military-companies-idUKKBN28E07F
21:32 | AMD Will Continue Reference RX 6800-Series Card Production, for Now
Last month, AMD (via Scott Herkelman) stated that it would be winding down production of its reference RX 6800-series cards in early 2021. This in itself isn’t unusual; often this is the case once AIB partners get their designs off the ground and into retail channels. However, some users felt that AMD’s reference board production was a bit short-lived, especially as many still can’t get custom or reference cards yet.
Additionally, some users said they had purchased waterblocks in anticipation of being able to get a reference card, and were worried they wouldn’t be able to use it. Keep in mind that AMD still supplies a reference board to add-in board partners, so you can still potentially get a compatible PCB.
At any rate, after the outcry on social media, Herkelman and AMD did a 180, saying “We've just officially extended the reference design builds indefinitely due to popular demand. Thanks for the feedback,” on Twitter.
AMD’s RX 6800-series has gone the same way as Ryzen 5000, Nvidia’s RTX 30-series, and the new Xbox Series X|S and PlayStation 5 consoles, in that they’ve seen stock evaporate. So, AMD keeping its board production online for the foreseeable future is probably a good thing, at least until stock and demand normalizes a bit.
https://twitter.com/sherkelman/status/1336684121830137858
23:45 | Rumor: Intel Sapphire Rapids Could Use MCM Design
An image of what is purported to be a Sapphire Rapids Xeon-SP has been spotted over at ServeTheHome. The image is interesting, as it isn’t like the traditional Xeon parts we’ve seen. Instead, it appears to have two integrated heat spreaders fused into one, two identical sets of capacitors on the bottom of the chip, replete with all the necessary LGA pins separated into two halves. The IHS is marked with a “QTQ2 2.00GHz” stepping.
Sapphire Rapids and its Eagle Stream server platform are expected to succeed the Whitley platform, which is where Cooper Lake-SP and Ice Lake-SP belong. Sapphire Rapids is expected to be built on Intel’s 10nm SuperFin process, and will see the platform make use of the emergent DDR5, PCIe Gen 5, and CXL 1.1, among other new technologies.
It’s entirely possible Intel may be planning a chiplet/MCM-based approach for 10nm, instead of using a monolithic die. At this point, it’s well documented how chiplet-based designs can be deployed to better control costs and improve yields, among other things. As such, Intel has invested heavily in off package and off chip interconnects, such as Co-EMIB, Foveros, Omnidirectional Interconnect (ODI), and MDIO.
Source: https://forums.servethehome.com/index.php?threads/lga-4677.31075/
25:17 | AIDA64 Beta 6.30.5523 Lists New RTX 30-Series Cards
The newest beta version of AIDA 64 seems to have outed more RTX 30-series cards, as several new IDs are listed in the release notes. The new GPUs are as such:
- GeForce RTX 3050 (GA107)
- GeForce RTX 3060 (GA106)
- GeForce RTX 3060 Mobile (GA106M)
- GeForce RTX 3070 Mobile 16GB (GA104M)
- GeForce RTX 3070 Mobile (GA104M)
- GeForce RTX 3070 Mobile/Max-Q (GA104M)
- GeForce RTX 3080 Ti (GA102)
Interestingly, the listing shows three different configurations for the alleged RTX 3070 Mobile, one with a 16GB VRAM buffer at that. We haven’t heard much about what form Ampere is going to take in notebooks, but this listing seems to suggest that RTX 3070/3060-powered laptops could be imminent. Additionally, the release notes also add more credence to the rumored RTX 3080 Ti and entry-level RTX 3050.
Nvidia just introduced the RTX 3060 Ti (see our review), and hasn’t said much more about the non-Ti variant yet.
Source: https://twitter.com/KOMACHI_ENSAKA/status/1337557121626374147
26:42 | Newegg Backtracks on GPU Combo Policy
Just like with the Vega 56 launch, Newegg again started bundling new GPUs with other components (such as memory and motherboards) in an effort to combat bots and scalpers snatching up new graphics cards. This isn’t an unusual thing to do on its own, but Newegg implemented a caveat that if customers didn’t need the extra hardware, they could return it, according to this archived policy ( via reddit user HookEm2013).
However, the new policy has some updated language that now states that some combos or kits must be returned in their entirety. Going through the policy and reading comments on reddit, it seems that the policy isn’t retroactive. Assuming a purchase predates the new policy, it seems there shouldn’t be much trouble returning it.
Presumably the company ended up processing more RMAs than expected. The idea of tying new GPU purchases to bundles is debatable to begin with.
As HotHardware points out, some other retailers seem to be faring better with different policies.
“Other retailers have decent enough policies in place where people are not encountering issues. For example, Microcenter has apparently been taking IDs and limiting how many GPUs (or CPUs) people can buy in a month. Best Buy, although a hassle, requires in-person pickups for GPU orders. While both options do not necessarily work for Newegg, they could be adapted perhaps,” says HotHardware.
Source: https://hothardware.com/news/newegg-yanking-customers-around-with-gpu-purchases
28:04 | Celeron D 347 Crosses 8GHz Barrier
If you recall Intel’s NetBurst CPUs, you may recall their aptness for overclocking potential -- especially as it relates to extreme overclocking. Intel NetBurst-based Celerons spanned a few different iterations, such as Prescott-256 and Cedar Mill, with the Celeron D 347 belonging to the latter.
Overclocker Ivanqu0208 just recently managed an 8362 MHz (8.36GHz) overclock on the Celeron D 347, using an X38 chipset-based Asus P5E64 motherboard. As usual, LN2 was used. Interestingly, this isn’t the highest overclock a NetBurst CPU has hit.
The Celeron D 352 holds the record at 8543 MHz, as Tom’s Hardware points out.
Source:
https://www.tomshardware.com/news/celeron-d-347-overclocked-to-8-37-ghz
HWBot: https://hwbot.org/submission/4623615_ivanqu0208_cpu_frequency_celeron_d_347_8362.21_mhz
Editorial: Eric Hamilton
Video: Keegan Gallick, Andrew Coleman
Host: Steve Burke