This past week was slammed for us. We posted reviews of the AMD Ryzen 5 5600X, Ryzen 7 5800X, Ryzen 9 5900X, and Ryzen 9 5900X, all available on our YouTube channel. In the time since, we’ve been working on Ryzen memory benchmarks, including an upcoming piece featuring Wendell of Level1Techs and Buildzoid of AHOC. That piece will focus on ranks, channels, and Zen 3’s newly exaggerated behaviors with regard to interleaving and scaling. More on that soon. For hardware news this week, the big story is a GN-exclusive about NVIDIA MSRP targets and BOM cost suggestions for an RTX 2060-style replacement.
02:23 | Partners on NVIDIA MSRP Target: “We’d Be Bankrupt”
One of NVIDIA’s next upcoming GPUs will target price levels of the previous RTX 2060 and GTX 1660 Ti cards and currently has internal cost targets that, partners tell us, are unrealistic to hit while maintaining the estimated MSRP. The name of the card could still change, but the goal is something in the $250 to $300 price class. To give some perspective on how things work for NVIDIA’s board partners, it goes like this:
Early on, partners are given a technical specs sheet that provides a suggested BOM cost for each part in the product. That’d be a suggested BOM cost for line items of PCB, heatsink-fan, and of manufacturing and assembly. Partners generally know far better than NVIDIA or AMD what the appropriate BOM costs are for each of these parts, but the companies try to help well ahead of launch to ensure certain price points can be met. Of course, the biggest cost is always the GPU and the memory, which are sold as a bundled package to the partners. It is possible for partners to source their own memory supply, but because the volume is so much lower than crowd-sourcing it through the GPU maker, it’s typically bundled together. This also helps NVIDIA obscure the true GPU silicon cost since it’s harder to separate the combined price.
Currently, NVIDIA’s goal to hit an MSRP of $250 to $300 is predicated on the usage of a GPU heatsink-fan that costs between $4 and $5. GN has received the exact pricing from partners, but in order to protect our sources, we’ve slightly obfuscated the range so that NVIDIA can’t pinpoint the leak by looking at exact dollars and cents values. For perspective, we’re told that a $4 to $5 cooler would get you an Intel stock-cooler style solution, commonly called a “flower cooler” by the supply factories we’ve toured in China, except applied to a GPU. We were told by partners that their high-end models, often just below the flagship, cost around $50 for the cooling solution. That includes the backplate, the shroud, the RGB LEDs, the thermal pads (this cost can be high if using good pads), the past, the metal finstack, any present vapor chambers, and the tooling. Specifically, the tooling cost is what balloons the total cooler cost, but more on that in a moment. A low-end cooler as found in existing 1650-class GPUs costs about $9 for the absolute cheapest ones on the market, with a $13 BOM cost for the high-end GTX 1650 cards. For perspective, a $9 GPU cooler doesn’t usually cool the memory or VRM and only cools the GPU. Board partners explained to us that the lower-end cards might make them $1 here and there, but otherwise, it was described by one partner as a “wild west for the low-end cards. It’s hard to even say what the margins are because they’re always on and off promotions and don’t have a strict MSRP, so it’s more fluid.”
The BOM cost provided to GN by partners includes thermal pads, screws, sometimes LEDs, the shroud, the backplate (which is about $1 for an aluminum plate), the fans, and the heatsink. The fans cost about $1 to $1.40 each for PWM fans with custom blades, whereas a low-end PC case fan might cost about $0.70.
None of these complaints include the PCB or assembly cost. For perspective, assembly on an RTX 2060-class GPU with a mid-range cooler is typically in the range of $14. This relates to assembly and often tooling, depending on the partner. Sometimes tooling is counted in the cooler cost, sometimes it’s counted in manufacturing, but it’s ultimately divided out against the order quantity.
In asking the partners what they expected from us when providing all of this information, the answer was that they want NVIDIA to either set a higher MSRP that’s realistic -- in other words, something they can actually hit while making a good product -- or for NVIDIA to take a margin cut and sell the GPU and RAM to the partners for cheaper. This obviously indicates a separate motive for the board partners.
Partner motives here benefit the customer by way of seeking pressure on NVIDIA to bring MSRP to meet higher, realistic expectations, because the partners are concerned that customers will become upset at a lack of cards at MSRP for the next RTX 30 card. We recognize that partners might not be giving us the complete picture when asking NVIDIA to reduce its margins, so we next reached-out to some of our contacts in factories in China to ask about the actual sale price to manufacturers.
Our understanding is that, after speaking with factories, the cooler BOM prices quoted to GN from board partners have to include tooling cost divided out by the number of products being made in almost all instances. Tooling can easily cost tens of thousands of dollars, depending on the amount of pre-production tools are made, and is a major investment. The tools are reusable basically forever -- at least, for as long as the life of these shortly lived products -- but their high cost means high volume is needed to make-up the cost. Tools are what’s installed in the large stamp machines and other machines that we’ve shown in our factory tours. Cooling factories think that the price of $4 to $5 isn’t realistic for a dual-fan GPU cooler unless removing the backplate, reducing the fin density significantly, reducing the heatpipe count or removing them, using less copper, and using cheaper thermal interfaces.
We then looped-back with the partners with our new information, and were told that they’d have to offer mail-in rebates to hit NVIDIA MSRP targets for the 2060-class replacement at present. This is because MIRs have a redemption rate of 30-40%, and so a $20 MIR really is only reducing the effective global cost of the card by $7 per unit, not $20 per unit.
Ultimately, it sounds like tricks will have to be played with MDF and rebates in order to hit MSRP. We’ll have to follow-up after the launch to see if NVIDIA ended up changing its pricing to benefit the partners, and thus make it possible for partners to hit MSRP without MSRP just being an unattainable goal. AIBs told us that, worst case, they’d make short-lived models to hit MSRP and then switch only to the better products.
“If we sold only that, we’d be bankrupt,” one told us, saying that the current guidelines from NVIDIA are not realistic.
Source: GamersNexus exclusive
13:43 | AMD Ryzen 5000 Availability
This one is just in the video -- but it's timestamped!
17:07 | PCI SIG Release PCIe 6.0 Specification, Version 0.7
A quick update on the brewing PCIe 6.0 specification, as the PCI-SIG has officially released its PCIe 6.0, version 0.7 specification to members. This should be the last milestone ahead of a full ratification in 2021 (although there will be a version 0.9 before hitting 1.0), as this version has finalized bandwidth, electrical specifications, signaling, etc.
PCIe 6.0 will double previous Gen5 bandwidth to 63 GT/s, and PCIe will continue to rely on PAM4 signaling. PCIe 6.0 will also make use of low-latency forward error correction (FEC), and will be fully backwards compatible with PCIe 1.x through 5.0.
Bear in mind that, like DRAM, interface and bus technologies have a long gestation period. PCI-SIG released the final PCIe 5.0 spec in May of 2019, and consumers won’t see PCIe 5.0 capable hardware until, at soonest, the second half of 2021.
One of the more interesting pieces of hardware announced recently (that isn’t a CPU or GPU) is a new 420mm variant in Arctic’s Liquid Freezer II line up. We looked at the Arctic Freezer II 280, which was heavily requested from our readers/viewers, and we ended up rather impressed with it, despite some assembly difficulty. Nonetheless, we ended up naming it the best CPU cooler for Ryzen.
So, it’s interesting that Arctic is adding a 420mm radiator option, which may be appealing to those with the case compatibility to mount a 420mm radiator. Aside from the increased radiator size, the other feature to note is the integrated 40mm fan that resides in the pump block. This fan is aimed at adding additional cooling to the surrounding VRM components.
Arctic is still touting the use of its in-house developed pump solution, which you can learn more about in our previous Arctic Freezer II Tear Down video. Both the pump and the VRM fan are PWM-controlled, and the cold plate is your standard copper, micro-skivved affair. Arctic uses its own MX-4 thermal compound for the cold plate. As users have likely deduced, the radiator uses three 140mm fans. This unit should sell for around ~$140, as EU pricing is currently at €120.
As usual, Mercury Research has provided the x86 market share numbers, which show AMD snatching some of its biggest market share numbers since 2007.
These numbers represent quarter 3 2020 and were taken before the launch of Ryzen 5000. AMD makes up 22.4% of the x86 market share as a whole. For AMD, this is an increase of 4.1 points over Q2 and 6.3 points over Q3 2019.
In the x86 desktop space, AMD is showing it’s 12th consecutive quarter of growth -- with a .9 point quarterly increase bringing them to 20.1% of the desktop market. This is a 2.1 point increase YoY. For notebooks, AMD broke their record for notebook market share, now at 20.2%. This is up from their previous record of 19.9% which was Q2 2020.
This is impressive growth for AMD, who now has their strongest presence in the desktop space since 2013, and their largest presence in the overall x86 market since 2007.
The server market, as usual, gets a little muddy. AMD historically bases its share number on IDC’s research, which only captures the 1P and 2P market, whereas Mercury Research captures all x86-class servers. That said, Mercury shows AMD sitting at 6.6%, marking a 0.8 point increase QoQ and a 2.3 point increase YoY.
Massachusetts Expands Right to Repair Law, New Precedent
While not strictly hardware news, right to repair is an important measure that affects all consumers in some way or another. Furthermore, unless you’re new here, you likely know that the right to repair is something of extreme importance to us at GN. It’s something we support and follow closely.
To that end, Massachusetts has long been an important battleground for right to repair. It was Massuchestts’ 2012 right to repair legislation that gave way to The Motor Vehicle Owners' Right to Repair Act, which eventually forced the automobile industry -- and its various coalitions and organizations -- to sign a Memorandum of Understanding, acknowledging that they would support the law in all 50 states.
Now, in the face of the deluge of personal and telemetry-related data on drivers, generated by their cars, Massachusetts just passed a very important new law. Question 1, or the “Right to Repair Law” Vehicle Data Access Requirement Initiative, states that independent repair shops are permitted access to wirelessly transmitted repair and diagnostic data that has since been funneled directly to manufacturers. This new law is set to take effect with model year 2022 vehicles.
iFixit’s CEO commented: “Modern cars can send maintenance information directly to the manufacturers, cutting out local mechanics. Question 1 makes sure that consumers can continue to fix their own vehicles, or get them fixed at the shop of their choice.”
The war over data generated by cars has been brewing for some time. It’s the typical money versus privacy trope: Car makers want to control and monetize the data, and privacy and right to repair advocates rightly take issue with that. This new law could help galvanize another nationwide effort to crack open car data, and pry it away from manufacturers. Separately, it could very well have knock-on effects for digital right to repair intaives, such as for smartphones, consoles, laptops, etc.
Arm recently took the wraps of what we believe is its first CPU aimed directly at laptops, the Arm Cortex-A78C. This comes during Apple’s transition to custom Arm silicon, and in increasing interest from Microsoft and its Windows 10 on Arm systems.
Arm as much as teased that it was looking to dip its toes into the water with x86-competing offerings when it announced its Cortex X Custom Program and the Cortex X-1. The CXC and X1 are borne out of the idea of building chips beyond Arm’s traditional ecosystem and PPA (Performance Per Area) requirements, building custom chips that are scalable to different form factors and larger devices.
While the Cortex-A78C CPU isn’t itself derived from CXC -- rather, it’s a derivative of Arm’s most recent Cortex A-series IP -- it demonstrates that Arm seems serious about punching ever higher with its compute performance.
The Arm Cortex-A78 supported configurations of 4 little cores and 4 big cores. Now, the Cortex-A78C will support 8 big CPU cores with its “octacore” configurations. The Cortex-A78C continues Arm’s heterogeneous, multi-core computing trend with configurations scaling up to eight “big” cores. The Cortex-A78C also uses a noticeably bigger 8MB L3 cache, aimed at improving performance in data-dense applications. The Cortex-A78C also introduces Pointer Authentication (PAC) which ARM says will “minimize the attack surface to ensure data on the device is kept secure.” PAC will reduce Return-Orientated-Programming (ROP) and Jump-Orientated-Programming (JOP) attacks by 60 and 40 percent respectively.
Next-Gen Consoles Purchases Will Be Different This Year
Due to growing Covid-19 risks, Sony has already stated that no PS5 consoles will be available in stores on launch day. However, Best Buy has gone as far as not permitting customers in stores on launch day or throughout the holiday season to buy consoles in person. Instead, purchase will be online only. Although, customers are still able to pick up their order via curbside pick up. Target beat Sony to punch, stating on Twitter that the new consoles would only be available in stores to fulfill online orders.
Wal-Mart has also joined Best Buy in this practice, stating that its inventory of next-gen consoles will only be available online. We fully expect other retailers will commit to this in an attempt to avoid long lines and crowds during a global pandemic. If you insist on having a new console this year, plan accordingly.
Microsoft has made no formal statement, but for Best Buy, Target, and Wal-mart the policy applies to both consoles.
Much to the chagrin of vintage Windows users, it seems Microsoft is set to kill the classic Control Panel menu, seemingly in favor of its Settings app for Windows 10.
In the most recent October 2020 Windows update, it seems that clicking on the “System” setting within the Control Panel points users to the Settings app. Microsoft is also taking measures to prevent previous shortcuts and workaround for accessing retired Control Panel pages, as well. For whatever reason, Microsoft has decided to dismantle the Control Panel menu piece by piece, and if preview builds are anything to go by, the “Programs and Features” page will be the next part of the Control Panel to get the axe.