Variations of “HBM2 is expensive” have floated the web since well before Vega’s launch – since Fiji, really, with the first wave of HBM – without many concrete numbers on that expression. AMD isn’t just using HBM2 because it’s “shiny” and sounds good in marketing, but because Vega architecture is bandwidth starved to a point of HBM being necessary. That’s an expensive necessity, unfortunately, and chews away at margins, but AMD really had no choice in the matter. The company’s standalone MSRP structure for Vega 56 positions it competitively with the GTX 1070, carrying comparable performance, memory capacity, and target retail price, assuming things calm down for the entire GPU market at some point. Given HBM2’s higher cost and Vega 56’s bigger die, that leaves little room for AMD to profit when compared to GDDR5 solutions. That’s what we’re exploring today, alongside why AMD had to use HBM2.
There are reasons that AMD went with HBM2, of course – we’ll talk about those later in the content. A lot of folks have asked why AMD can’t “just” use GDDR5 with Vega instead of HBM2, thinking that you just swap modules, but there are complications that make this impossible without a redesign of the memory controller. Vega is also bandwidth-starved to a point of complication, which we’ll walk through momentarily.
Let’s start with prices, then talk architectural requirements.
AMD’s Long-Term Play & Immediate Risk
AMD’s pricing structure for Vega uniquely leans on bundle packs to help improve the company’s value argument in a competitive market. MSRP is $400 on RX Vega 56, $500 on RX Vega 64, and an added $100 upcharge in exchange for two games and some instant discounts. AMD’s intention with this is to offer greater value to gamers, but clearly also will help the company increase margins and move more Ryzen parts, thereby recouping potentially low or negative margins on Vega. This is aided particularly with game bundles, where AIB partners pay AMD about $29 for the game codes, though that is often waived or offered in exchange for MDF. AMD also stated desire to stave off some mining purchases with increased bundle prices, as this would offset the value proposition of the card. Since the bundles are sold as standalone SKUs and can’t be broken by consumers (into parts), it seems that this is potentially an effective solution at keeping miners at bay.
This move is also mystified by complex economics surrounding potential long-term plays by AMD, with the company having already lost on a bet that HBM2 pricing would be cheaper by the time Vega rolled-out; in fact, price increased at least once within the past year, and Hynix ultimately failed to deliver on AMD’s demands in a timely fashion. In the meantime, AMD could have its outlook set on increased supply driving down cost to build Vega GPUs. This might mean taking a slight loss or running on slim margins for now, but hoping for a payoff down the line. Part of this hinges on Hynix potentially coming online with HBM2 at some point, which could help reduce the cost figures we’ll go over in this video. Ultimately, AMD also needs to reclaim gaming marketshare, and part of that reclamation process will be aided by gritting teeth through painfully slim margins or even losses at launch.
The Cost of HBM2
There are two major costs with a video card: The GPU die and the memory, with follow-up costs comprised of the VRM and, to a lesser extent, the cooler.
Let’s start with HBM2 and interposer pricing, as that’s what we’re most confident in. Speaking with David Kanter of Real World Tech, the analyst who broke news on Maxwell’s tile-based rasterization and who previously worked at Microprocessor Report, we received the following estimate: “The HBM2 memory is probably around $150, and the interposer and packaging should be $25.” We later compared this estimate with early rumors of HBM2 pricing and word from four vendors who spoke with GamersNexus independently, all of which were within $5-$10 of each other and Kanter’s estimate. This gave us high confidence in the numbers. Taking his $175 combined HBM2 + interposer figure, we’re nearly half-way to the MSRP of the Vega 56 card, with the rest of costs comprised of the VRM, GPU, and dime-a-dozen electrical components. It’d cost a “normal person,” for instance, about $45 to build the VRM on Vega – that’d include the $2.70 per-phase cost of the IRF6894s and IRF6811 hi- and lo-side DirectFETs, about $8.80 for all six of the IR3598 drivers, and roughly $4 on the IR35217 (from public sellers and datasheets). AMD is a large company and would receive volume discounts. Even as individuals, we could order 10,000 of these parts and drive that cost down, so these numbers are strictly to give an idea of what it’d cost you to build the VRM.
We’re not sure how many AMD ordered and aren’t going to speculate on what the company’s discount would be, but those numbers give an idea for what someone might pay if not a major corporation. This primarily helps explain why AMD opted for the same PCB and VRM on Vega: FE, Vega 64, and Vega 56, especially given the BIOS and power lock on V56. Although Vega can certainly benefit from the advanced VRM, the necessity of it lessens as we get to V56. The increased volume from V56 orders could offset cost across the entire product stack, to a point of it being cheaper to overbuild Vega 56 than to order two or three completely different sets of VRM components and PCBs.
Regardless, we’re at about $150 on HBM2 and $25 on the interposer, putting us around $175 cost for the memory system.
We did speak with numerous folks, including Kanter, on estimated GPU die pricing, but the estimates had a massive range and were ultimately just educated guesses. Without something more concrete to work with, we’re going to just stick to HBM2 and interposer pricing, as that’s the figure we know. GPU cost and yield cost are only really known by AMD and GlobalFoundries, at this point, so no point in working with total speculation.
The Cost of GDDR5
The next question is what GDDR5 costs. A recent DigiTimes report pegs GDDR5 at about $6.50 for an 8Gb module, though also shows pricing for August onward at $8.50 per module. With old pricing, that’s around $52 cost for an 8GB card, or $68 with new pricing. We do not presently know GDDR5X cost. This puts us at around 3x the cost for HBM2 which, even without factoring in yields or the large GPU die, shows why AMD’s margins are so thin on Vega. We also know that AMD is passing along its HBM2 cost to partners at roughly a 1:1 rate – they’re not upcharging it, which is what typically happens with GDDR. There’s no room to upcharge the HBM2 with Vega’s price target.
Ignoring GPU cost and cost of less significant components, like the VRM and cooler, we’re at $100-$130 more than 8GB of GDDR5 cost to build. This is also ignoring other costs, like incalculable R&D or packaging costs. Again: We’re just focusing on memory today.
Why AMD Had to Use HBM2
(Marketing image from SK Hynix)
Now that we know how much HBM2 costs, we need to talk about why AMD decided to use it. Like most of AMD’s hardware, the company is partly trying to make a long-term technological play in the market. This started with Fiji and has progressed through Vega.
There’s more to it, though. HBM2 critically allows AMD to run lower power consumption than GDDR5 would enable, given the Vega architecture.
Speaking with Buildzoid, we know that Vega: Frontier Edition’s 16GB HBM2 pulls 20W max, using a DMM to determine this consumption. This ignores the voltage controller’s 3.3v draw, but we’re still at 20W memory, and no more than an additional 10W for the controller – that’s less than 30W for the entire memory system on Vega: Frontier Edition.
We also know that an RX 480 uses 40-50W for its 8GB, which is already a significant increase in power consumption per-GB over Vega: FE. The RX 480 also has a memory bandwidth of 256GB/s with 8GB GDDR5, versus Vega 64’s 484GB/s. The result is increased bandwidth, the same capacity, and lower power consumption, but at higher cost to build. In order for an RX 480 to hypothetically reach similar bandwidth, power consumption would increase significantly. Buildzoid calculates that a hypothetical 384-bit GDDR5 bus on Polaris architecture would push 60-75W, and an imaginary 512-bit bus would do 80-100W. For this reason alone, HBM2 saves AMD from high power budget that would otherwise be spent solely on memory. This comes down to architectural decisions made years ago by AMD, which are most readily solved for with HBM2, as HBM2 provides greater bandwidth per watt than GDDR5. HBM is effectively a necessity to make Vega at least somewhat power efficient while keeping the higher memory bandwidth. Imagine Vega 56, 64, or FE drawing an additional 70-100W – the world wouldn’t have it, and it’d be among the hottest cards since the GTX 480 or R9 290X.
The Vega architecture is clearly starved by memory bandwidth, too: Overclocking HBM2 alone shows this, as its gains are greater than just core clock increases. AMD didn’t have another choice but to go with HBM2, even though costs would be roughly one-third on the memory. GDDR5 might be possible, but not without blowing power consumption through the roof or losing on performance by limiting bandwidth.
AMD provided GN with a statement pertaining to choices revolving around HBM2, which reads as follows:
“AMD chose HBM2 memory for Vega because this advanced memory technology has clear benefits on multiple fronts. HBM2 is a second-generation product that offers nearly twice the bandwidth per pin of first-generation HBM thanks to various refinements.
“As we noted in the Vega whitepaper, HBM2 offers over 3x the bandwidth per watt compared to GDDR5. Each stack of HBM2 has a wide, dedicated 1024-bit interface, allowing the memory devices to run at relatively low clock speeds while delivering tremendous bandwidth. Also, thanks to die stacking and the use of an interposer, Vega with HBM2 achieves a 75% smaller physical footprint for the GPU die plus memories versus a comparable GDDR5 solution.
“The combination of high bandwidth, excellent power efficiency, and a compact physical footprint made HBM2 a clear choice for Vega. We have no plans to step back to GDDR5.”
AMD ended up opting for two stacks of HBM2 on the current Vega cards, which limits its bandwidth to Fury X bandwidth (2x 1024-bit Vega vs. 4x 1024-bit Fury X), ultimately, but AMD does benefit in the bandwidth-per-watt category. That’s the crux of this decision.
Battling for Marketshare & Margin
As for cost, knowing that the memory system gets us up to nearly $200 as a starting point, it is inarguable that AMD has lower margins on Vega products than could be had with GDDR5 – but the company also didn’t have a choice but to use HBM2. NVidia forced AMD’s hand by dropping the 1080 Ti in March, followed by 1070 and 1080 MSRP reductions. That’s ignoring the current insane GPU pricing (which inflates the 1070s & V64s into overpriced oblivion) and just looking at MSRP, as that’s ultimately where the two companies battle under normal conditions. AMD and nVidia also do not see a dollar of the upcharge by distributors and retailers, so margins for either company are theoretically unimpacted by the inflated consumer-side pricing. That’s why, ultimately, we’re looking at MSRP – AMD and nVidia sell their product to the AIB partners for a cost which should theoretically be largely unchanging. AMD is able to make back some of these margins with bundle packs, where a pair of games can be sold to AIBs for ~$29, then to consumers for $100, or where Ryzen and motherboard parts help recoup margins. Each motherboard sold is another chipset moved, and Ryzen sales go completely to AMD. Either way, AMD has to increase its GPU marketshare, and fighting through early losses or slim margins is part of that. The long-term play is clearly hoping that increased demand and supply will lower cost to build, so it remains to be seen how that’ll play-out.
AMD’s investing a lot of effort to try and recoup some of the margins: Bundle packs are a major boon, either through direct cost of games sold or through accompanying Ryzen product sales, and reusing the same PCB & VRM further helps slim margins. This is particularly true with a low volume part like Vega: FE, as using the same board will help meet MOQ thresholds for discounts, aided by higher volume V64 and V56 parts. Without immediate production runs of significant quantities on each SKU, it makes more sense for AMD to reuse the VRM and board than to design a cheaper V56 board, as cost across all three SKUs is lowered with higher quantities. This particular move has the upshot of benefitting V56 consumers (though the benefit would be more relevant with unlocked BIOS), as you end up with a seriously overbuilt VRM for what’s needed. The VRM can easily handle 360W through the card and is more than the V56 will ever draw stock or with a 50% offset. We’ve even tested up to 406W successfully (with a 120% offset) through V56, though it’s probably inadvisable to do that for long-term use.
But that gives us a starting point, and helps to better contextualize what people mean when they say “HBM2 is expensive.” It is – about 3x more than GDDR5, give or take $20 and assuming high yield – and that’s rough for AMD’s profitability. That said, the company really couldn’t have reasonably gotten GDDR5 onto Vega’s present design without severe drawbacks elsewhere. It wouldn’t compete, and we’d be looking at another Polaris mid-range GPU or much higher power consumption. At less than 30W for 16GB of HBM2, GDDR5 just can’t compete with that power consumption under command of the Vega architecture. It’d be cheaper to make, but would require significantly higher power consumption or a smaller bus, neither of which is palatable. AMD isn’t using HBM2 to try and be the “good guy” by pushing technology; the company was in a tough spot, and had to call riskier shots out of necessity. Although it’d be nice if every GPU used HBM2, as it is objectively superior in bandwidth-per-watt, both AMD’s architecture and market position pressure the company into HBM adoption. HBM2 would benefit a 1070 by way of lower power consumption, but the 1070 doesn’t need HBM2 to get the performance that it does – the architecture is less bandwidth-hungry, and ultimately, nVidia isn’t in the same market position as AMD. In a 30-to-70 market, AMD has to make these more expensive plays in attempt to claw back marketshare, with hopes of enjoying better margins further down the generation.
Editorial: Steve Burke
Video: Andrew Coleman