Quote:Update: The story initially mentioned that the $10,000 per card from three cards and up was an actual store policy, and it has been updated to reflect its nature as a deterrent instead.

Quote:If Vega 56 is able to stick near the 1070’s price, it’s AMD’s strongest argument from the Vega line. The biggest downside is the boosted power consumption, but if that’s not a concern to you, Vega 56 is a good buy if assuming a similar price between them. Prices are so volatile right now that we’ll refrain from hard numbers, and just suggest checking that the cards are relatively close. We’d strongly encourage solving for thermals with an aftermarket cooler or a board partner card, then overclocking. Vega 56 can outmatch or equal Vega 64 with the right mods, including powerplay tables and BIOS mods. For these gaming workloads, the only reason Vega 56 would underperform versus Vega 64 is AMD’s power limit, which is higher on V64. You can fix that with a BIOS flash or registry mod.

As for the shaders, it looks like there’s not a big difference for the games we tested. There’s probably an application out there that likes the extra shaders, but for gamers, we’d say hard pass on Vega 64 and strongly consider Vega 56 as a highly modifiable counter.

Quote:XFX and Sapphire confirmed that they both have custom boards in the works, but they could not say when they might be ready. PowerColor said that it will have its own custom cards, with mass production scheduled for the beginning of November, but it hasn't yet received the DRAM it needs. (VisionTek didn’t immediately reply to our queries about their future offerings.)

AMD also has partnerships with Asus, Gigabyte, and MSI to build Radeon graphics cards, but these three companies don’t have exclusive deals with AMD. As such, they aren’t driven by necessity and have the luxury of choosing which components to support. We spoke with all three companies, and their responses indicated that their support for the Vega architecture is less definitive than AMD’s exclusive partners.

We already knew that Asus is on board with Vega. The company announced in August that it would be releasing a pair of ROG Strix Vega cards with Asus’ custom cooling solution. Asus confirmed that those cards are still coming, although the release date has been pushed back from September to early October.

Although a Gigabyte rep said it’s likely that the company would be producing a custom Vega card, they would not or could not confirm with 100% certainty that it will. If it does, we likely won’t see it until the end of the year, or later.

MSI’s response surprised us. The company traditionally offers re-engineered graphics cards with custom PCB designs for all high-end GPU platforms, but it appears to be skipping the Vega lineup. A company representative told us that MSI “won’t be making a custom card anytime soon,” but could offer no additional information.

So what gives? Sources tell us that there is too much variance in the quality of the chips AMD is providing. AIB partners are unable to figure out a stable overclocked GPU frequency that works for all cards, and therefore cannot provide any sort of warranty on factory-tuned cards. Further, there continues to be discrepancies between the temperatures the GPU is reporting and what AIB partners are finding in actual measurements. This is true of the actual GPU and the capacitors below the GPU. We have some follow-up testing that will reveal more about these issues.

Finally, as we reported last month, there have been issues due to the different packages for Vega, making it difficult to efficiently mass produce custom Vega cards. We were seeing Vega with molded and unmolded packages, which we noted impacted package height. We were even seeing a third package--we assume, using SK hynix HBM. As we wrote then:
...
Generally speaking, AIB partners seem optimistic about shipping Vega cards in 2017, and some pointed out that custom Polaris cards came a couple months after the reference card launch. By that timing, we should be seeing some custom Vega cards at the end of September, or at least in October. We’re not getting a strong feeling that will be the case, however.

We’ve reached out to AMD for comment, but the company didn’t immediately reply.

Quote:Yes, this is the third post today about AMD introducing multi GPU support for RX Vega with their Crimson ReLive 17.9.2 beta drivers but it had to be made. First up, the caveats- we were only able to test the driver on a CrossFire setup involving one RX Vega 64 and one RX Vega 56 GPU so results with two of the same may differ. Secondly, these are beta drivers so there is a level of lenience here I am willing to afford to AMD. That said, the driver which came with its own announcement and internal results had to show something good and this means showing good scaling across multiple games.
...
Enabling CrossFire with GPUs that are based off the same die and family but with a different shader count has been something AMD has done for a few generations now, and we continue to appreciate this remaining an option. However, this is no excuse for the performance we are seeing here, be it a driver issue or otherwise. As can be seen from the chart above, a lot of games not only show poor scaling relative to a single RX Vega 64 but actually negative scaling. The negative scaling can potentially be a result of Vega 56 slowing down the Vega 64 card here, but then we are back to zero to minimal scaling again. Of the four games that do, three were in AMD's results chart that went around earlier today as well. This is really disappointing performance, and to make it worse we also faced visual artifacts and display corruption in a few games including Witcher 3 (pictured), Prey and Dishonored 2. We also tried having RX Vega 56 be the primary card to see if that changes anything, and the results were within error margins more often than not with a few results actually being higher (5-10%) than with RX Vega 64 as the primary card. There was also a blue screen that greeted us during the switch, so do not attempt to do so often.

There is another interesting train of thought departing this station of results. AMD has had CrossFire profiles for the previous Polaris (and older) architecture GPUs for most of the games tested here, and they showed consistent, if not great, scaling with those older cards. But based on the inconsistent scaling across game titles here, it appears that AMD may have to create new CrossFire profiles for all games with their RX Vega GPUs based off the Vega microarchitecture. This merits more analysis and testing, but we are definitely curious whether this is indeed the case and, if so, what is the reason behind it knowing that Vega is an evolution of GCN, and not a complete departure from it.

Quote:I’m not going to pretend to have a secret source inside AMD on this one, but it’s hard to look at Vega and not wonder if HBM was a fundamentally bad call. The company’s molding issues are directly related to the height of the HBM2 memory. Meanwhile, it took AMD over two years to launch Vega. That’s the longest gap AMD or ATI has ever gone between high-end GPU refreshes, which previously took 12 – 16 months at most. HBM2’s rollout has been slower than anticipated in general, particularly the highest speed memory.

We don’t know for certain that HBM is the culprit here, but it certainly seems like the most obvious place to look. Because there’s always a considerable lag between when a GPU design project kicks off and when the final product tapes out, it’s entirely possible that AMD was too far along in the design process to start over when HBM2’s growing pains became clear.

The big signal here will be whether AMD’s future GPUs are based on GDDR6. If HBM2 was the culprit for the problems Vega seems to have, AMD won’t keep using it. AIBs are still expected to bring custom Vega cards to market in general, but those samples may not be in-market until closer to the holiday season.

• Retail sample aftermarket Vega 64 from ASUS
  
• Theoretically only 2–7% faster than reference
  
• In practice, with stock voltage his sample OCed worse than his stock card, even with 17.8.2 drivers that aren't affected by a bug that harms OCing
  
• Undervolting helps with noise reduction and OCing
  
• The issue with OCing is the power limit
  
• Manual OCing doesn't work very well, automatic OCing is the way to go
  

Quote:There’s some scaling in 3DMark Firestrike GT1, which is poly and tessellation intensive, and scaling in Unigine synthetic benchmarks. Even when there is scaling at the more realistic upper-end of performance, though, it’s not much – we’re talking 1-3% for an extra $100-$150. Not at all worth it, and often not replicable in gaming scenarios. Again, there are likely compute applications and some very specifically-made games that could benefit from the CU increase, but 97% of the performance comes down to clocks – if not more.

Given how easy it is to flash V56 and overclock, we’d recommend just going that route. Save the money, OC Vega 56, and walk away with more money and functionally equivalent performance.

Quote:Apparently, Gigabyte has received a new batch of AMD RX Vega GPUs just in time for the mid-October expected release window of RX Vega custom cards, and are already at work on a Gaming OC custom version of the graphics card with a dual fan configuration (likely a revised WindForce 2X cooler). The card is expected to feature an output configuration of 3x HDMI + 3x DisplayPort, and in Videocardz image comparison, occupies slightly more space in our usual three dimensions than the AMD reference design (as expected).

Quote:8x MSAA was used in all configurations, since "the game isn't all that demanding". Being it demanding or not, the fact is that AMD's solutions are one-upping their NVIDIA counterparts in almost every price-bracket in the 1920 x 1080 and 2560 x 1440 resultions, and not only by average framerates, but by minimum framerates as well. This really does seem to be a scenario where AMD's DX 12 dominance over NVIDIA comes to play - where in CPU-limited scenarios, AMD's implementation of DX 12 allows their graphics cards to improve substantially. So much so, in fact, that even AMD's RX 580 graphics card delivers higher minimum frame-rates than NVIDIA's almighty GTX 1080 Ti. AMD's lead over NVIDIA declines somewhat on 2560 x 1440, and even further at 4K (3840 x 2160). In 4K, however, we still see AMD's RX Vega 56 equaling NVIDIA's GTX 1080. Computerbase.de contacted NVIDIA, who told them they were seeing correct performance for the green team's graphics cards, so this doesn't seem to be just an unoptimized fluke. However, these results are tremendously different from typical gaming workloads on these graphics cards, as you can see from the below TPU graph, taken from our Vega 64 review.

Quote:Consulting the work of our friends over at Techgage, it looks like Rob Williams reported similar results back in August, so that hasn’t changed much with the 17.10.1 driver push. We did find uplift in Superposition, but the trouble is that this uplift begins to emerge primarily when the card is beginning to struggle for other reasons – like frequency, shader count, or ROP limitations. It’s difficult to fill an 8GB framebuffer, so AMD’s claims of 50% uplift in AVG FPS may make more sense or better come to fruition were a 4GB card to exist. At this time, we do not have a means to create a 4GB framebuffer, so we can’t validate those claims in a direct comparison. We have some ideas, but they may not work. We’ll see.

In the interim, there’s really not much reason to disable HBCC, and it occasionally provides a 0.5-4.0% performance increase. Limited system memory would be a reason, of course. It might be worthwhile to enable HBCC and just leave it in the background, hoping occasional uplift will emerge. We can only speak for the applications which we’ve tested, naturally, and there’s potential for HBCC uplift in production applications (untested) or memory hog games (user mods for Skyrim and FO4, for instance, are often VRAM pigs). In the event you encounter one such game, HBCC enablement could help; of course, if stability or system RAM issues are encountered, there’s obviously not much loss to disabling HBCC, either.

As it stands now, we’re not presently able to observe appreciable performance uplift in any of our tested games, with one exception being the synthetic applications. We observed repeatable uplift in Superposition (~4%), with similarly repeatable (but insignificant) ~1-1.4% uplift in Firestrike. That’s probably good news for competitive benchmarkers, it’s just not meaningful for most users.

Let us know if you find an application where this technology is better leveraged. We may attempt some other applications in the future, but will probably wait for more driver updates to push.

Quote:There’s not a huge difference, at the end of the day, but there is a quality control variation present. Patterns drawn from this initial sample size would indicate more potential for vapor chamber quality differences than anything else – but we have more coming soon, now that we just got more cards in. The contact issues seen above are resolved largely by thermal paste, and the vapor chamber’s benefits more than make-up for its contact pressure variations, which are normal for a vapor chamber (those cylindrical shapes would happen with every vapor chamber, as the cylinders are used internally to prevent the chamber from being crushed). Just make sure there’s thermal paste in there -- nothing different from any other GPU, really. Thermally, we have not yet seen a major difference in packaging performance, but we also haven’t yet shown any packages with the earlier reported 40um lower HBM2. Again, more on that very soon, but it’ll take time to process the next round.

So far, from these cards tested, there’s not a lot to be worried about. The initial reports from Tom's and Videocardz got a lot of interest, but were never looked into beyond that. Thus far, from a small sample size, there is variation. Sure, there are some seeming quality control concerns (that also correlate with our earlier samples, so it may be resolved on later runs), but nothing that’ll break the cards or even perform appreciably worse.

Quote:Apparently, the new AMD RX 500X series is bringing some extra performance to laptops via nothing more than an OEM-exclusive rebadge. It's not a new SKU or product for AMD's AIB partners to play with; it's an OEM-only product that seems to bring AMD's mobile solutions - such as the RX560M - up to par with the company's desktop GPUs. So, a higher performing RX 560M will be called the RX 560X (with a 73 MHz core clock increase, mind you) in some mobile solution. Lending credence to this report is that the product page for Acer's own Nitro 5 laptop, for example, has already updated its graphics card listing from an RX 560 to the RX 560X.

Quote:AMD's David Wang also announced that the company is striving to produce a new graphics product every year for the next three years, which bodes well for a consumer graphics market that has become somewhat stagnant in terms of recently released high-end graphics cards.

AMD is currently sampling the 7nm Vega GPU to its partners and will launch it to the general market in the second half of 2018. That's a full quarter ahead of expectations. The card comes bearing the fifth-gen GCN microarchitecture, but it will bring many of the benefits borne of the 7nm process. In addition, true to the GPUs' stated mission, the card has a number of optimizations specifically for AI workloads.
...
AMD's new tactic extends the fabric outside of the GPU to speed peer-to-peer (P2P) communication between GPUs. This approach is similar to Nvidia's NVLink implementation, so it purportedly reduces latency and boosts throughput between graphics cards. We've seen several other P2P implementations take root in the broader component landscape, such as P2P fabrics that speed communication between storage devices and GPUs, but it appears the industry is stratifying into proprietary solutions.

In either case, it's logical to expect the Infinity Fabric to eventually extend to communication between the CPU and GPU, which could provide AMD yet another advantage as the only producer of both x86 processors and GPUs. Interconnects are increasingly becoming more of a strategic asset as companies transition to heterogeneous computing architectures, and it's a positive sign that AMD's Infinity Fabric continues to pay dividends.
...
AMD originally planned to release 12nm GPUs, which it announced last year, but made the strategic decision to remove those products from the roadmap and skip to the 7nm process instead. AMD has 7nm Navi up next on its roadmap, followed by new graphics cards with a 7nm+ process that should arrive before the end of 2020.
...
Neural networks are becoming increasingly complex, and, as a byproduct, voluminous, so AI-driven architectures increasingly require more memory capacity to keep pace. AMD's 32GB of HBM2 should put the company on par with Nvidia's recent adjustment that boosts the Tesla V100 up to 32GB, but we aren't sure if AMD will also offer versions with 16GB of HBM2. That could be an asset given the current sky-high pricing for HBM2 memory. AMD didn't reveal specifications or pricing for the new cards or a definitive timeline for mainstream graphics cards.

Quote:One such slide that created quite some flutter reads that "Vega" shipments are over 10 times greater than those of the "previous generation."

Normally you'd assume the previous-generation of "Vega" to be "Polaris," since we're talking about the architecture, and not an implementation of it (eg: "Vega 10" or "Raven Ridge," etc.). AMD later, at its post event round-table, clarified that it was referring to "Fiji," or the chip that went into building the Radeon R9 Fury X, R9 Nano, etc., and comparing its sales with that of products based on the "Vega 10" silicon. Growth in shipments of "Vega" based graphics cards is triggered by the crypto-mining industry, and for all intents and purposes, AMD considers the "Vega 10" silicon to be a commercial success.

Quote:The device ID 66A0 has been never seen before and is sufficiently far away from Vega 10 (687F) to indicate that AMD treats Vega 20 as a really new product and not as a minor revision. We expect Vega 20 to be used on a Radeon Instinct card first, with 32 GB of HBM2 memory and 4096 shaders, a professional workstation card called "Radeon Pro Vega 20" could also be possible, as indicated by other leaks, but I'm skeptic whether AMD really wants to put the "20" from the GPU codename into the official marketing name, going against what they used for "Vega 64".

The second card listed is Vega 12, for which very little is known. Its device ID 69A0 is also from a new, unused value range. It is rumored that Vega 12 is targeted at lower-end, mobile markets, possible to offer an alternative to Polaris. We also wouldn't be surprised to see improved power management on Vega 12, to cater better for mobile devices, which require higher energy efficiency. Please note that Vega 12 does not indicate the number of CUs (like on Vega 64 for instance), but rather represents the internal GPU codename, like Vega 10.

Quote:A close inspection of the latest AMDGPU Linux driver includes PCI-Express link speed definitions for PCI-Express gen 4.0, which offers 256 Gbps of bandwidth per direction at x16 bus width, double that of PCI-Express gen 3.0. "Vega 20" got its first PCIe gen 4.0 support confirmation from a leak slide that surfaced around CES 2018. AMD "Vega" architecture slides from last year hinted at a Q3/Q4 launch of the first "Vega 20" based product. The same slide also hinted that the next-generation EPYC processor, which we know are "Zen 2" based and not "Zen+," could feature PCI-Express gen 4.0 root-complexes. Since EPYC chips are multi-chip modules, it could also hint at the likelihood of PCIe gen 4.0 on "Zen 2" based 3rd generation Ryzen processor family.

Quote:AMD is reportedly prioritizing its first 7 nanometer silicon fabrication allocations to two chips - "Rome," and "Vega 20." Rome, as you'll recall, is the first CPU die based on the company's "Zen 2" architecture, which will build the company's 2nd generation EPYC enterprise processors. "Vega 20," on the other hand, could be the world's first 7 nm GPU.

"Vega 20" is not a mere die-shrink of the "Vega 10" GPU die to the 7 nm node. For starters, it is flanked by four HBM2 memory stacks, confirming it will feature a wider memory interface, and support for up to 32 GB of memory. AMD at its Computex event confirmed that "Vega 20" will build Radeon Instinct and Radeon Pro graphics cards, and that it has no plans to bring it to the client-segment. That distinction will be reserved for "Navi," which could only debut in 2019, if not later.

Quote:xGMI (inter-chip global memory interconnect) is a cable-capable version of AMD's Infinity Fabric interconnect. A line of code in the latest version of AMDGPU Linux drivers reveals that "Vega 20" will support xGMI. This line tells the driver to check the state of xGMI link. A practical implementation of this could be inter-card high-bandwidth bridge connectivity that would otherwise saturate the PCI-Express host bus; similar to NVIDIA's usage of the new NVLink bridge for Quadro and Tesla products based on its "Volta" and "Turing" GPU architectures.

Quote:It would appear AMD's 7nm Vega 20 has been benchmarked in Final Fantasy XV. While the details are scarce, what we do know is the hardware device ID 66AF:C1 can be linked to Vega 20 via the Linux patches back in April. Now considering AMD has not confirmed any 7nm Vega graphics cards for consumers, It is more likely this version is an engineering sample for the new Radeon Instinct or Pro series cards.

While the Final Fantasy XV benchmark is not exactly an AMD friendly title considering the heavy use of NVIDIA's Gameworks. The fact remains, it does give us a glimpse into its performance compared to the currently available Vega based graphics cards on the market. Looking at the charts, 66AF:C1 offers performance that typically beats out the previous Vega offerings including, Vega 56, Vega 64 and even the Frontier Edition. Even so, it still lags behind NVIDIA's older GTX 1080. A few possible reasons for these results exist, first, the GPU may not be operating at peak clock speeds which is indeed possible if it is an engineering sample. However, another likely possibility is that AMD is using 7nm as a means to lower power consumption, rather than to increase performance. Like all leaks of this nature it is best to take them with a grain of salt.

Quote:We’re resurrecting our AMD RX Vega 56 powerplay tables mod to challenge the RTX 2070, a card that competes in an entirely different price class. It’s a lightweight versus heavyweight boxing match, except the lightweight has a gun.
...
This mod absolutely loses in the power consumption department, but that was never really a question. It’s not difficult to install the powerplay tables mod (grab the registry files on overclock.net or somewhere similar), but it is potentially dangerous. For cards with lower quality PCBs, VRMs, and VRM cooling, it is feasible that the VRM could sustain long-term damage from significantly increased power throughput. An example might be some of the caps in the Vcore capacitor bank, where it would be possible to exceed 85-105C (depending on capacitor spec) on lower-end coolers, resulting in earlier deaths of those caps. For well-cooled cards with a good VRM, this is a non-issue and not a concern. The only remaining concern would be one of long-term impact on the Vega GPU. It is possible that such power input could kill the card early, but we simply don’t know without any endurance testing sample data to pull from.

It is fun, though, and it’s definitely possible to minimally match or marginally outmatch the RTX 2070 stock performance. This is definitely one of those enthusiast mods you do just to show you can, although the performance gains are real. The other side of the story, illustrated in some of these charts, is that the less-liked Vega variant – 64 – is able to compete with the 2070 when both are in stock configurations.

We still like Vega 56 a lot for its “mod-ability” and overclocking potential, and still believe it is one of the better cards at its price-point when considering the kind of enthusiast fun you can have with it.

One final note: We'd be able to achieve an even bigger overclock if we put Vega under a waterblock, like we did once before, but left them to air for this one.

Quote:According to GamersNexus, the first "Vega 20" products won't launch before the turn of the year, and even in 2019, one can expect product launches till the end of Q1 (before April). GamersNexus cites reliable sources hinting at the later-than-expected arrival of "Vega 20" as part of refuting alleged "Final Fantasy XV" benchmarks of purported "Vega 20" engineering samples doing rounds on the web. Lisa Su stressed the importance of data-center GPUs in AMD's Q3-2018 earnings call, which could hint at the possibility of AMD allocating its first "Vega 20" yields to high-margin enterprise brands such as Radeon Pro and Radeon Instinct.

Quote:AMD has applied for a trademark covering what many assume is a logo for Vega 2 graphics. The new mark closely remembers Vega's existing logo, which is essentially just a stylized "V," but with the right side modified to look like the Roman numeral "II." (Or, as the company described it in the trademark application, "a smaller triangle and four alternating light and dark horizontal stripes.") The implication is clear.
...
All of which means red team faithfuls await graphics news from AMD with bated breath. It's tempting, then, to assume AMD's application for this trademark means it plans to reveal new consumer-focused GPUs sooner than later. But it's worth remembering that companies often apply for trademarks long before they have to use them, or make sure their bases are covered legally even if they never actually release the product. The company might also plan to use this updated logo for the new Radeon Instinct accelerators but not for its upcoming consumer graphics cards.

Does that mean we expect the RX Vega line to be the last of AMD's consumer graphics cards? No, but it does mean that applying for a trademark doesn't necessarily imply a certain timeline, so it's best not to get hopes up. (Of course, us saying that means AMD's probably going to do a surprise debut of its new graphics cards just in time for the holidays or something.) This logo is a breadcrumb to follow; don't mistake it for a true reveal. Especially when people have already raised questions and circulated rumors about what AMD has planned for its next-gen Navi architecture.

Quote:It's clear that AMD's working on something graphics-related. The company also recently filed to trademark a logo that appears to be for Vega 2 graphics, and in October, chief executive Lisa Su assured Barrons that AMD would remain competitive in the graphics market. She didn't say how the company planned to do that--such as whether or not it'd be through the next-gen Navi architecture--so perhaps Vega 2 is the answer for now.

For now, the addition of these IDs to AMD's Linux drivers could indicate that new products are imminent, or simply mean the company has moved on to internal testing for whatever it has planned next. Maybe we'll hear more from the company in the new year as much of the industry flocks to conventions like CES or simply latches on to whatever news it can to make sure people stay hyped after the holiday season.

Quote:AMD has temporarily slashed the price of its Vega 56 — or at least one Vega 56 model — ahead of Nvidia’s next mainstream GPU launches, cutting the GPU down to $280. The new move could bring the older card back into competition in terms of overall performance, though it also puts sharp pressure on AMD’s recently-released RX 590.
...
Vega 64 prices have actually increased — the GPU is currently $500 — but the Vega 56 has dropped to $280, at least in theory. We say “in theory,” because the cheapest GPU at Newegg is currently a $400 card listed as being on sale, and the specific link that AMD sent us to the GPU that’s supposed to be cheaper has the card on Auto Notify as of this writing.

We’re not going to speculate much on how Vega 56 will perform in this position because we haven’t seen how the 1660 Ti will fare. Presumably AMD picked a price point that the GPU could actually compete at to start with. What’s less clear is what impact this will have on the RX 590. A price cut to that GPU — say, down to $230 or so — would position it much more effectively against both the Vega 56 and the older RX 580.

Quote:The "66AF" GPU, thought to have been Navi, is actually registered under Linux AMD GPU drivers as "Vega 20," making the Navi conclusion even more suspect.

Also, comparing this 66AF:F1 GPU to Radeon VII on CompuBench, nearly everything in the OpenGL information is identical. A notable difference is that Radeon VII has an additional tag under "GL_EXTENSIONS" called "GL_AMD_gpu_shader_half_float2," which may be the tag that specifies Radeon VII's reduced floating point performance compared to other Vega 20 GPUs, like the Radeon Instinct MI60.

While it is hard to tell what exactly this GPU is, if Linux's driver IDs can be trusted, it doesn't appear to be Navi. Even if the GPU is from the Navi lineup, it's hard to glean useful performance data and GPU specifications due to the nature of the CompuBench benchmark. For now, it appears more likely this is just another Vega 20 GPU, perhaps even a new WX Pro series model.