AMD Teases First Navi GPU Products: RX 5700 Series Launches in July, 25% Improved Perf-Per-Clockby Ryan Smith on May 26, 2019 11:30 PM EST
- Posted in
- PCIe 4.0
- Computex 2019
- RX 5700 Series
While the bulk of this morning’s AMD Computex keynote has been on AMD’s 3rd generation Ryzen CPUs and their underlying Zen 2 architecture, the company also took a moment to briefly touch upon its highly anticipated Navi GPU architecture and associated family of products. AMD didn’t go too deep here, but they have given us just enough to be tantalized ahead of a full reveal in the not too distant future. The first Navi cards will be the Radeon RX 5700 series, which are launching in July and on an architectural level will offer 25% better performance per clock per core and 50% better power efficiency than AMD’s current-generation Vega architecture. The products will also be AMD's first video cards using faster GDDR6 memory. Meanwhile AMD isn’t offering much in the way of concrete details on performance, but they are showing it off versus NVIDIA’s GeForce RTX 2070 in the AMD-favorable game Strange Brigade.
A Peek At the Navi GPU Architecture
Compared to today’s Ryzen information AMD is being a lot more modest – there’s no specific SKU information for a start – but they are offering us our first architectural details on Navi since we learned in 2016 that it would bring “next gen memory” support. First and foremost, AMD has been tinkering with their GPU architecture to improve both the effective IPC and memory bandwidth efficiency. In what the company is calling their Radeon DNA (RNDA) archtiecture – I should note that it’s not clear if this is a branding exercise to downplay the GCN family name or if it's a more heavily overhauled architecture – AMD has revealed that Navi’s compute units have been redesigned to improve their efficiency. AMD's press materials also note that, regardless of the above changes, the size hasn't changed: a single CU is still 64 stream processors.
Altogether, a Navi core/CU should be 25% faster than a Vega core on a clock-for-clock basis. The devil is in the details of course – AMD's data is based off of their internal testing, taking the geomean of 30 games tested at 3840x2160 with Ultra settings and 4x AA, so it's not a holistic view of just the core architecture – but it’s still potentially one of AMD’s largest GPU IPC gains in the last several years. AMD’s presentation also noted that this was part of a larger streamlining of the graphics pipeline, which is designed for both higher IPCs and high clockspeeds. For reference, on the same process AMD has been able to push Radeon VII well past its comfort/efficiency point to 1750MHz, so it will be interesting to see how Navi compares.
Feeding the beast is a new multi-level cache hierarchy. AMD is touting that Navi’s cache subsystem offers both higher performance and lower latency than Vega’s, all for less power consumption. AMD has always been hamstrung a bit by memory/cache bottlenecks, so this would be a promising development for AMD’s GPU architecture. Meanwhile for a bit of reference, Vega already implemented a more modern cache hierarchy, so it would seem unlikely that AMD is changing their cache levels or what blocks are clients of which caches.
AMD's memory controllers themselves have also been updated. Long expected, and confirmed in the press release that went out after Dr. Lisa Su's keynote, the upcoming RX 5700 series cards use GDDR6 memory, which should give AMD's cards a hearty bandwidth bump over their comparable GDDR5 Polaris cards.
It's also worth noting that, fittingly, the new Navi parts support PCI Express 4.0 as well. This will actually be AMD's second GPU with PCIe 4.0 functionality – Vega 20 was first, but it's only enabled in the Radeon Instinct parts – so this is the first time it's enabled in a consumer part. Of course to make full use of it you'll need a PCI 4.0-capable host, which AMD is happy to sell you as well. As for the performance impacts, these remain to be seen. Thus far high-end AMD/NVIDIA parts haven't been significantly bottlenecked in games by PCIe 3.0 x16 (or even x8), so it will be interesting to see how much the extra bus bandwidth helps these first generation of parts.
Last but certainly not least of course is overall power efficacy. Thanks to the combination of AMD’s architectural improvements and TSMC’s 7nm process, AMD is promoting a 50% increase in performance per watt for Navi. Power efficiency is AMD’s Achilles heel relative to NVIDIA, so this is another area where we’re eager to see AMD catch up. It should be noted however that this statement didn’t come with a qualifier – if it’s 50% more efficient at the same clockspeeds as Vega or a given total card TDP – which can impact the meaningfulness somewhat. For Polaris/Vega, AMD opted to push their cards well up the voltage/frequency curve in order to maximize performance at a cost to power consumption, so power efficiency is fluid based on what clockspeeds AMD ships at.
Update: Since there have been some questions about what the efficiency number is in reference to – if it's against the 7nm Vega 20 GPU or the 14nm Vega 10 GPU – I went back over AMD's keynote presentation and transcribed it. Here is what Dr. Lisa Su specifically said on the subject (emphasis mine).
And then, when you put that together, both the architecture – the design capability – as well as the process technology, we're seeing 1.5x or higher performance per watt capability on the new Navi products
So while I will hold off on calling this the definitive word until we've had a full technology briefing, at first glance it would seem that the 50% efficiency gain is a combination of architecture and the move from GloFo 14nm to TSMC 7nm.
Radeon RX 5700 Series: Coming in July
Alongside the architecture teaser, AMD also offered up some high-level details about the first Navi video cards. Navi will be sold under AMD’s RX 5000 series of video cards – that’s right, after Vega and Radeon VII, they’re going back to distinct series numbers. The RX 5000 series means that AMD is making a big jump in their numbering system, going back to 4 digits and back to the 5000 series. Officially, this is because AMD is celebrating its 50th year in business this year – so of course the product numbers need to start with a 50. However long-time observers will note that it’s been 10 years now since AMD’s previous 5000 series of video cards, the well-received Radeon HD 5000 series, and AMD may be trying to capture a bit of that.
Of the RX 5000 cards, the first series will be the RX 5700. These will be launching in July. AMD is also showing off the specific Navi GPU that will be used here, though as far as AMD’s numbering system goes, we don’t know whether this is Navi 10 or another GPU model number.
This is a traditional, monolithic die with a significant pin count package. So if I were a betting man (ed: you still owe me a steak), then I’d expect it to be paired up with GDDR6. (ed again: this has since been confirmed by AMD's press release)
Update: Thanks to our own Andrei Frumusanu for doing the leg work, we now have a die size estimate based on these and some unpublished photos. Our working guess right now is 275mm2, give or take a few percent. Though that's going to be less precise than what a proper micro-caliper measurement will turn up, so this is very much still just an estimate.
Finally, while AMD isn’t giving us concrete performance information, the company did use its Computex keynote to briefly demonstrate performance on the card versus NVIDIA’s GeForce RTX 2070, a $500 video card. While this is a game that favors AMD to begin with, it’s a promising sign that they’re able to pull ahead of mid-to-high-end NVIDIA card by 10%. Which means we’re all going to be eagerly awaiting more information on the video card as AMD ramps up for their July launch.
AMD's next big gaming event will be their E3 2019 Next Horizing Gaming Event, and AMD tells us that we'll find out more about the RX 5000 series there. So stay tuned.
Post Your CommentPlease log in or sign up to comment.
View All Comments
TEAMSWITCHER - Monday, May 27, 2019 - linkIn name only.
tipoo - Monday, May 27, 2019 - linkThe SIMD setup change is a pretty fundamental redesign. How much can change while still calling it GCN?
WinterCharm - Tuesday, May 28, 2019 - linkLisa Su, CEO of AMD -- "Its a ground up redesign"
Armchair internet commenter -- "It uses the same ISA so it's not!"
Guess what? The i9 9900k, AMD Bulldozer, and AMD Zen all use x86/64. they are not the same design.
mode_13h - Tuesday, May 28, 2019 - linkComparing CPU with GPU ISA is a bad analogy. CPUs have a front-end ISA and a back end with micro-ops, shadow registers, etc. GPUs have no such indirection. This is largely because virtually no one programs GPUs in their native assembly language, so the native ISA is much easier to change.
CiccioB - Wednesday, May 29, 2019 - linkIt's a ground up redesign because they used the same ALUs and just arranged them in a different way.
It seems that they simply doubled graphics engine, for 4 to 8, that is now each engine can have up to 5 CU instead of 10, and each engine can now have more ROPs with respect to old GCN versions.
If so this does not make it a completely new architecture, as it does not introduces nothing new but a "ground up redesign", that is a different way to use the old blocks with same capabilities, it just improve the number of some of those blocks.
Poòaris 10 is not a new architecture with respect to Polaris 11 just because it uses more blocks, and Polaris 10 is not a new arch with respect to Ellsmere just because the number of ACES or geometry engines has increased.
Here we have that to increase even more the number of uncore elements (as it seems the max number of ALUS is still 4096 as old GCN) they had to redesign the graphics engine and doubling their number.
I'm quite curious to see ho big these changes are going to impact on die size.
We have seen that nvidia had a quite die size impact on its changes at shader (with the introduction of the INT32 unit), SMX level with their doubling of their front end and with the new scheduler able to work at thread level. This on net of the other die size increase due to tensor and RT core, as we can see for TU116 and TU117.
Never the less these changes brought an IPC increase which is about 30% with respect to Pascal and with a reduction is power consumption.
silverblue - Wednesday, May 29, 2019 - linkIt's not perfect though; the 1660 loses to the 1060 6GB in The Witcher 3 and Middle Earth: Shadow of War, and on average it's more like 15% faster for about the same power consumption, certainly not the 30% increase in performance for a reduction of power consumption as you're stating (source: https://www.techspot.com/review/1813-geforce-gtx-1... ).
Santoval - Monday, May 27, 2019 - linkNo they didn't. Their first post-GCN card is going to be Arcturus.
bridgmanAMD - Wednesday, May 29, 2019 - linkArcturus is just the first chip we started working on after switching back to using engineering code names rather than names like "Navi" that were discussed publicly in roadmaps etc...
I don't think we have ever said anything to suggest that it will be post-GCN or even post-Navi.
mode_13h - Wednesday, May 29, 2019 - linkThanks for that.
Now, I'm unclear on the distinction between the two classes of names - is it just that Arcturus hasn't been publicly announced?
brakdoo - Monday, May 27, 2019 - link"Thanks to the combination of AMD’s architectural improvements and TSMC’s 7nm process, AMD is promoting a 50% increase in performance per watt for Navi"
TSMC deserves no credit if those 50% are just about the architecture....