AMD Announces Radeon RX 5700 XT & RX 5700: The Next Gen of AMD Video Cards Starts on July 7th At $449/$379
by Ryan Smith on June 10, 2019 7:20 PM ESTAs part of a jam-packed day of AMD product news, moments ago AMD’s CEO, Dr. Lisa Su got off the stage, wrapping up her suite of announcements. The highlight of which is AMD’s new family of video cards, the Radeon RX 5700 series. AMD first teased these back at the tail-end of Computex a few weeks ago, and while the cards won’t actually launch until July, AMD has opened the floodgates on information about these cards – pricing, expected performance, architecture – so let’s get to it.
The Radeon RX 5700 series – which I’ll call the 5700 series for short – are AMD’s new family of mid-to-high end video cards. Within AMD’s product stack these cards essentially replace AMD’s previous RX Vega 64/56 parts, offering similar-to-better performance at lower prices, lower power consumption, and with newer features. To be clear, these are not flagship-level video cards, and at some point in time Vega 64/56 will get true successors in the form of faster, more powerful high-end video cards. But within AMD’s product stack and in the broader market, this is where the new cards will land.
These new cards from AMD are part of their first wave of cards based on their new RNDA architecture family. We’ll get into (excruciating) detail about that at a later time, but at a high level RDNA makes some pretty radical shifts in how AMD’s underlying GPU architecture works, more than earning the new name and realigning our performance expectations for AMD video cards. RDNA ultimately seeks to boost both AMD’s workload efficiency – that is, getting more work done with the same resources – as well as their power efficiency, in order to improve their competitiveness in the PC video card market. Pioneered in the Navi family of GPUs, the RDNA architecture will be the basis of AMD products for a long time to come; and not just PC GPUs, but consoles (Xbox and Playstation), mobile, and whatever other deals AMD can land.
But getting back to the matter at hand, AMD is launching two 5700 series cards here next month. At the high end we have the fully enabled Radeon RX 5700 XT (yes, those insufferable suffixes are back), which sports 40 CUs and a peak clockspeed of over 1900MHz. It’s partner in crime will be the suffix-free Radeon RX 5700, which is the traditional second-tier part that cuts back on some functional units and performance in the name of offering a lower-priced card (and letting AMD salvage Navi chips). These parts, AMD tells us, will be competitive with the GeForce RTX 2070 and RTX 2060 respectively, though of course this is something we will determine for ourselves once we have them in for testing.
AMD Radeon RX Series Specification Comparison | ||||||
AMD Radeon RX 5700 XT | AMD Radeon RX 5700 | AMD Radeon RX 590 | AMD Radeon RX 570 | |||
Stream Processors | 2560 (40 CUs) |
2304 (36 CUs) |
2304 (36 CUs) |
2048 (32 CUs) |
||
Texture Units | 160 | 144 | 144 | 128 | ||
ROPs | 64 | 64 | 32 | 32 | ||
Base Clock | 1605MHz | 1465MHz | 1469MHz | 1168MHz | ||
Game Clock | 1755MHz | 1625MHz | N/A | N/A | ||
Boost Clock | 1905MHz | 1725MHz | 1545MHz | 1244MHz | ||
Throughput (FP32) | 9.75 TFLOPs | 7.9 TFLOPs | 7.1 TFLOPs | 5.1 TFLOPs | ||
Memory Clock | 14 Gbps GDDR6 | 14 Gbps GDDR6 | 8 Gbps GDDR5 | 7 Gbps GDDR5 | ||
Memory Bus Width | 256-bit | 256-bit | 256-bit | 256-bit | ||
VRAM | 8GB | 8GB | 8GB | 4GB | ||
Transistor Count | 10.3B | 10.3B | 5.7B | 5.7B | ||
Typical Board Power | 225W | 180W | 225W | 150W | ||
Manufacturing Process | TSMC 7nm | TSMC 7nm | GloFo/Samsung 12nm | GloFo 14nm | ||
Architecture | RDNA (1) | RDNA (1) | GCN 4 | GCN 4 | ||
GPU | Navi 10 | Navi 10 | Polaris 30 | Polaris 10 | ||
Launch Date | 07/07/2019 | 07/07/2019 | 11/15/2018 | 08/04/2016 | ||
Launch Price | $449 | $379 | $279 | $179 |
To start things off, as always we have the specs. It’s best to be clear now that while the raw specifications are helpful in understanding the basics of these cards, due to the RDNA architectural transition the numbers can be deceiving. In particular, RDNA incorporates a number of changes to improve both compute efficiency/utilization and reduce memory bandwidth needs. So if anything, these specifications understate the 5700 series cards by a decent degree, as in practice they’re going to be more efficient per clock than their Polaris predecessors.
And I’m using AMD’s Polaris cards here as my point of comparison – despite the fact that these new cards will perform more like Vega 64/56 – because at a hardware level Navi 10 replaces Polaris 10 as a mid-range(ish) GPU. AMD’s first 7nm GPU for this market, which is being fabbed over at TSMC, measures in at 251mm2, pacing in 10.3 billion transistors into that modestly-sized die. This is a bit larger than the 232mm2 Polaris 10, and incorporates 80% more transistors. So there’s a whole lot more hardware at work here, which for AMD should translate into a good deal more performance.
Diving into the 5700 XT, AMD’s full-fledged Navi card will attempt to put its best foot forward in terms of securing a new spot for AMD in the video card market, and in showing off the RDNA architecture. This is a 40 CU part, with clockspeeds peaking at 1905MHz. New to the Navi generation is a figure AMD is calling their “game clock”, which is analogous to NVIDIA’s turbo clock, and is a conservative estimate of what the average GPU clockspeed is during normal games. To be sure, AMD’s clocking behavior hasn’t really changed – they still try to boost as high as they can, as much as power and thermals allow – but this value is intended to offer better guidance to buyers about what the hardware will typically do.
Looking at these clockspeed values then, in terms of raw throughput the new card is expected to get between 9 TFLOPs and 9.75 TFLOPs of FP32 compute/shading throughput. This is a decent jump over the Polaris cards, but on the surface it doesn’t look like a huge, generational jump, and this is where AMD’s RDNA architecture comes in. AMD has made numerous optimizations to improve their GPU utilization – that is, how well they put those FLOPs to good use – so a teraflop on a 5700 card means more than it does on preceding AMD cards. Overall, AMD says that they’re getting around 25% more work done per clock on the whole in gaming workloads. So raw specs can be deceiving.
Meanwhile on the front and backends respectively of the new GPU, 5700 XT can spit out 4 rendered polygons per clock, and more still when primitive shading is employed. In order to consume all of the pixels that will come flowing out of that process, the GPU ships with 64 ROPs on the backend, twice as many as on AMD’s Polaris cards (or the same number as the Vega cards. We’ll get into architectural matters later, but AMD has put some work in here in order to improve the throughput of these blocks, so the 5700 XT on paper looks like a reasonably well-balanced architecture.
Feeding the beast that is AMD’s Navi 10 GPU are the company’s new GDDR6 memory controllers. While AMD as a company has their arms deep in the development of GDDR memory, for product cadence reasons they are becoming the second company to employ the new memory type, behind NVIDIA. So as we’ve seen in other products, GDDR6 stands to significantly improve the amount of memory bandwidth AMD has to play with; going from 256GB/sec on comparable Polaris cards to 448GB/sec on these new Navi cards. Compounding this, AMD is now employing delta color compression throughout virtually their entire chip, so memory bandwidth efficiency as a whole is improving. Along these lines, the new GPU employs a new cache heiarchy. The nuts and bolts of this we’ll save for another time, but it ultimately keeps traffic local to the GPU and better avoids using expensive off-die GDDR6 bandwidth when it can be avoided. Overall between its 64 ROPs and significant compute throughput, 5700 XT can eat a lot of bandwidth, and AMD intends to be well-prepared to feed it.
Finally, let’s talk about power consumption. For this generation AMD is sticking with their Board Power figures, which means they’re largely comparable to past AMD cards. In the case of the 5700XT, this is a 225 Watt card, similar to the RX 590 and (on paper) a bit more than the Vega 56. The card draws its power from a combination of the PCIe slot and external power connectors, relying on an 8-pin + 6-pin configuration there. At 225W it’s definitely not a lightweight card when it comes to power consumption, and in our full review we’ll have to see how this translates to real-world performance, and if AMD has tuned this card more towards performance than power efficiency.
Cooling this card will be a largely traditional AMD blower. AMD is employing an aluminum shroud and backplate here (similar to the Vega 64), with a vapor chamber drawing heat up from the GPU to the heatsink. AMD tells us that the blower itself has been further optimized for air flow and noise, and the company has set the default acoustic limit to a relatively low 43dB. Ultimately AMD is trying to strike a new balance between the benefits of the blower design and noise; blowers work far more consistently, which AMD considers desirable for their wide range of customers, but by tuning the blower and capping the noise a bit lower, they’re trying to keep it from being quite so audible in quiet environments.
Radeon RX 5700
Not to be entirely overshadowed, below the Radeon RX 5700 XT we have the vanilla Radeon RX 5700. This card is largely cut from the same cloth as its faster XT sibling, trading off some performance for lower power consumption and lower pricing.
The lesser of the 5700 cards ships with 36 CUs enabled, and more modest clocks. The average game clock is rated for 1625MHz, with a maximum boost of 1725MHz, meaning that for compute/shader/texture workloads it should deliver around 87% of the 5700 XT’s performance. Meanwhile for ROP and geometry throughput, we’re looking at 93% of that performance.
Meanwhile nothing changes for the 5700 relative to the XT card when it comes to memory. It gets the same 8GB of 14Gbps GDDR6. So the 5700 should be even better fed than its full-fledged counterpart, relatively speaking.
The overall drop in performance also means power consumption has come down. The card is rated for a board power of 180W, which is comparable to what the RX 580 was rated for. For anyone crunching the numbers at home, the 45W drop in rated power consumption is even greater than the rated drop in performance, so it’s likely that the 5700 will end up being the more power efficient of the two cards.
Product Positioning & the Competition
Last but certainly not least of course is the details of next month’s launch, and how AMD’s new cards stack up to the competition – both AMD and NVIDIA.
Like the rest of AMD’s new 7nm consumer hardware, the two Radeon cards will be launching on July 7th. The 5700 XT will hit the shelves at $449. Meanwhile it’s smaller sibling, the 5700, will be a $379 card.
Next month’s launch is a traditional, driven-from-the-top full reference card launch. This means AMD’s blower-style reference cards will be what you find on the shelves on the first day. Semi-custom and custom cards are of course in the works, but those will come at a later time. Meanwhile, unlike NVIDIA, AMD isn’t doing a Founders Edition program here, so once custom cards do come out, barring any market changes they should be priced the same as AMD’s reference cards.
Interestingly, even though this is a new GPU family on a new architecture (on a new process), AMD is doing a game bundle of sorts. Bundled with the Radeon RX 5700 series cards is a 3 month subscription for Microsoft’s new Xbox Game Pass program. The company’s new all-you-can-eat game subscription service, despite the name it applies to PC games as well. I would be lying if I said I wasn’t a bit off-put by the idea of including what amounts to a trial subscription as a bundle – as opposed to games you own – but it’s certainly different. And I suspect there was some wheeling & dealing by Microsoft to promote the new service.
Looking at the AMD product stack then, the new Radeon 5700 cards are an interesting addition to AMD’s lineup. They will remain below the Radeon VII as AMD’s fastest card – the Vega derivative is still a tier above – but they are supplanting the Vega 64/56 and then some. According to AMD’s own data, the 5700 XT is on average 14% faster than the Vega 64, and closer to 30% faster than the Vega 56. AMD doesn’t provide similar data for the 5700 (vanilla), but with those numbers the card should easily best the Vega 56.
AMD has been drawing down Vega card inventory for a while now, so if you look at the few cards left on the market, you’ll find that they’re a fair bit cheaper than the new Navi cards. So for the moment they are options as cheaper alternatives, but as is usually the case here, this isn’t a situation that will last (and won’t come with any of Navi’s benefits, obviously). Still, I’m curious to see just how close (or far apart) the two families really end up.
Instead the big competitive question is going to be how all of this compares to NVIDIA’s current-generation GeForce RTX 20 series cards. NVIDIA kicked off that launch almost a year ago, and the unimpressively priced cards have been ruling the roost for a while now.
According to the slides AMD has provided, the $449 5700 XT should beat the $499 RTX 2070 by a few percent. That said, vendor benchmarks must always be taken with a sizable grain of salt, as vendors like to put their products in a good light. Credit to both AMD and NVIDIA here, they’ve actually been pretty decent as of late, so I suspect AMD’s numbers are close to what we’ll find with the hardware. In which case I’m expecting an earnest 2070 competitor, though we’ll see if the 5700 XT can consistently beat it.
326 Comments
View All Comments
Cellar Door - Monday, June 10, 2019 - link
Ryan - do you know if vega 56 and 64 are EOL?RaV[666] - Monday, June 10, 2019 - link
Can you think of one reason to make them ?I mean they will be made, as in vega10 chips for datacenters but for gaming, theyre gonna have higher MUCH higher asps on 5700
AshlayW - Monday, June 10, 2019 - link
Yes Vega 10 is being made for Google Stadia gaming as they use PRO V340 cards with dual "Vega 56" GPUs. GCN still is better for Compute and HPC I think, but Vega 20 will largely succeed that in HPC. GCN is not going anywhere.Oh, I do not think Vega 10 is cheaper to make than Navi 10. Yes the process is mature and cheaper, but the die is almost 2X the size and you factor HBM2 and interposer cost into that and the price is largely in the same ballpark.
Navi 10 cannot do a "V340" style card easily, or as effectively, as Google Stadia needed graphics density and the on-package memory on Vega 10 makes the overall space requirements much smaller, so yes Vega 10 is likely to be made still, and itself is much cheaper than Vega 20.
mode_13h - Monday, June 10, 2019 - link
Why would 2x Navi's be so much worse than 1x Hawaii? You're talking about 512-bits of memory data bus, in each case - just one compute die vs. two.AshlayW - Tuesday, June 11, 2019 - link
What? where did you get Hawaii from? V340 uses "Vega 10" which has on-package HBM2 instead of GDDR5. That is a major advantage for space saving when putting multiple GPU packages on the same card.mode_13h - Wednesday, June 12, 2019 - link
My point was that Hawaii cards use 512-bits of datapath on a single card, so perhaps 2x 256-bit Navi's can fit.Regarding density, I don't see your point. From a server's perspective, a PCIe card is a PCIe card, unless it's low-profile, which I don't think it is.
olafgarten - Monday, June 17, 2019 - link
They might not be using standard PCIe or maybe putting multiple chips on a single card. Either way density helps.Acreo_Aeneas - Sunday, June 30, 2019 - link
Servers would likely have to use PCI-e LP or mini PCI-e. In whichever case, most servers are built more for memory I/O and storage performance and capacity (with also a focus on CPU performance) rather than on how powerful the GPU is onboard. Most servers are even headless and don't even have GPUs on board. The few that do usually use theirs for interfacing with a terminal.This does not include server or server farms built specifically for mutli-gpu setups. Usually that is scientific/graphics oriented or with the increasing niche of bitcoin mining.
WaltC - Thursday, July 4, 2019 - link
Pretty sure I heard Su mention that NAVI included (at the "Rdna" level) improvements to compute. We'll know in a few days, of course. (Impatience makes time drag, eh?...;)) I don't usually do this, but *provided I can buy a 5700XT for either $499 (20th anniversary) or $449 MSRP* I'll be buying one next week. I'm going to be rather ticked off if the prices for the card a grossly inflated! Here's hoping AMD will control this much better than than happened at the RX-480's debut--what makes me shudder a bit is that I just read some days ago that Bitcoin stock was on the rise again! Stadia servers are likely only using Vega now because NAVI simply wasn't available when they began. Should change in a couple of weeks, possibly.Ryan Smith - Monday, June 10, 2019 - link
As far as consumer cards go, they've been drawing down inventory from the market for a couple of months now. I don't know if they've been formally discontinued, but they may as well be de-facto done for.