The AMD Radeon VII isn’t a GPU we ever anticipated to see as a shopper product. AMD repeatedly indicated it had no specific plan to deliver a 7nm to this area, preferring to maintain its first run of 7nm hardware reserved for servers and the AI/ML area. Then, at CES 2019, AMD CEO Lisa Su introduced AMD would deliver a brand new GPU to market to compete face to face with the RTX 2080 at the $700 worth level. Quick ahead to the current day and right here we’re.
The Radeon VII is predicated on the similar 7nm silicon as the Radeon MI50 and MI60. AMD shrank the die considerably between the two GPUs and used the area financial savings to squeeze in one other set of HBM2 chips, doubling out there RAM bandwidth. When you care about sheer reminiscence capability in your graphics card — and to be clear, there are skilled and scientific purposes that profit from giant GPU buffers — the Radeon VII is the solely card on the market that provides you this a lot RAM for beneath a thousand dollars.
Boosted FP64 Efficiency
The Radeon VII has one other benefit over different playing cards on the market, although AMD stored this one tight to their vest. Till immediately, the highest FP64 efficiency you might purchase in a shopper GCN GPU was the Radeon 7990, a short-lived dual-GPU product that’s almost 5 years previous. Initially, AMD communicated that its Radeon VII GPU was able to simply zero.88 TFLOPS (1/16th of FP32 efficiency). As an alternative, the Radeon VII is able to a whopping Three.46 TFLOPs, or simply beneath half of the MI60’s most efficiency.
This vital repositioning places a brand new spin on the Radeon VII, positioning it extra clearly for double-precision compute workloads. Sadly, this information wasn’t communicated till proper earlier than the evaluation NDA, which suggests we gained’t have time to take a specific examination of the GPU in that context. We intend to write down a separate article investigating compute and scientific workloads on the Radeon VII, which can give us a chance to look at this aspect of the GPU extra completely. At this time’s assessment will give attention to the shopper aspect of the equation.
As with the first 7nm GPU, all eyes are going to be on the Radeon VII to see how nicely it performs on the course of node. We have already got some info on this level. As the desk above signifies, the Radeon VII die is ~67 % the measurement of the Vega 64 die. Clock charges have improved solely modestly, with the base clock up by 1.09x and the increase clock growing by 1.13x.
Most of the enhancements between Radeon Vega and Radeon VII boil right down to the big improve in RAM bandwidth and complete VRAM buffer. The clock good points from the 7nm shift are very modest. AMD did notice to us that it had decreased the GPUs inner latency and made a number of performance-enhancing tweaks to the structure, however knowledge right here was restricted. These hoping that AMD would have a brand new main AI or machine studying initiative to announce — one thing akin to DirectX ray tracing help or new antialiasing strategies — can be dissatisfied.
New Thermal Monitoring
The Radeon VII makes use of a brand new sort of thermal monitoring system moderately than the previous, edge-mounted GPU thermistors it had beforehand deployed. Radeon VII has a complete of 64 temperature sensors mounted throughout the die, 2x the variety of Vega 64. Going ahead, Radeon VII GPUs will use the most temperature measured throughout the die, referred to as the Junction temp, to regulate GPU conduct. In accordance with AMD: “Controlling based on Junction Temperature from the extensive sensing network allows each GPU to reliably maximize its performance potential while reporting an additional temperature that is more representative of the hottest parts of the GPU.”
AMD notes that Junction temps can be larger than what players are used to seeing in the previous. This shouldn’t be thought-about an issue, and the GPUs are designed to carry these temperatures safely. Junction temps of 110C aren’t uncommon or thought-about problematic.
The Story So Far…
Earlier than we dive into efficiency figures, let’s revisit how the GPU market has advanced in the previous six months. Nvidia’s RTX refresh cycle this previous fall didn’t do a lot to enhance performance-per-dollar. Of its new high-end GPUs, solely the RTX 2080 Ti genuinely moved the ball ahead on efficiency. The RTX 2080 and RTX 2070 are barely quicker than the GTX 1080 Ti and GTX 1080 respectively (assume Eight-12 %), however carry greater costs than their predecessor GPUs did. Nvidia’s justification for these worth will increase has been to level at its new ray tracing function as justification. ExtremeTech traditionally takes a really dim view of shopping for hardware for options you possibly can’t use, for causes we explored in-depth as a part of our RTX 2080 and 2080 Ti assessment.
However therein lies the rub. Even for those who agree that Nvidia’s RTX know-how is a dangerous guess, AMD hasn’t baked something equal into the Radeon VII. In the previous, one GPU vendor typically zigs when the different zags — Nvidia went all-in for Three-D glasses and screens a couple of years again, whereas AMD threw its weight behind Eyefinity and multi-monitor gaming. On this case, RTX and DLSS have been main launch options for Nvidia, whereas AMD’s main launch function is the simple promise of extra efficiency relative to its earlier era of graphics playing cards.
Players hoping that AMD would convey a GPU to market that shook up the established order or at the very least pressured Nvidia to decrease its costs are going to be disenchanted (once more). AMD is concentrating on the RTX 2080’s efficiency at the RTX 2080’s $700 worth level utilizing Vega’s 300W energy envelope.
All testing was finished on an Asus Prime Z370-A utilizing an Intel Core i7-8086Okay with 32GB of DDR4-3200 and the newest model of Home windows 10. A Thermaltake Toughpower 80 Plus Titanium 1250W PSU was used for testing.
All of our outcomes may be parsed in the slideshow under. Click on on every graph to open it in a brand new window.
AMD promised that the Radeon VII can be able to tackling the RTX 2080, and that’s usually what we see right here. In our full suite of checks, the Radeon VII provided 96 % the efficiency of the RTX 2080. That’s inside a 5 % margin of error, and shut sufficient to declare that sure, the two options are usually aggressive.
The Radeon VII tends to exhibit superior scaling to the RTX 2080, which is to say, it tends to lose much less efficiency than its competitor as decision will increase.
Energy Consumption, Warmth, and Noise
Word: My GPU check rig makes use of a V3 Voltair CPU cooler, which features a Thermo-Electrical Cooler, or TEC. This consumes further electrical energy. In consequence, our idle and energy consumption figures could also be larger than elsewhere on the net. Operating 32GB of RAM in XMP at DDR4-3200 additionally considerably will increase energy consumption in contrast with inventory voltage and DDR4-2400.
Our energy consumption figures are taken from the third loop of a Metro Final Mild Redux benchmark run at 1920×1080. All element settings conform to these used for our GPU evaluations.
AMD’s absolute energy consumption has scarcely budged in contrast with Vega 64, however its efficiency per watt has improved considerably. Utilizing the automated undervolting choice on our Radeon VII decreased energy consumption by ~7.5 % with out harming body charges in any respect.
When you think about the Radeon VII’s elevated efficiency, the GPU is certainly considerably extra environment friendly. The Radeon VII consumes roughly 75 % as a lot energy as the Vega 64 per body of animation drawn. Activate its underclocking function, and this drops to 70 %. However the RTX 2080 consumes simply 63 % the energy of the Radeon Vega 64. AMD’s 7nm GPU attracts roughly the similar quantity of energy as its Nvidia rival, however it isn’t fairly as environment friendly on the entire.
Lastly, there’s noise. I don’t personal a dB meter, however people — Radeon VII ain’t quiet. General, it’s similar to the Vega 64, however there are moments when the followers on the Radeon VII kick more durable. Additionally they are likely to ramp up quicker. Everybody has their very own private tolerance for this kind of factor, however I contemplate the noise profile of those playing cards to be a big damaging.
At this level, the noise state of affairs has turn into ridiculous. Ever since at the least Hawaii, reviewers have hit AMD for the noise profile of its reference designs. To its credit score, the firm has a minimum of tried to deal with this, however its most high-profile try to repair the drawback with a water cooler created a good greater mess. Vega 64 was a loud GPU, louder than I’m personally snug putting in in my very own system. Radeon VII doesn’t enhance on this in any respect. At this level, it’s clear AMD doesn’t even have any curiosity in constructing or outfitting its reference playing cards with coolers that match the efficiency of what Nvidia ships (and what Nvidia ships isn’t all the time nice, both, thoughts you). Hawaii launched over 5 years in the past. Why are we nonetheless ready for AMD to truly repair this in a high-end GPU that isn’t the Radeon Nano?
I anticipate a $700 GPU to have a greater noise profile than the $12 field fan I purchased at Aldi.
The Radeon VII is a tough match for the RTX 2080’s efficiency. It additionally lacks one in every of the RTX 2080’s particular liabilities. When Nvidia launched the RTX household, it requested clients to swallow a big worth hike relative to the firm’s then-current GPU lineup. AMD isn’t making that mistake. The Radeon VII gives extra efficiency than any AMD GPU earlier than it. It’s 1.33x quicker than the Vega 64 with no different pesky last-gen playing cards to muck up the stack.
In fact, provided that the Vega 64 presently sells for $400, you’re shopping for 1.33x extra efficiency for 1.75x the cash. That’s precisely the reverse of the sort of ratio we choose to see. It’s onerous to not marvel what 7nm Vega’s efficiency may need seemed like if the GPU had fielded extra ROPs and texture mapping models to accompany its monumental bandwidth improve — once more, we suspect that shopper video games merely weren’t the main marketplace for this card and that its general design displays that reality. We intend to put in writing a separate article specializing in GPU compute and can report again with how the Radeon VII’s further reminiscence bandwidth aids it in these contexts.
For me, the backside line is that this: I’ve persistently argued that Nvidia erred in elevating costs when it launched the RTX household. I’ve argued that there have been too many causes to consider RTX options gained’t be notably helpful throughout Turing’s lifespan to justify shopping for into the household, notably when older GTX playing cards based mostly on Pascal provided such aggressive efficiency.
Now we now have the Radeon VII. It lacks RTX/DXR options however doesn’t supply an alternate. Its bigger HBM2 buffer and interposer prices might clarify its pricing — I’ve questioned earlier than if Nvidia’s choice to boost costs would truly make it make sense for AMD to deliver 7nm Vega to the shopper market — however they don’t make it a very good worth. How a lot worth do I placed on ray tracing proper now? Not a lot. However a function that’s solely helpful in a tiny variety of video games continues to be at the very least arguably extra helpful than no function in any respect. Neither is there a lot cause to consider that the 16GB HBM2 buffer will get a exercise any time quickly. Recreation builders will all the time goal their titles to the GPUs that folks virtually personal; having one shopper GPU with 16GB of onboard reminiscence isn’t going to maneuver the needle when it comes to future recreation VRAM necessities.
If AMD had managed to deliver Radeon VII in at $600, it might have a real worth argument to make relative to Nvidia’s product stack. If it had outpaced the RTX 2080 at the similar worth, it might argue for superior rasterization efficiency with greater noise ranges as a suitable trade-off. As an alternative, what we’ve right here — no less than in the shopper market — is a loud RTX 2080-equivalent with out the admittedly doubtful options Nvidia tried to make use of to justify its worth will increase.
You realize what’s worse than an RTX 2080 with doubtful options and a nasty worth level? A loud RTX 2080-equivalent with no new options in any respect and the similar dangerous worth level.
The story isn’t all dangerous right here. AMD was capable of reap the benefits of its shift to 7nm to enhance its general aggressive standing towards the RTX household, and the Radeon VII competes towards the GTX 1080 Ti / RTX 2080 extra successfully than Vega 64 fared towards the older GeForce 1080. However I can’t hammer Nvidia for months over the worth will increase and positioning it launched with Turing solely to show round and laud AMD for delivering a GPU that roughly matches on perf however presents fewer options and better noise, unsure as the worth of these options could also be.
This isn’t the 7nm GPU you’re on the lookout for. We’ll have extra to say on compute particularly in the close to future.
(perform(d, s, id)
var js, fjs = d.getElementsByTagName(s);
if (d.getElementById(id)) return;
js = d.createElement(s); js.id = id;
js.src = “//connect.facebook.net/en_US/all.js#xfbml=1”;
(doc, ‘script’, ‘facebook-jssdk’));