Friday, June 30th 2017
New Performance Benchmarks of AMD's Vega Frontier Edition Surface
You probably took a long, hard read at our article covering a single-minded user's experience of his new Vega Frontier Edition. Now, courtesy of PCPer, and charitable soul Ekin at Linus Tech Tips, we have some more performance benchmarks of AMD's latest (non gaming specific) graphics card.
Starting with 2560x1440, let's begin with the good news: in what seems to be the best performance scenario we've seen until now, the Vega Frontier Edition stands extremely close to NVIDIA's GTX 1080 Ti video card in Fallout 4. It trails it for about 10 FPS most of the test, and even surpasses it at some points. These numbers should be taken with a grain of salt regarding the RX Vega consumer cards: performance on those models will probably be higher than the Frontier Edition's results. And for the sake of AMD, they better be, because in all other tests, the Frontier Edition somewhat disappoints. It's beaten by NVIDIA's GTX 1070 in Grand Theft Auto V, mirrors its performance in The Witcher 3, and delivers slightly higher performance than the GTX 1070 on Hitman and Dirt Rally (albeit lower than the GTX 1080.)
At 4K (3840x2160), the Vega FE trails the GTX 1080 by about 3 FPS (at 57 FPS, just shy of 60 FPS) on Dirt Rally; trails it again (this time with a 7 FPS difference) in Fallout 4, at around 42 FPS; delivers around 66% of the GTX 1080's performance on Grand Theft Auto V, and less than 50% of the GTX 1080 Ti's performance on the same game. In Hitman, the Vega FE delivers around 83% of the GTX 1080's performance (around 50 FPS versus the 1080's 60), and delivers almost the same result on The Witcher 3, barely maintaining a 30 FPS performance towards the end of the run.
Do note that all of these tests will apparently be re-run by PCPer, and the publication is looking to publish their results later today. Also keep in mind the Vega Frontier Edition isn't a consumer graphics card, and isn't officially meant for gaming. Instead, it's meant for professionals or prosumers who do some professional workloads as well as some gaming, and want to have the ability to test their development fruits with the same graphics card they developed with. Power draw was rated at around 280 W while gaming, with only 25 of those being taken from the PCIe slot, which seems somewhat disproportionate.
Apparently, there was some testing done on mining software as well, and performance is reported as disappointing (as in, "very low".) This probably speaks to the HBC (High Bandwidth Cache) and HBCC (high Bandwidth Cache Controller), which probably will require a lot of fine tuning from mining software (remember the GTX 1080 is generally poor in mining workloads compared to the GTX 1070 because of the higher latency of its GDDR5X memory implementation.) Perhaps these news come as a relief, however, since availability of RX Vega cards will likely be limited without miners taking up the whole supply.
Sources:
Ekin @ Linus Tech Tips, PC Perspective, PC Perspective YouTube Channel
Starting with 2560x1440, let's begin with the good news: in what seems to be the best performance scenario we've seen until now, the Vega Frontier Edition stands extremely close to NVIDIA's GTX 1080 Ti video card in Fallout 4. It trails it for about 10 FPS most of the test, and even surpasses it at some points. These numbers should be taken with a grain of salt regarding the RX Vega consumer cards: performance on those models will probably be higher than the Frontier Edition's results. And for the sake of AMD, they better be, because in all other tests, the Frontier Edition somewhat disappoints. It's beaten by NVIDIA's GTX 1070 in Grand Theft Auto V, mirrors its performance in The Witcher 3, and delivers slightly higher performance than the GTX 1070 on Hitman and Dirt Rally (albeit lower than the GTX 1080.)
Apparently, there was some testing done on mining software as well, and performance is reported as disappointing (as in, "very low".) This probably speaks to the HBC (High Bandwidth Cache) and HBCC (high Bandwidth Cache Controller), which probably will require a lot of fine tuning from mining software (remember the GTX 1080 is generally poor in mining workloads compared to the GTX 1070 because of the higher latency of its GDDR5X memory implementation.) Perhaps these news come as a relief, however, since availability of RX Vega cards will likely be limited without miners taking up the whole supply.
166 Comments on New Performance Benchmarks of AMD's Vega Frontier Edition Surface
The real problem here are the sheer number of people who never read these professional card descriptions are suddenly doing so, and trying to apply their views onto something that is foreign to them.
As an example, I've played plenty of mods in different games in which the creator had a low level GPU. Does that mean I cannot crank up awesome details on what he or she produced? Nope. He provides that capability and only needs to test that is works as intended, not see it at the same level as I can.
No, they just make the program and make sure it works as intended as they go along with enough detail for them to see their work. The tests of their implementation of the game engine comes later with gaming cards.
Other than performance, it doesn't have any of Quadro bonuses - an existing pro support for example.
And BTW: is AMD planning a Vega-based Radeon Pro? What about the FirePro line?
Will there be a competitor to GP100 (i.e. high FP64)?Maybe AMD said it wasn't a gaming card, because it doesn't perform well in games? Sadly, it's also not good at anything else. It's just "some card".
And some here are even praising AMD officials for openly saying that this card is awful and they should wait for RX. Just how twisted is that? :eek:
AMD lineup becomes a mess. They had gaming cards and pro cards. It all worked. The cards even used to be pretty good few years back.
Now they released a new segment, which is so confusing that even they got lost (just check their websites).
So when you compare Vega FE to similar cards based on Pascal, Pascal wins.
But you can't compare, because Vega is different. It's so different that it sits alone in it's cave of misunderstood and overlooked greatness. They should borrow competitor's naming convention and call it AMD Copernicus. Of course other than the fact that Copernicus did something forever important and this is just a card.
If NVIDIA chooses to join the race and make a Pascal-based "pro Titan" (which - looking at the Vega benchmarks - would only need a rebranding) we'll finally be able to officially call Vega s..t.And until now they were using normal gaming GPUs. Why do we need a "pro" card for game development? What does this card offer over 1080?
Of course, it could be that your premise is flawed, and that a number of game devs are not using gaming cards, but some form of pro card.
So why has AMD done this with the drivers for Vega FE? Because it has little to no impact on non-gaming performance or professional workloads. It just, intentionally, gimps Vega FE for gaming purposes. And they really do NOT want Vega FE to appeal to gamers(not that it would at the $1000+ price point anyway). They need gamers to buy up the RX Vega when it's released. By that time the drivers will suddenly be "fixed". DBSR will work with FE and RX. FE owners will get a nice gaming performance boost as a result. But in comparison to how well RX is going to perform it will still not justify having bought an FE for gaming. So there, hopefully, won't be any pissed off gamers wishing they'd saved their money for later and bought the much cheaper and better for gaming RX, instead of the FE. Because they were repeatedly warned ahead of time not to buy the FE for gaming. And shown very convincing benchmarks proving they shouldn't. If they did it anyway, then that's their own dumbass fault and they have nobody to blame but themselves. And that's why AMD is doing this. To keep gamers, and Vega owners in general(FE and RX), happy. Plain and simple. Because Vega FE will never game like the RX Vega will. They know already that. And they're doing their best to keep gamers from making a bad decision by purchasing the Vega FE for gaming. And, all the while, still letting "prosumers" have a generous piece of the Vega gaming pie too. And they will be praised for doing so when all is said and done. And not talked shit about for not doing so. Because they did. It was the smart thing to do. It was the right thing to do. And in the end = happy gamers, happy Vega owners all around(FE and RX), happy AMD. Everyone's happy.
Sounds stupid to you? Well it's a good thing I'm not asking you then isn't it. I'm telling you this is the way it is. If you don't believe it I don't really care. Time, and only time, will tell if what I've said here is true or isn't.
As for the second paragraph, thats the idea behind FE. Not to have to use a gaming card to see how the game engine REALLY performs. Let me be more specific here. How it really performs under a gaming card with the same specs, using the same architecture, with only frequencies probably being different.You need Instinct for this kind of jobs.
Radeon Instinct™ MI Series
I still maintain my stance of wait until the consumer cards have arrived and get benchmarked.
I vividly recall that the "realistic" guess was that it(the actual gaming card with drivers some months old) would preform between 1070 and 180 speed.
I see now that the FE with like alpha drivers is achieving those values already.
So add the extra speed of the Gaming edition and add the fact that it isn´t the actual decent gaming driver. Let this soak in some months of driver optimization (against the 1080 whose drivers are now solid and wont improve that much anymore)
- Gaming edition
- Drivers
- 1080 drivers already optimal
- AMD drivers age much better
- ...pricing(cost/performance)?
My guess is that the intel+nvidia combination will be beaten by the AMD+AMD combination :)
Clear win for AMD here folks
forum.hwbot.org/showthread.php?t=142320
The driver issues are clear.