Today we’re bringing you the confirmed specs for AMD’s upcoming Fiji powered enthusiast graphics cards. The Radeon Fury X and Radeon Fury. We’ve been working very closely with our sources to confirm the specifications for AMD’s upcoming Fury X, powered by the Fiji XT GPU, and Fury powered by the Fiji Pro GPU.
Needless to say these newcomers will be AMD’s fastest and most advanced graphics chips to date. The Fiji GPU, both the XT and Pro variants, form the foundation of AMD’s new ultra enthusiast “Fury” brand. There are several reasons as to why AMD decided to resurrect the “Fury” brand name with Fiji. Chief among which is that with Fiji AMD will be introducing the world’s first GPU featuring 3D stacked High Bandwidth Memory, otherwise known as HBM. Another reason is the sheer performance of the chip, which AMD has proclaimed to be the fastest in the world. To crown these achievements AMD decided that a simple numerical product name will not do this chip justice and thus will be bringing back the Fury brand name.
We have a lot of information to share with you today about AMD’s upcoming flagships. So without any further delay let’s get to it.
Let’s briefly discuss the differences between Fury X and Fury. Both cards feature AMD’s brand new Fiji GPU and stacked High Bandwidth Memory technology. Fury X will be AMD’s highest end offering, the Radeon flagship. The card is based on Fiji XT, the fully unlocked variant of the Fiji GPU. Fury is also based on Fiji silicon, however it will be powered by Fiji Pro a slightly less powerful variant with a few GCN compute units disabled. So in essence the relationship between Fury X and Fury is quite similar to the one between the R9 290X and R9 290 or the HD 7970 and HD 7950.
Another difference, we’re told, is that Fury X will come with both water cooled and air cooled reference designs. Fury on the other hand will only feature an air cooled reference design. However that doesn’t preclude AMD’s add-in-board partners such as Sapphire, HIS, Powercolor and so on from making custom water cooled versions of Fury. Only that AMD’s reference design is going to feature air cooling. Powercolor for example has shown a water-cooled version of the upcoming R9 390X at Computex called the Devil 13. And we wouldn’t be at all surprised if they launch a water cooled Fury. We’ll discuss the cooling solutions later in the article but first let’s take a look at the specifications of AMD’s Fury X and Fury graphics cards. As well as the two variants of the Fiji GPU powering them.
Fiji, and the Fury line of cards which are based on it, feature notable improvements across the board. Performance is obviously significantly improved. Fury X is faster than the R9 290X by a minimum of 54%. Which brings us to the second major improvement. Fury X achieves this performance improvement with a TDP that’s a meager 10W higher. Which makes Fury X 48% more power efficient than AMD’s previous single GPU flagship the R9 290X, which is quite remarkable.
We’ve also been told that since Fiji’s die is measurably bigger than that of Hawaii, it’s considerably easier to cool. Because the heat will be distributed across a larger surface area, which will allow it to dissipate more readily. This is good news, especially considering that AIBs will also be bringing out newer and more effective cooling solutions.
On an even brighter note, we’ve been told that AMD’s reference air cooler for Fury X and Fury is actually quite beefy. And will keep the chips cool even with an overclock. The cooler features three axial fans blowing air onto a large heatsink. A concept that’s been popularized by AIBs such as Gigabyte who have employed it for many years. And is quite similar to what Asus has shown at Computex with their brand new triple fan Strix cooler. Which the company has hinted towards being compatible with AMD’s Fiji GPU.
Going back to the technical specifications of Fiji. AMD has doubled the Render Output Units, ROPs, with Fiji compared to Hawaii. So even though the core count has gone by less than 50% AMD decided to double the ROP count. Which shows a major focus on performance at high resolutions. In which ROPs are taxed the most.
AMD has also managed to significantly improve the area efficiency. We don’t have an official die size for Fiji yet but we do have a fairly good idea about how large this chip is. And what we are able to conclude is that AMD has managed to improve the performance by over 50% with Fiji vs Hawaii. While only expanding the silicon area, i.e. size of the chip, by roughly 25%. Which means that AMD has achieved the holy trifecta of semiconductor engineering. Greater performance, greater power efficiency and greater area efficiency.
AMD’s launching the Radeon 300 series and the Fury series of graphics cards on June 16th at E3 and the event will be live streamed over the web. Until then we’ll make sure to keep you updated as more information is made available to us, so stay tuned.
Sounds good.
I'm pleased by the new reference air cooler,about time they redesigned it. Remains to be seen if it's good. Then I wouldn't have to wait for custom coolers.
I'm hoping the price for normal Fury is around 600-700 euros. But I'm guessing it will be around 800 and Fury X is about 1000 euros. Wait continues indeed....not long now.
There was some talk about the issue in TechReport's article on HBM:
Quote:
This first-gen HBM stack will impose at least one limitation of note: its total capacity will only be 4GB. At first blush, that sounds like a limited capacity for a high-end video card. After all, the Titan X packs a ridiculous 12GB, and the prior-gen R9 290X has the same 4GB amount. Now that GPU makers are selling high-end cards on the strength of their performance at 4K resolutions, one might expect more capacity from a brand-new flagship graphics card.
When I asked Macri about this issue, he expressed confidence in AMD's ability to work around this capacity constraint. In fact, he said that current GPUs aren't terribly efficient with their memory capacity simply because GDDR5's architecture required ever-larger memory capacities in order to extract more bandwidth. As a result, AMD "never bothered to put a single engineer on using frame buffer memory better," because memory capacities kept growing. Essentially, that capacity was free, while engineers were not. Macri classified the utilization of memory capacity in current Radeon operation as "exceedingly poor" and said the "amount of data that gets touched sitting in there is embarrassing."
Strong words, indeed.
With HBM, he said, "we threw a couple of engineers at that problem," which will be addressed solely via the operating system and Radeon driver software. "We're not asking anybody to change their games."
The conversation around this issue should be interesting to watch. Much of what Macri said about poor use of the data in GPU memory echoes what Nvidia said in the wake of the revelations about the GeForce GTX 970's funky 3.5GB/0.5GB memory split. If Nvidia makes an issue of memory capacity at the time of the new Radeons' launch, it will be treading into dangerous waters. Of course, the final evaluation will be up to reviewers and end-users. We'll surely push these cards to see where they start to struggle.
In any case, it's the same as always - wait for reviews and see how they will handle it in reality. I think there were already some benchmarks that showed that 4GB cards could struggle at 4K, for example, so there is something to compare the results with.
Though when will 4k be playable at 60fps with a single card?!
Since DX12 *should* double the vram in SLI/X-Fire, the 4GB per card would be sufficient for 4k gaming
Proud member of Frustrated Association of International Losers Failing Against the Gifted and Superior (F.A.I.L.F.A.G.S)
But that doesnt mean that you can automatically use double the amount of actual information. So it probably on how and what you scale with SLI/X-Fire, to see what assets need to be present. I mean its not like both cards should get completely different data, there should be some overlap, because they're rendering similar scenes.
Unless I'm missing something.
But that doesnt mean that you can automatically use double the amount of actual information. So it probably on how and what you scale with SLI/X-Fire, to see what assets need to be present. I mean its not like both cards should get completely different data, there should be some overlap, because they're rendering similar scenes.
Unless I'm missing something.
I think that's true, but then can't you just apply the logic of how much memory a game would need, in total, at running half the resolution of 4K? That is what the card would be rendering. It shouldn't be that much, I don't think. Then take into account not all the data would need to be in VRAM, so it should be less. If you get what I mean.
Ryzen 5 5600, ASUS ROG STRIX B550-F GAMING WIFI II, Corsair Vengeance RGB RT 32GB 3600MHz C16, MSI RTX 5070 Ti Ventus 3X OC , Corsair RMx Series RM750x. AOC AGON AG324UX - 4K 144Hz 1ms
But that doesnt mean that you can automatically use double the amount of actual information. So it probably on how and what you scale with SLI/X-Fire, to see what assets need to be present. I mean its not like both cards should get completely different data, there should be some overlap, because they're rendering similar scenes.
Unless I'm missing something.
According to AMD slides it should double:
Spoiler:
Proud member of Frustrated Association of International Losers Failing Against the Gifted and Superior (F.A.I.L.F.A.G.S)
@sausje It will double the available memory but he is talking about duplicate data being on each card. If two identical objects are on both halves of the screen then the data for the object will be in both cards memory.
Ryzen 5 5600, ASUS ROG STRIX B550-F GAMING WIFI II, Corsair Vengeance RGB RT 32GB 3600MHz C16, MSI RTX 5070 Ti Ventus 3X OC , Corsair RMx Series RM750x. AOC AGON AG324UX - 4K 144Hz 1ms
Yeah but it will share far lesser assets than before, where the same scene had to be fully rendered on each card. Plus it depends on the dev now according to these slides, they can optimize it properly and point out what card should load what. Technically, they could use the full 4GB on each card, making it 8GB total. Though that could mean bad things too if they trying to limit stuff...
Proud member of Frustrated Association of International Losers Failing Against the Gifted and Superior (F.A.I.L.F.A.G.S)
I'll have to disagree
Especially if those price points are confirmed.
For "4K" gaming 4 GB might be a bit low for upcoming games in 2016 and beyond but as mentioned I think it's a limitation of this first-gen HBM design with the second-gen allowing was it 8 GB or so?
(Which I believe is what Nvidia will start with when they move to HBM and I guess AMD's 20nm GPU might also feature this improvement.)
EDIT: Was it 20nm or was it 14 with finfeet?
(I think it might have been either 14 or 16)
@sausje Yeah it should, and I'm hoping, will be a massive improvement on the way VRAM works. Something like it should have been done a long time ago, but I guess there was no incentive for MS or Nvida/AMD to implement it. Now with SteamOS trying to compete, however. I wonder if the graphics card companies will push developers to use it or not, it's not like it would help sell new cards, just the opposite.
Ryzen 5 5600, ASUS ROG STRIX B550-F GAMING WIFI II, Corsair Vengeance RGB RT 32GB 3600MHz C16, MSI RTX 5070 Ti Ventus 3X OC , Corsair RMx Series RM750x. AOC AGON AG324UX - 4K 144Hz 1ms
I'm more scared of games that are "sponsored" by nvidia, them forcing devs to allocate more than 4GB resources per card and thus bottlenecking AMD hardware like that :\
You would see shit like: card 1 using 5GB and card 2 3GB, instead of 4/4
Proud member of Frustrated Association of International Losers Failing Against the Gifted and Superior (F.A.I.L.F.A.G.S)
No, you're looking at it wrong with regards to DX12/Vulkan multi-GPU stuff (which isn't enabled in the current drivers for both AMD and Nvidia yet btw).
Each GPU will only have in memory what it needs for whatever it's rendering. However, some of it will still be the same data. It won't be exactly doubled.
And HBM Gen 1 isn't limited to 4GB per se. AMD and SK Hynix developed a dual-link interposer, which allows them to do something between RAID-0 and clamshelling.
Clamshelling is how they put 8 GB memory on a 256-bit bus. On a bus of that size, you can only put 8 GDDR5 chips, as each chip has a 32-bit interface. So with 4Gb chips, that's 8 * 4 = 32 Gbit = 4 Gigabyte. With clamshelling, they put an extra chip on the other side of the board and they share those 32 lanes between two chips. The Titan X' 12 GB is in fact also clamshelled. It's a 384-bit bus; that means 12 chips maximum and up until January this year, there were no 8Gb GDDR5 chips (nothing uses them). It still uses those 4 Gb chips, so it needs 24 of them to get to 12 GB. When you go over 50% of VRAM usage on such a clamshelled interface, you will lose some performance, although not as much as the GTX 970's last 512 MB for example.
They haven't been very clear about how the dual-link interposer works, but it's not entirely like clamshell, obviously each HBM stack has a whole load more lanes available so apparently it borrows some concepts from RAID. At any rate, 8GB is possible on HBM Gen 1.
Not that 4 GB wouldn't be enough, especially not with how fast HBM is.
That shit comes from WCCFTech anyway. They also "confirmed" 8 GB HBM several times before already. Clickbait site
We can only thank Nvidia about the prices if they are true. But anyway, AMD Fury X sounds good, 4GB VRAM is not issue unless you run something fucking crazy like 4K (which 99% of humpers can only dream of). That resolution is anyway too demanding for any single card out there and you will need 2 high end GPU´s minimum to run it and with the upcoming DX12 VRAM is the least of the issues since you can stack it. I can already see guys coming to this thread and say LOLOL TITAN X has 12GB VRAM, lets face it, can the card run anything that could require that kind of VRAM usage?.
I have seen only one damn game that has given me issues with VRAM and that was Dying Light on GTX 970 GPU where 3.5GB got filled and that stuttering started on my 2560x1080 resolution, but with 4GB card there is no issues at all. Ofcourse you can max out 4GB if you run something shit like 8XMSAA, but there are far better and more efficient ways doing AA than that demanding shit. Personally im leaning towards 980Ti unless AMD is going to shock us all with the prices.
That shit comes from WCCFTech anyway. They also "confirmed" 8 GB HBM several times before already. Clickbait site
The slides are from AMD though, so AMD is bullshitting about double the amount of vram in multi-gpu too?!
It's not bullshit - you do get the option to use all the combined capacity of the GPUs. The problem is that realistically, it would be very difficult to split a scene in such a way that there would be absolutely no overlap in assets. So some parts of memory in each GPU will have same data stored on them, technically reducing total effective capacity, if you could call it that.
One has to understand that it's not some magic that doubles your VRAM.
The way it worked with Mantle (and I assume DX12 is pretty much the same in that regard) is that the memory of all GPUs is exposed for developers to use. Essentially the entire combined capacity is available for them to do whatever they wish with it. This means that while in theory you can double the amount of available memory with dual-GPU setups, developers are the ones to decide how to use it. This requires some work specifically targeted at optimizing rendering for multi-GPU setups and most would not bother when AFR "just works".
SFR is sort of a middle ground between AFR and some lower-level multi-GPU rendering, but you are unlikely to get near 200% performance scaling as with AFR just "out of the box", so it's not like AFR with double the VRAM. It would also need some work to be utilised efficiently.
I think that's true, but then can't you just apply the logic of how much memory a game would need, in total, at running half the resolution of 4K? That is what the card would be rendering.
That makes sense. Any ballpark figures as to how VRAM usage increases when you increase render resolution by a factor of 2 (strictly in one direction), which halfes the fov in one direction as well?
Can such a statement even be made
edit: fov should play a role
Breezer_ wrote:
I have seen only one damn game that has given me issues with VRAM and that was Dying Light on GTX 970 GPU where 3.5GB got filled and that stuttering started on my 2560x1080 resolution, but with 4GB card there is no issues at all.
So your showing that 4 GiB are utilized to at least 87.5% already and thus we dont have a problem?
I think that's true, but then can't you just apply the logic of how much memory a game would need, in total, at running half the resolution of 4K? That is what the card would be rendering.
That makes sense. Any ballpark figures as to how VRAM usage increases when you increase render resolution by a factor of 2 (strictly in one direction), which halfes the fov in one direction as well?
Can such a statement even be made
edit: fov should play a role
Breezer_ wrote:
I have seen only one damn game that has given me issues with VRAM and that was Dying Light on GTX 970 GPU where 3.5GB got filled and that stuttering started on my 2560x1080 resolution, but with 4GB card there is no issues at all.
So your showing that 4 GiB are utilized to at least 87.5% already and thus we dont have a problem?
I really dont know how much games really require VRAM, but caching and real usage are 2 different stories. What i have seen that next-gen console ports can straight fill VRAM, yet they run flawlessy (unless its that GTX 970 with its slow ass 512mb partition). Seems that devs doesnt know how to manage memory usage, since all they know is that games need to be played with controller and PS4 / Xbone has 8GB of unified memory available .
all depends on a price. if 800-1000 EUR is true, 4GB is simply unacceptable. you shouldn't have to worry about such thing as hitting the VRAM limit when you spend that much on a GPU.
all depends on a price. if 800-1000 EUR is true, 4GB is simply unacceptable. you shouldn't have to worry about such thing as hitting the VRAM limit when you spend that much on a GPU.
You also shouldn´t have to worry about maxing out games @ 60fps on a 1K € GPU advertised as "4K gaming card" with 1080p monitor, yet there are frame drops everywhere on these days shit ports. But yeah, that is true.
Ran on a single 980ti .. though, like with most impressive supposedly-single-card-honest-guvnor-totes-furreal super high-end visual powerhouse tech demos, it's not available to download or test for ourselves.
Ran on a single 980ti .. though, like with most impressive supposedly-single-card-honest-guvnor-totes-furreal super high-end visual powerhouse tech demos, it's not available to download or test for ourselves.
Not impressive. All I see is nvidia physx smoke and lots of chromatic aberration. oh and shitload of terring and poor performance for what is being shown
Guess we will never again see gfx jump like the one from dx7 to dx9. Stalker illustrated the change so well. And I miss those old demos which came with every new major card from nvidia and ati back at the time. Those ruby demos were so impressive
srsly.. I understand dof and motion blur(proper object based) but CA has to go. Even cheapest dslr's don't produce this effect now. My canon 700d don't have ca with it's stock lense unless I zoom in 15x on an edge of a picture
5080 | ps5 pro
Sin317-"im 31 years old and still surprised at how much shit comes out of my ass actually ..."
SteamDRM-"Call of Duty is the symbol of the true perfection in every aspect. Call of Duty games are like Mozart's/Beethoven's symphonies"
deadpoetic-"are you new to the cyberspace?"
Last edited by KillerCrocker on Wed, 10th Jun 2015 22:24; edited 1 time in total
Signature/Avatar nuking: none (can be changed in your profile)
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
I think you are right tbh.
