AMD Fury X Far Cry 4 game performance from AMD

[bogus]

Found my next GPU I think. Still wish they would of did a 4GB HBM + 4GB GDDR5 thing though for a total of 8.. but hopefully this wont matter too much!

 

Cant wait to get one when they get cheaper lol

Why would they have to complicate the architecture, probably to get worst performance and stutters (looks at 970 3,5GB+0,5), just to put a sticker in a box claiming it has 8GB?

If 4GB HBM is apparently performing better then 12GB GDDR5... why do you need that extra 4GB of GDDR5?

 

Swag?

[Doobeedoo]

Those are great numbers, nice! Can't wait to see it across other games.

[LAwLz]

Why would they have to complicate the architecture, probably to get worst performance and stutters (looks at 970 3,5GB+0,5), just to put a sticker in a box claiming it has 8GB?

If 4GB HBM is apparently performing better then 12GB GDDR5... why do you need that extra 4GB of GDDR5?

 

Swag?

I agree with you that mixing memory doesn't sound like a good idea, but you can't say 4GB of HBM will perform better than 12GB of GDDR5. Memory just doesn't work that way. For some things you need fast memory, and for some things you need a lot of memory. Increasing the speed of the memory won't help if the bottleneck is the amount of memory you got, and vice versa.

 

 

I just don't know how mods allow pages and pages of this non sense.

Same reason they allowed pages and pages of the GameWorks bashing non-sense, and 970 memory non-sense, and all the other times something happens in the GPU world. It always ends up being a shit storm because both AMD and Nvidia got a ton of fanboys who can't accept that their beloved company isn't perfect.

[d3sl91]

Great numbers.

 

I want to see

 

1. benchmarks from reviewers across multiple games. 

2. Overclockability. 

[SOLO_AVENGER]

I guess they decided to rebrand because they had too much stock left over from the 2xx series. But Fiji still looks promising, I'm interested in R9 Nano.

[RagnarokDel]

in the conference at e3 earlier today they called 4k 4096 by 2160

Who cares, that's more displayed pixel which means it should be more demanding and therefore if true, theses cards are beasts.

[parasemic]

I guess they decided to rebrand because they had too much stock left over from the 2xx series. But Fiji still looks promising, I'm interested in R9 Nano.

 

I'm pretty certain HBM is too expensive to create low-mid end cards at all and Fiji is strongly tied with the technology, so cutting Fiji down to create a new range of low-mid cards wasn't possible or made no sense in terms of cost or end user experience. Also it would make no sense to create even another GPU solely for low-mid range.

 

The prices are way too high, though. I was expecting 390X to undercut 970 but I guess the 8GB GDDR5 isn't free after all. Just tells the tale of AMDs profit margins in 2XX (and 3XX) series.

[zMeul]

5120x1440 is a lower resolution than 3840x2160. Try using a calculator.

only if VR would resume to the resolution [facepalm]

what about frametimes? framerate consistency above or at 60FPS - unimportant, eh!?

[Techhog]

only if VR would resume to the resolution [facepalm]

what about frametimes? framerate consistency above or at 60FPS - unimportant, eh!?

You get so aggressive in these threads. Chill bro, they're just video cards.

[Aniallation]

if that is true then amd may have a win,

but man those rebrands

 

So pretty much the Fury was the only good thing they released?

 

Same with last year... only 290/X were new, everything else was a rebrand.

 

If it ain't broke don't fix it I guess, at least until they can come up with a revolutionary lineup like Nvidia did with Maxwell.

 

Wonder if I can flash my 7870 (currently flashed to a 270X) to a 370.

[bogus]

I agree with you that mixing memory doesn't sound like a good idea, but you can't say 4GB of HBM will perform better than 12GB of GDDR5. Memory just doesn't work that way. For some things you need fast memory, and for some things you need a lot of memory. Increasing the speed of the memory won't help if the bottleneck is the amount of memory you got, and vice versa.

 

 

Same reason they allowed pages and pages of the GameWorks bashing non-sense, and 970 memory non-sense, and all the other times something happens in the GPU world. It always ends up being a shit storm because both AMD and Nvidia got a ton of fanboys who can't accept that their beloved company isn't perfect.

Well I think there's still more information to be seen (and surely tested) about HBM, and how it manages the 4GB. AMD since a long time they always market for a new resolution when new graphics cards come out - so for them to brag about it at 4K and even 5K, maybe there's more to it then we know.

About the rest - I just think that if not delete, they should move such flame wars to a "Drama" section, where all the hate could be release.

[79 Potatoes]

So AMD managed to pull another Titan killer out of their asses. And if its cheaper than the Titan X.....

And it's a single GPU card

[Sauron]

only if VR would resume to the resolution [facepalm]

what about frametimes? framerate consistency above or at 60FPS - unimportant, eh!?

Frametimes and framerate have nothing to do with vram quantity, and I'm pretty sure two of these could easily stay above 80 fps. And yes, vr is just as hard to run as any screen with the same resolution. The scene splitting is mostly done by the cpu and occupies no extra vram at all - in fact, with proper optimizations it should take LESS vram because some assets are shared between the two screens.

Wonder if I can flash my 7870 (currently flashed to a 270X) to a 370.

why would you though?

[Elapo]

If this is true and not AMD telling us it's Ultra while it's not, I'll be damned. Glad they can finally push back Nvidia, they needed this. We need this competition. Bravo

 

Yups, would make me excited to see what Nvidia's answer is

[Dabombinable]

exactly

 

and my point was also this: because of the 4GB limitation, AMD made a mistake using HBM1

they should've kept using GDDR5 in a 8+GB configuration with a good price point, since they advertise the Fury for VR - that 4GB limit will start to be a real problem very soon

they should've waited for HBM2 or a new way to implement HBM1 that didn't limited them to only 4GB

why I say that: it very much appears that nVidia's Pascal with HBM2 is 1st headed to HPC market and not at desktop graphics market; it may be very late 2016, or ealy 2017, when we'll catch a glimpse of Pascal for PCs

Your not even joking are you? System only counts when the graphics card runs out of spare vRAM. Which so far hasn't been the case with the Fury X, apparently because of using HBM instead of GDDR5.

[Dabombinable]

And it's a single GPU card

No dual GPU card was needed to kill the original Titan-the R9 290X did a fine job on its own.

[zMeul]

Your not even joking are you? System only counts when the graphics card runs out of spare vRAM. Which so far hasn't been the case with the Fury X, apparently because of using HBM instead of GDDR5.

is HBM magically increases in size? NO!

so where did you even got that idea?  :blink:

[Dabombinable]

is HBM magically increases in size? NO!

so where did you even got that idea?  :blink:

Its been said before. Above 4GB of vRAM is not needed at 4K with HBM due to the sheer amount of bandwidth. However with GDDR5 8GB is needed because it has lower bandwidth and therefore can't have data transferred as rapidly. Its no coincidence that an R9 290X 4GB with higher memory bandwidth than a GTX 970 does better at higher resolutions.

[Notional]

is HBM magically increases in size? NO!

so where did you even got that idea?  :blink:

 

Alpha compression however magically decreases the data size. Like AMD said, their engineers, has been focused on better memory management, so this won't be an issue. Gotta ask you though: Did NVidia make a mistake with their 980 and 970 with only 4GB and 3,5+0,5GB of vram? (ok the latter one was a dumb idea).

[slay english]

I hope it wrecks 1080p. 

[ItsAFeature]

Its been said before. Above 4GB of vRAM is not needed at 4K with HBM due to the sheer amount of bandwidth. However with GDDR5 8GB is needed because it has lower bandwidth and therefore can't have data transferred as rapidly. Its no coincidence that an R9 290X 4GB with higher memory bandwidth than a GTX 970 does better at higher resolutions.

 

Potentially. But if developers insist on loading over 4GB into VRAM at the same time rather than streaming it, it won't help to have HBM. Between DX12 and HBM, game developers have some stuff to learn.

[Dabombinable]

Potentially. But if developers insist on loading over 4GB into VRAM at the same time rather than streaming it, it won't help to have HBM. Between DX12 and HBM, game developers have some stuff to learn.

There is that. But part of the current problems are quite literally the use of AA at 4K. Which is pointless.

[Opcode]

is HBM magically increases in size? NO!

so where did you even got that idea?  :blink:

HBM works a bit different then GDDR5 with a much wider interface, double the command input, single bank refreshing and a much shorter RLC. This is why you see a lot of people are saying "4GB HBM acts like 6GB of GDDR5". We won't know these improvements factor into gaming performance until we get Fury X on a test bench and run some numbers. Personally, I don't foresee 4GB of HBM being of any limitation. AMD went with HBM simply because GDDR5 is becoming too slow to accommodate the GPU. Hawaii was already pushing the limits of GDDR5 on a 512-bit bus. How many more modules of GDDR5 do you want to see on a graphics card in order for it to keep it fed properly? Another thing to keep in mind is GCN 1.2 brought forth delta compression which allows for the frame buffer to be compressed. So there's going to be fewer bytes per frame allowing the GPU to use what it needs to and to then get rid of it much quicker (shorter RLC). Needless to say through all of the jabber I would urge anyone who thinks 4GB is not enough to go look at GTX 980 SLI vs TITAN X 4k benchmarks. Them alone will shut a lot of people up about this whole density situation.

 

Potentially. But if developers insist on loading over 4GB into VRAM at the same time rather than streaming it, it won't help to have HBM. Between DX12 and HBM, game developers have some stuff to learn.

By the time games get that demanding (outside of the scope of running on current mainstream hardware) Fury X will be irrelevant. We'll have the next best thing in the form of Pascal and Arctic Islands both packing 8GB or more of HBM2. AMD is dropping a 4GB Fiji right now because they know 4GB is not a limitation at this point in time and when it becomes a limitation then it will turn into that you should be running our new model Fury and not the one from last year.

[Stagea]

A synthetic bench seems to show that a 4GB buffer will be really short at very high resolutions (in this case, 5K and 8K):

[Derranged]

A synthetic bench seems to show that a 4GB buffer will be really short at very high resolutions (in this case, 5K and 8K):

No card has the proper horsepower to run properly at either 5K and 8K, and if you do play at those resolutions you will have two or more of these cards. My point here is with DX12 memory pooling it will allow for 8 to 16GBs of HBM to become available.