AMD Radeon R9 300 Series MegaThread: FAQ and Resources

Page 20 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
So basically mobile apu's? Alright so judging from all this nvidia's not gonna have a new card until pascal next year. Yeah i was wondering why i'd never heard of this new socket. I know 900 series architecture is maxwell, and the one before that was kepler. It just seemed strange i'd never heard of another gpu like tegra before, thought it was some sort of nvidia dooms day weapon or gpu. A break glass in-case of type thing. That said, i do have another question pertaining to amd's fury x.

So it say's that fury is going to be using new 4096 bit memory interface on account of hbm. Would my mother board be able to handle that? I only have 2.0 gen port, and i do realize that 3.0 cards are backwards compatible with 2.0 pcie ports. But would my mobo not be able to handle that influx of data 8.5 teraflops seems a bit much. Though i might be thinking of something entirely different.
 


It looks like Fury X is going to be a little better than a Titan X. It's not a second coming of GPU horsepower. It's just the typical progression. 20-50% increase on the previous generation. If your board can handle a Titan it will be able to handle a Fury X.
 

not necessarily. thanks to the scalability of gpu architectures, nvidia can release a card between a titan and a 980ti. it can also release a dual titan gfx card using two GM200 gpus. too early to tell.

not a socket (not sure you mistyped or not..), an SoC. search in wikipedia, you'll get plenty of info on socs as well as on tegra.

yes, 4096bit memory bus is possible. it's made possible through chip stacking technology and interposers. simply put, the memory chips are "soldered" very near the GPU itself i.e. on the same substrate where the GPU sits. the memory wirings go directly from the GPU to the memory dies through a base logic die and the whole thing sits on the main gpu package and communicate through a passive interposer.
since the gpu and video ram are packaged together, the gfx Card pcb design is simplified. the pcb has the power delivery and other components but it doesn't have to hold the RAM chips anymore.
the memory bus width won't affect motherboard compatibility. you should be able to run a fury x card without issues as long as your motherboard, case, power supply etc. are adequate. make sure to check the card dimensions and fitment with your pc beforehand. we don't know yet if or how fiji's bandwidth will affect pcie 3.0's throughput though it may be unlikely.

your profile info says you're running an i7 4770k, that'd mean your motherboard very likely has a pcie gen 3.0 slot for gfx cards. only way your mobo will have a gen 2.0 slot is if your mobo is one with h81 chipset and a pcie gen 2.0 slot.
 


GM200 will be the best nvidia have to offer until pascal arrive. if they want to fight it off head to head with Fury then they have to increase their clock more and possibly disabling their 'limiter' to open maxwell true potential. reviewer like hardocp has mention that the cards like 980 potential was blocked with nvidia imposing restriction to their cards.
 


You have a lot of mixed thoughts and concepts there, I believe.

1.- Tegra is an ARM based System on Chip (SoC) that is aimed to mobile and ultra portables / low power devices. Think nVidia Shield and tablets. They are basically a full computer on a chip (or close, really) that nVidia has developed using another CPU interface (ISA) from a competitor of Intel and AMD called ARM (hence the "ARM based SoC" moniker). The GPU portion of this SoC is a *very* stripped down version of regular GPUs. In fact, the architecture is different and they'll be in sync when Pascal arrives (or so I remember). So you could link Tegra GPUs to Desktop, full blown, GPUs from a uArch point of view. These SoCs are not socket compatible with anything on Desktop side; or at least, I haven't read, heard or seen any adapters for putting a Tegra SoC inside a PCIe port inside a regular PC.

2.- The memory sub-systems of a Video Card is enclosed to the actual AIB (Add-in-Boards) and is different from the CPU & MoBo memory sub-system (SDRAM). In particular, current crop of Video Cards uses GDDR5 which is a form of SDRAM derived from DDR3, but tweaked for high bandwidth and computers use either DDR3 or newer DDR4 nowadays. In AMDs case, they co-developed a new type of memory sub-system that goes by the acronym HBM and it has a very high bandwidth and a very wide "lane" connection to the actual GPU. The number of "lanes" of any memory sub-system is usually linked to the maximum theoretical bandwidth each memory tech can achieve per generation. I won't go any deeper, but the idea is that HBM equipped cards have nothing to do with *how* you plug in your Video Card inside your computer. That is called "PCI Express" slot and they're somewhat universal for computers; in particular, all Video Cards connect to computers using this standard and it has nothing to do with the memory sub-system of a Video Card.

3.- PCI Express version has to do to how fast you move data across the motherboard (sort of speak), so it doesn't matter if the video card has 100 PetaFLOPs or 1 FLOP, as long as the PCIe is not saturated with data, the performance of the card won't vary. At least, not by a significant amount. Now, you do have one thing right. If the memory sub-system is *way* too fast, the Video Card *might* suffer from data starvation if it relies too much on PCIe communication with other sub-systems of the computer. In particular RAM. This shouldn't be noticeable until we're playing on 8K resolutions and should only show for multi Video Card systems (XFire or SLI). It would be interesting to see how it scales though.

Cheers!
 
AMD Radeon R9 Fury X With Fiji XT GPU Detailed- Specifications, Fiji GPU, PCB Shots, Cooling Design Pictured and Detailed
http://wccftech.com/amd-radeon-r9-fury-x-official-specifications-pcb-shots-cooling-design-detailed/
AMD Radeon 300 Series Officially Launches - R9 390X, R9 390, R9 380, R7 370 and R7 360 Performance, Specifications Detailed
http://wccftech.com/wipamd-radeon-300-series-officially-launches-r9-390x-r9-390-r9-380-r7-370-r7-360-performance-specifications-detailed/
AMD Radeon R9 Nano Officially Detailed - Features Fiji GPU With 175W TDP, Single 8-Pin Connector and SFF Design That's Faster Than Hawaii
http://wccftech.com/amd-radeon-r9-nano-detailed-features-fiji-gpu-175w-tdp-single-8pin-connector-sff-design-faster-hawaii/
AMD Project Quantum Dissected - Powered by Radeon R9 Fury X2, Core i7 Processor and Fully Liquid Cooled Concept PC
http://wccftech.com/wipamd-project-quantum/

AMD Radeon 300-series explored
http://hexus.net/tech/news/graphics/84089-amd-radeon-300-series-explored/
AMD shares some R9 Fury X Far Cry 4 UHD performance charts
http://hexus.net/tech/news/graphics/84080-amd-shares-r9-fury-x-far-cry-4-uhd-performance-charts/

AMD Exposes Fiji to the World: HBM for the Enthusiast
http://www.pcper.com/reviews/Graphics-Cards/AMD-Exposes-Fiji-World-HBM-Enthusiast

AMD releases specs, pricing on Radeon Fury, showcases Project Quantum platform
http://www.extremetech.com/gaming/208461-amd-releases-specs-pricing-on-radeon-fury-showcases-project-quantum-platform
more of the same... a bit differently..
 


Yes'r! Yahoo's the main financial reporter on the web. You can overlay the charts also. Those few cents add up if you bought in at the low I was talking about earlier. It the price gets back up to and historical spike, even 25% of such a spike, I'll pay off my grad school debt! But the greatest part is, if nothing happens, it's still pretty safe given that they're not going lower unless they go out of business, which is highly unlikely in the short term. C'mon AMD!

Google Finance is also pretty good

https://www.google.com/finance?cid=327
 
Sheesh...I was hoping for better performance in Witcher 3 4k res. Drat. Will definitely be holding off for another model. *scratches 390x off the list*



 
http://www.techpowerup.com/213575/amd-fiji-silicon-lacks-hdmi-2-0-support.html

Do you guys think that sapphire or others will do better reference cards that actually come with hdmi 2.0 support? I currently own a 45 inch 1080p tv, which i plan on replacing with a 4k tv in the near future. If i can't get 60 hz on it this is disappointing. Or does the 30hz only pertain to amd's first revision refference card. I really can't make an investment into a card like this if i can't use it to play 4k on a tv.

This is pertaining to fiji cards as in fury x, right? I mean if i'm dropping 650 on a card the least they could do is put in hdmi 2.0 right?
 


nope, well for me it is, i'm still using a gigabyte oc 3gb 7950 windforce, i believe the single percision floating point is around 2.79 tflops to 3.3 tflops. Going from that to 8.6 i believe i'll be seeing some pretty massive gains.
 


Yeah i got my socket types mixed with chips. i also got the numbers for memory bandwidth and bus mixed up. Because i'm fairly certain a card with a bandwidth of 4 gigs wouldn't be able to work on a pcie 2.0 port. Yeah the pc in my description is for a different pc i use. The one i'm on currently is fairly old, from a time when intel still made motherboards i think it's an H57 set. Only 2.0 sadly.
 

we have to wait until reviews come out. i think that.. fiji won't saturate 8 lanes of pcie 3.0. amd is saying that managing HBM will be different from managing GDDR5 (from developer's perspective) so 4GB may not be a limitation when dx 12 comes out. in reality though, majority users are running dx11.x and 4GB will act as a limitation in certain scenarios (expect the green team to blow these out of proportions).

did you mean H67 chipset? sounds like sandy bridge cpu. those will have pcie gen 2.0.
 


Nope it's an h57 chipset with i believe the old lynnfield cpu's. Released around 2010. Nehalem micro-arch, before Westmere and sandy bridge. I wanna say it was a weird off shoot instead of a real tick or tock, during the early days of the i7 800' series cpu's.
 

my mistake. i totally forgot about the h55, h57 and p55 chipsets.
 


What you need is 2 of them...

 
Was running dual 290x in crossfire, still only getting 45ish fps average in witcher 3 at 4k. I need to see 60fps, makes a huge slick difference in the gliding look on my samsung 8550 when i drop down to 2k res at get a smoothed 60fps. Not sure if that happens at 4k as well but if it did, I would pay up the butt to never see 50's. Problem is that the 8550 doesnt have a display port, so i guess im SOL :[