AMD Radeon R9 300 Series MegaThread: FAQ and Resources

Page 18 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.


Wut juzt happened?! I'm an Nvidia user (currently), but this card looks like it's going to destroy, and I hope it does!

This marks one of less than five time AMD came to market with a significant tech before its competitors. Five-times the computer performance of Titan Z is insane! Stoked to see how this unfolds 😀

Shows Titan Z's computer performance -> http://www.extremetech.com/computing/183242-nvidias-titan-z-arrives-titanic-compute-but-at-3000-the-amds-r9-295x2-is-a-better-choice
 

dual gpu/multi gpu/ crossfirex has always been a weak spot from amd's driver side. even though the hardware is crazy powerful (so was r9 295 x2, 7990 before that, and so on) timely, showstopping bug-free driver support will be absolutely important. mind you dx12 is still a month away and windows 10 will be raw and brand new (massive updates post launch), most people (including high end gfx's target demo) are using ye olde windows 7 and 8 OSes. if amd can't follow this up with driver support, nvidia will kill in publicity.

afaik, amd has always gotten newer memory. nvidia seems to get new gpu tech though.
 


Even if no support, the Fury X shows promise. Plus the application for HBM. I hope they patented the $|-|iT out of that, so they will make money on licencing it out. Everything before has only addressed 8 bits, but HBM address 1024. That's way out there.
 

i don't know about patents. both amd and sk hynix follow jedec's standards for implementing hbm. nvidia will combine hbm with it's new pascal lineup.
 


Also a good thing. That means GPUs will get real fast, real quick, and it also give AMD time to make some money (maybe) and keep up the competition. Pascal isn't coming out any time soon. Maxwell's still not even matured out yet. It's got at least another year before getting to the GM210 chips, which is traditionally the most mature in an architecture (Gx210).
 

yup, gpus wil get real fast. but how soon ... depends on the foundries. all of them crashed and burned (sorta) with 20nm, so that affected both amd and nvidia, and may affect the bigger maxwell. nvidia might internally cancel gm210 and move to gp100 instead. both amd and nvidia gushed at finfet processes' promises. as long as the foundries don't screw up we might actually see the new gpus later next year.
 


If AMD's stock is still capable of spiking (who knows), now seems like a good time. All other major spikes in the company's history have been from AMD's major first-to-markets. HBM is a major first-to-market. All that bandwidth should keep it on the forefront for several years. At the very least, AMD won't fold. Investing some money in them appears right now to range from "safe" to "great timing." I think I'll take the gamble on them. The worst case I see is pulling out without any real loss. Interesting.
 


Comeback for the high-end market, yes, but I feel Nvidia will pass them up in the low and mid market. 950Ti could be a 370 killer for sure. But Fiji is fantastic, definitely an Nvidia killer.
 


I think we just saw it. The question becomes, "How long will it last?"

Also, hi-end is where the money is. There's a really low margin in lower-tiered hardware. Moving to Fury will be good because AMD can phase out the R9 200- and 300-series stuff to streamline manufacturing with HBM across the board. Right now, they have to manufacture two processes, which gets expensive. They'll have to phase in the new Fury line, which will let them saturate the lower markets with a single process. Smart.
 
Before declaring Fiji a winner and the second coming of Jesus Christ, wait for benchmarks. Plus, keep your expectations in check.

Also, the market that is most profitable is not the high end. It's the mid 200 USDs. Anything above 300 is "premium" by all definitions of the word. You have to compare what 300 USD is in terms of computer parts. 99% of the builds your friends have and think of are targeting 1K USD. That means, the more expensive the Video card, the less money for all other components and that's not a trade off most of us are willing to accept.

If you think down the road, 4K will become "mainstream", just like FullHD (1920x1080) screen became mainstream a good 3-4 years after the first ones became readily available (2005-ish I believe and 2010 was mainstream). That means, 4K+ still has some time left for it to happen. Current 4K monitors that are half decent are still over 400 USD. Even 1080p monitors that have 120Hz+ are still around 200ish USD.

So, in short: wait for benchmarks before declaring Fiji a winner and keep expectations in check so you don't get from lover to hater in 5 minutes.

Cheers!

EDIT: Typo.
 


Preliminary synthetic benchmarks are already out, and it looks okay.

http://wccftech.com/amd-radeon-r9-fury-r9-nano-fury-unveiled-fiji-gpu-based-hbm-powered-649-priced-small-form-factor-powerhouse/

Also, I'm not the one saying low-end makes less money than high-end. It's the GPU companies' CEOs.



http://www.tomshardware.com/news/amd-ceo-lisa-su-interview,29327.html

Last, I'm not worried about hating AMD. If their stock doesn't go up, no big deal. I'll just sell it back. That's the magic of publicly traded companies :)
 
And here they are people. AMD unveiled them today.

radeon-r300-100591487-orig.png



http://www.pcworld.com/article/2936630/amd-reveals-hbm-powered-radeon-fury-graphics-cards-new-r300-series-gpus.html

"The $199 Radeon R9 380 was designed for 1440p gaming, AMD says, and packs up to 4GB of memory. Meanwhile, the $329 Radeon R9 390 and $429 Radeon R9 390X each pack 8GB of RAM, presumably for a better gaming experience at 4K resolution.

All of these new R7 and R9 300-series graphics cards will available to buy Thursday. That imminent launch is actually a bit worrisome. For the past few months, the rumor mill’s been adamant that the new 300-series Radeon GPUs are actually built around barely-tweaked GCN silicon that first made an appearance in the Rx 200-series and even the older Radeon 7000-series graphics cards. In other words, these may not actually be “new” graphics cards.

Even though the new cards are launching a mere two days from now, PCWorld hasn’t received review samples for any of the R300-series cards (or the Fury X, for that matter). Considering that, the rumors, and the lack of technical details provided for these new 300-series graphics cards, you’ll definitely want to wait for reviews to hit before you pick up one of these, even if you could buy one Thursday."
 


That quote is for APUs/CPUs and not GPUs.

To put it into context, the price of the actual GPU is big, but the actual cards (AIBs) have a lot more components than just the GPU AMD sells to the OEMs. The CPUs and APUs on the other hand, rely on AMD for production, so they can position them a tad better and get more per unit sold. In other words, Lisa Su is totally correct there. Selling a 300 APU/CPU is on another "profit" league than a 300 Video Card. To have the same profit margin per GPU they have to go even higher with Video cards, but there's not many people out there willing to spend more on a GPU and even CPU past the 250 USD mark. There's also the context of market share. Even though nVidia still dominates the GPU "gaming" market, the disparity is not as awful as in the CPU space with Intel. AMD has been cornered to a position they might be getting pennies per APU/CPU sold while Intel reeks in the cash like a broken dam flooding the desert.

Also, it's not the same when you talk about "higher profit margin market" than "most profitable market". In the consumer or even Pro level markets, most expensive stuff usually reeks in more cash per unit sold, but you also have quantity. So, there's usually a trade off. AMD wants to get a hold of the high profit margin market now, in both CPU and GPU.

Cheers!
 
So GPU battle get more interesting? I just hope AMD can bring the same excitement to mid range and low end. Fury nano look impressive. So does Fury have support for 12_1?

Edit: i've look at the official spec for Fury. Only seeing single presicion performance. Did AMD mention DP performance for Fury?
 
Doesn't matter if Fury beat titan x or not. Because those are not in my price bracket. Hoping to see something interesting on the mid range but all we have is tonga and pitcairn all over again. I will hold onto my 960 until16nm gpu are here 😀
 
E3's First Ever 'PC Gaming Show' Features Loads Of New Titles
http://www.tomshardware.com/news/e3-first-pc-gaming-show,29394.html
this source claims the dual fiji card will have 16 TFLOPS /edit

AMD unveils R9 Fury X, Fury, and Nano graphics cards
http://arstechnica.com/gadgets/2015/06/amd-unveils-r9-fury-x-fury-and-nano-graphics-cards-2/
better pics

SK Hynix confirms mass production of first-gen HBM memory
http://www.kitguru.net/components/graphic-cards/anton-shilov/sk-hynix-confirms-mass-production-of-first-gen-hbm-memory/

Alleged AMD-Radeon-R9-Fury-X-Specifications
http://cdn3.wccftech.com/wp-content/uploads/2015/06/AMD-Radeon-R9-Fury-X-Specifications.png

/edit2


DP performance will likely be crippled for the consumer cards...1/32th or something. i am expecting same spec as r9 290x.
 
There is no limitation with tahiti. Both the firepro and radeon version have the same DP performance which is 1/4th of tahiti SP performance. Hawaii DP was rated at 1/2 of it's SP exceeding even GK110 theorical max DP but 290X the DP performance was cut down to 1/8 making even 7970/280 better in DP compared to 290X/290. It will be interesting to see how Fury goes as a pro card espcially in HPC.
 


Yes, I agree.
The 980 Ti has a serious 6GB GDDR5, the modern Maxwell architecture and a solid 2816 CUDA cores and a fast memory clock. It's practically a Titan X.
But then you have the HBM-wielding Fury X! With blazing memory speeds, 4096 SPUs and a revolutionary new Fiji core.

The stage is set. This is gonna be a good one - grab your popcorn, people. Two cards. One price. One winner. Now we wait for benchmarks.