Long post incoming.
Originally Posted by shadow85
I need faster GPU's, 2x GTX 980 gamings is not giving me smooth frames @ 4K. So hurry AMD and release this beast so I can throw my money at 2 of them!
The 980 is literally
designed to dominate at 1920x1080. It isn't designed to work at 4K at all. It isn't designed to handle surround-gaming. It's for 1920x1080 and 1920x1080 only, and at 1920x1080 it's untouchable (at this time). Increase the resolution and even the R9 290 and the 770 will win. This is why a lot of people are (rightfully) calling the 980 a mid-end card; it is.
Originally Posted by CynicalUnicorn
2(512 bits * 5GHz) = X bits * 1GHz
X = 2(512 bits * 5GHz)/1GHz
X = 5120 bits
In a single cycle, that's 640 bytes or 62.5% of a KiB. Not bad! I wonder if latency will be an issue since the frequency is so much lower. Something like a single optical fiber has very low latency and bus width, but a truck full of hard drives has a much higher "bus width" at the expense of your ping times.
I have no idea what you where trying yo calculate here. The Fiji (R9 390x) is estimated to have a memory bandwidth of 550GB/s++. The R9 290X has a bandwidth of 320GB/s, the 780ti 336GB/s and the 980 224GB/s. This is what matters. It doesn't matter if its 256b, 512 or 4096 bit wide, all that matters (for absolute performance) is the raw bandwidth. From a power-performance point of view, you want a wide and slow buss, but that increases the complexity of the PCB a lot, and that drives the BoM up. HBM reduces the added cost a lot. This is also why HBM will lower the power-usage while increasing the performance.
Originally Posted by mtcn77
If this Fiji XT is indeed the new mid range as I assume it is, soon it will be lolfest when it comes out.
I don't know where you got the idea that Fiji is mid-range..? There was this one article a few weeks ago that had 2 of the new AMD GPUs and two of the new nvidia GPUs, but that article isn't solid. Do you really think that someone would get hold of ALL the next-gen cards at the same time, and
have time to bench them against each-other? That's a lot of stars aligning just perfect; too many for it to be believable.
Fiji is a top-end card (R9 390x) and will be with us Q2 unless there are further delays. Expect nvidia and AMD to launch within weeks of each other.
Originally Posted by Just a nickname
Well, they let the reader guess they are not sure at all of what they're saying. Look at the table, it's not even indicated. So either they had some info and try to attract more traffic by including 20nm or they just made this all out using the shipping data as supporting evidence of their claim.
They did what is commonly called writing a fairytale. They might be interesting and entertaining, and might even have some life-lesson and/or morals, but in the end, it's fiction. Nothing more, nothing less. See here
for a great article from Fudzilla about why there will be no 20nm GPU* from neither AMD or nvidia.
*dedicated GPU. APUs might be made at TSMCs 20nm node, but no high-end GPUs. Read the article, keeping in mind Fiji is expected to be 550mm2+ at 28nm or about 400mm2+ at 20nm.
Originally Posted by iamhollywood5
As far as the 20nm stuff goes, I really can't understand why AMD would go this
long between flagships if they're just putting out another 28nm gpu. I mean really, it would be pointless. If AMD wanted to do another 28nm GPU it would have been released already. If they've sat on the 290X for this long, it's because they're waiting for 20nm.
They wouldn't if they could choose. As it stands, and have been known for some time, TSMCs 20nm node is useless (yields) for any chip larger than about 200mm2. It takes at least 9 months from finished, bug free
design to launch-volume of said chip is possible. This is if every step of the process goes without problems or delays. 2nd rule of silicone, there is always a delay. This put AMD in the spot where they decided to launch the R9 285, in order to do some extended testing of the 28nm node, checking the yields of the new core design (GCN 1.2) and test some other new features (multi-orientation eyefinity). Sadly the performance of the 285 was below what was expected, and the price wasn't low enough to compensate. If it would have been called (and priced as) R9 275, I believe it would sell as fast as they could make them (remember the ATi 4770 or the 8800GT?). That way, when the next generation launches, it will be as smooth of a launch as can be (AMD 5000 series launch).
In the end, AMD was already more than ready to launch a GPU with HBM, and 20nm wouldn't work with that at all, see the Fudzilla article above, but they had to make it work at 28nm, so they did. Launching with both HBM and 16nm was considered to large of a gamble, so they had no choice. It was 28nm + HBM or no new cards at all.
Originally Posted by Leopard2lx
This might beat the 290x by about 15% at best....20% if lucky....maybe edge out the 980....
Really? That is a lot lower estimate that every other realistic estimate I have seen. Most hover around the 40% -> 60% increase in performance over the R9 290X. And as far as the 980 goes, if you dare go to a resolution over 2560x1600 it crumbles. It's simply starved of bandwidth. A R9 290X is equal to the 980 at 4K despite the 980 running circles around the R9 290X at 1920x1080. For memory on a GPU, it's not about the GBs, it's about the GB/s