TweakTown NewsRefine News by Category:
We already know that the Radeon R9 390X is going to be a special card, where according to rumors we can expect it to arrive with a 4096-bit memory bus, made possible with High Bandwidth Memory, or HBM. This card should have memory bandwidth of a truly next-gen 640GB/sec. But, what comes after that, is going to be even better.
The first generation of HBM, which SK Hynix is now shipping, provides 2Gb per DRAM die, 1Gbps speed per pin, 128GB/sec of bandwidth and is stackable in groups of four. This results in 4 x 128GB/sec = 512GB/sec, with another 128GB/sec coming from somewhere else on the card. But, the second-generation HBM is going to be lightning quick, even compared to the already-damn-impressive first-gen HBM tech.
The second-gen HBM technology will allow for 8Gb per DRAM die, up from 2Gb on the first-gen HBM for starts. This means we should see GPUs with 8-16GB of VRAM on-board. Second, the bandwidth increases to 256GB/sec (up from 128GB/sec) which should arrive as around 1.28TB/sec of available memory bandwidth. Considering NVIDIA's second-generation Maxwell-based GeForce GTX 980s only have 224GB/sec of memory bandwidth on a 256-bit bus, the 4096-bit wide memory bus with 1.28TB/sec of bandwidth will surely do some insane things, especially at Ultra HD and beyond resolutions.
This is where it will get interesting; over 1TB/sec of memory bandwidth on a 4096-bit bus is going to be an amazing sight to behold, especially at 4K or 8K. Even more so on something like VR with the Oculus Rift, where it could render one eye separate to the other, at 1440p each, and still have memory bandwidth to spare. Things are going to get exciting with the Radeon R9 390X and beyond, but I think that's just the beginning. The R9 490X is going to be when things really kick into second gear, not only spitting on 'next-gen' consoles, but it's going to be when game developers hopefully wake up and realize they should be coding for these GPUs with insane amounts of memory bandwidth, as they're truly next, next-gen technologies.
The last rumor we had to report on about AMD was the exciting new HBM-based, 4096-bit wide memory bus Radeon R9 390X, but now we're going to tell you a little about the lower-end cards that will most likely find their way to consumers in the first half of 2015.
These new cards will reportedly be AMD's new MXM-based GPUs, arriving as the Litho XT and Strato PRO. These two new mystery cards have just reached Zauba.com, with the Litho and Strato most likely standing for the space-driven names of Lithosphere and Stratosphere. Starting with the Litho XT, which will be a Type A MXM GPU with 2GB of GDDR5 memory, while the bigger Strato PRO will be a Type B MXM GPU with 4GB of RAM.
The XT side of things will be the powerful card, while the PRO will have a cut die. We could also see a Strato XT in the future, but we won't know anything until the new rumors start floating online.
NVIDIA has launched its new Tesla K80 compute card at the Supercomputing 2014 (SC14) conference in New Orleans, which is a dual GPU design, cramming in an insane amount of compute power into a single card.
One of our friends, Anshel Sag, over at Bright Side of News reported the news, explaining that "Logically, you would think that the K80 would naturally be two K40's smacked together into a single card, but that's not accurate. In order to build the K80, NVIDIA actually went with GPUs with similar shader core counts as the Tesla K20, but what's most important is that they actually did double the onboard memory of the K80 from the K40 to 24 GB of GDDR5". 24GB. OF. RAM.
NVIDIA is using 12GB of GDDR5 per GPU, for a total of 24GB of RAM. We have 4992 shader cores, with the company using two GK210 GPUs, instead of the GK110B's found on the Tesla K40. NVIDIA is claiming that the new Tesla K80 is capable of 8.74 teraflops single-precisiion, and 2.91 teraflops double-precision. These numbers are over double that of the K40.
Holy caps lock, Batman! MSI has just released its GTX 970 GAMING 4G GAMING EDITION video card, which includes quite a good card with all those capitalized letters.
The GTX 970 GAMING 4G GAMING EDITION features MSI's new Frozr V Thermal Design, which is a dual-slot cooler with dual 10cm Torx fans that pump through considerable airflow, keeping the factory-overclocked GPU nice and cool. MSI explains that the "heat sink itself is re-organized through the SuperSU design which increases heat-pipe efficiency, especially when combined with Airflow Control Technology".
When it comes to Base and Boost Clocks, we have 1165MHz and 1317MHz, respectively. The RAM is set at 7010MHz, and we have a change in display outputs, with two dual-link DVI outputs, a DisplayPort output, and a single HDMI out.
Today is the day for next-gen GPU news, where we just reported about AMD's Radeon R9 390X to feature an insane 4096-bit memory bus. Now, we're hearing about NVIDIA's new GeForce GTX Titan II, a beast of a GPU that would feature 12GB of VRAM.
This new beast would be baked on the GM200 silicon, on the 28nm process, feature 3072 CUDA cores, a 384-bit memory bus, and the aforementioned 12GB of VRAM. When it comes to clock speeds, we should see a Core clock of 1100MHz, while the Boost clock sits at 1390MHz and the memory at 6GHz. With the GTX 980 having 2048 CUDA cores, the GTX Titan II based on the current rumors has over 1000 additional CUDA cores.
We should see a David vs Goliath battle early next year for GPU supremacy between AMD and NVIDIA, it's just too bad that today's games run like absolute crap, even on great hardware.
Here we are again, another post about the AMD Radeon R9 390X, but this time we have some even more exciting news. There's more details leaking out on the codename Fiji XT board, thanks to a listing in the SiSoft Sandra benchmark database.
If the details are true, and right now we're classing them as rumors, we could expect the Radeon R9 390X to feature a whopping 4096 stream processors, with 64 compute units and 256 texture units - but the best bit, we've saved until last: a massive, game-changing 4096-bit memory bus. This insane jump in memory bus specs will have been a result of the HBM technology we reported on a little while ago now, with a 1024-bit input/output interface.
We should see the first R9 390X cards with 4GB of RAM, but 8GB of VRAM will be seen on these GPUs very quickly. If we do see four, first-generation HBM DRAM chips with a 4096-bit memory bus operating at around 1.25GHz, we can expect a huge memory bandwidth of 640GB/sec. Considering the still-fresh, and kick-ass NVIDIA GeForce GTX 980 has a 256-bit memory bus with 224GB/sec of bandwidth, AMD would be killing it at high-res. 4K and beyond would see AMD leaping ahead, which is something they need right now.
With 4096 stream processors and an insanely wide memory bus, it would be the highest bandwidth ever used on a single GPU. If these specs are correct, it means that AMD should have the fastest video card on the market, by a nice amount - even more so when it comes to high-res, and multi-monitor gaming.
NVIDIA already has two great GPUs in its GeForce GTX 970 and GTX 980 cards, but the mid-range market needs some lovin', too. This is where the GeForce GTX 960 comes into play, and according to the latest rumors, it won't be as cut down as previous generations have been.
The GeForce GTX 960 will reportedly maintain its GM204 core, as well as the 256-bit memory bus and 4GB of GDDR5 memory that both of its bigger brothers have. The GTX 960 will reportedly be clocked at 933MHz on the Core, and most likely 1408 Stream Processors, with 88 Texture Units behind it.
When it comes to pricing, we should expect some cut throat pricing from NVIDIA, with a price of under $249 in the US. When will it be unveiled? We should see it coming into the world in Q1 2015, so we don't have much longer to wait.
We can't let NVIDIA have all the next-gen GPU fun, can we? AMD has reportedly kicked off internal testing of its next-generation GPU, known as Fiji XT.
The news is coming from the Zauba.com database, which "monitors imports and exports to and from India, where high-tech companies employ many engineers" according to KitGuru. AMD's Indian office received a "printed circuit board assembly (video video card) C880 Fiji XT P/N.102-C88001-00" which could be the company's new GPU.
VideoCardz.com is also reporting that the Fiji XT architecture will arrive to us as the Radeon R9 390X, and could use the 28nm process (which is what the R9 290X is built on) or the smaller, more power-efficient 20nm process. We should expect a new GPU from AMD in the next six months or so. The last time we heard about the Radeon R9 390X, we heard it would arrive with HBM memory and baked onto the 20nm process.
It looks like the refreshed 16nm-based GPUs from NVIDIA could be delayed, due to TSMC not being able to hit its scheduled 16nm FinFET mass production schedule - which was due for Q1 2015. The reason? TSMC is busy pumping out A9 chips for Apple and its new iPhones, which is seeing volume 16nm production shifted a few quarters deeper into 2015.
For NVIDIA, this means that the company won't be able to bring its "Big Daddy" GM200 chip to fruition, at least on 16nm, until late 2015 or early 2016. The process shift is a lengthy one, something we saw with "Big Kepler" a couple of years ago. 28nm went into mass production over at TSMC in October of 2011, but the GTX Titan was released in February 2013. Two years of waiting for NVIDIA, with TSMC finally getting its act together for 28nm volume production, in order to built a chip that was reasonably priced, which resulted in the GeForce GTX 780.
TSMC has said that its 16nm process will result in around 7-9% of its total revenue in Q4 2015, which means that if any 16nm-based GPUs make it to consumers late next year, the amount of them will be quite small.
During NVIDIA's earnings call a few days ago, the company teased its upcoming next-gen GPU, even though it only released its GeForce GTX 900 series a couple of months ago - cards that are still kick ass. These cards are based on the GM204 GPU, but the new one coming is the GM200 core, which should provide a big boost in performance.
Previously, we thought that NVIDIA would release this GPU on 16nm, but the 16nm process is being chewed up at TSMC. NVIDIA's Maxwell architecture is so efficient, it could fit on the 28nm process, and be re-released next year, or early 2016 when the 16nm FinFET process is ready for it. This would allow for even more performance, allowing NVIDIA to strike at AMD twice, and probably still win all sides of the battle.
Just how fast should we expect these new GPUs to be? We could expect around 50% more performance than the GeForce GTX 980, which is already a damn impressive card, but 50% more performance? Yes, thanks, NVIDIA.