TweakTown NewsRefine News by Category:
Last August we reported on the 'Faraway Islands' GPU which was meant to be the Radeon R9 400 series but it looks like this is known as 'Arctic Islands'. You might have guessed from the word 'arctic' that AMD is looking to keep the R9 400 series cooler, with improved power efficiency, better aligning them with NVIDIA's Maxwell-based GeForce GTX 900 series GPUs.
Then in November we reported that the Radeon R9 490X (and any GPU close to it in the R9 400 series) could feature over 1TB/sec memory bandwidth, thanks to HBM. The latest rumors are pointing to Arctic Islands being baked onto the 20nm process, which should see the GPUs running much cooler than the R9 290X which runs quite hot, especially compared to a GTX 980.
When can we expect AMD to begin shipping the Radeon R9 400 series? Well, we are to expect the R9 300 series in Q2 2015 of this year (sometime between April and June) so we shouldn't see the R9 400 series until at least the second half of 2016.
SAPPHIRE was the first in the world with a Radeon R9 290X with 8GB of VRAM, with its Vapor-X model but now it's back with another model, the Tri-X card. The new card features SAPPHIRE's great triple fan cooler, increased clocks, and dual BIOS.
We have 2816 stream processing units on the SAPPHIRE Radeon R9 290X Tri-X 8GB, with an engine clock of 1020MHz. 8GB of GDDR5 is spread across a 512-bit memory bus, clocked at 1375MHz (5.5GHz effective). SAPPHIRE has used its original PCB, with a 6-phase power design pushing upwards of 240W, with two 8-pin PCIe power connectors which provide up to 375W of total power.
SAPPHIRE is pushing this card out to its usual e-tailers and retailers, but we've reached out to the company and requested a sample already. So expected a review from us in the very near future.
We reported yesterday that NVIDIA was having an issue with its GeForce GTX 970 GPU, with its VRAM only maxing out at 3.5GB even with 4GB on offer. Well, this seems to be getting a detailed explanation quickly from NVIDIA, with the Senior Vice President of Engineering Jonah Alben sending PC Perspective a diagram of the reason behind the GTX 970 acting weirdly.
As you can see in the diagram above, there are three four things greyed out. Three SMM blocks and a single L2 cache block. NVIDIA explained to PCPer: "Despite initial reviews and information from NVIDIA, the GTX 970 actually has fewer ROPs and less L2 cache than the GTX 980. NVIDIA says this was an error in the reviewer's guide and a misunderstanding between the engineering team and the technical PR team on how the architecture itself functioned. That means the GTX 970 has 56 ROPs and 1792 KB of L2 cache compared to 64 ROPs and 2048 KB of L2 cache for the GTX 980".
So the question is, is the GeForce GTX 970 a 4GB card or not? It has 4GB of VRAM, that much we know, but accessing the remaining 0.5GB of RAM is slower than the first 3.5GB.
NVIDIA only took the wraps off its its GM206-powered GeForce GTX 960 video cards last week, but now we're already being teased with the hint of 4GB models to be coming soon, as in just a few weeks away in March.
Computerbase is behind the rumor, posting a shot of Inno3D's latest offerings under their iChill range. As you can see, it says "4GB available in March - watch this space!", so unless Inno3D had this go out accidentally, or its a new form of guerrilla marketing, we know to expect GeForce GTX 960 cards to start shipping with 4GB of VRAM in March.
But even when, or if it happens, it will put the GTX 960 in a weird position. Unless AIBs can price them at under $250 (the GTX 960 with 2GB of VRAM sells for $199-$209) then it will be too close to the price of the GTX 970, which also has 4GB of RAM. I guess we'll see, with GTC 2015 kicking off in March, it could be the month for NVIDIA announcements, along with the rumored GeForce GTX Titan X.
During AMD's Q4 earnings call, CEO Lisa Su teased that the company will have "very good" graphics products that will ship in Q2 2015. Su confirmed that AMD has multiple product launches to spread out over "the next couple of quarters", with those quarters being Q2 and Q3. Most of which will be released between April and June.
Su did explain what is happening with the GPU side of the business, in that AMD had a large quantity of its Hawaii-based R9 290 series GPUs, most of which are rapidly declining. The company then increased its inventory in the anticipation of continued demand, but it fell off very quickly - mostly because the cryptocurrency market began to shrink rapidly. Thanks, Bitcoin miners. During Q4, Su said that company took the required actions to get the inventory levels back to normal, and moving into Q1 2015, AMD took even more drastic actions. I would dare say that NVIDIA's Maxwell-based GeForce GTX 980 and GTX 970 video cards had something to do with that.
Su continued: "Think on the computing and graphics business, we can improve our execution and there is a lot of focus, at least from my standpoint, to ensure that our 2015 product launches are quite strong. So that's important for us to really stabilize that business, and that will certainly be key to our second half performance. I think from where I see it going forward, we are very focused on correcting the channel and normalizing that business. I think we have some good products that are coming out for it. So I do see opportunities for a return to a more healthy channel business going forward".
We've been reporting on pretty much every GPU rumor we can, as they're all exciting, but the next-gen TITAN X is shaping up into something more real with each passing day. Now we have some unconfirmed news that the GeForce GTX TITAN X will have an MSRP of $1350.
The stock card will be arriving with 6GB of VRAM, with 12GB of VRAM to arrive at a later date, or through various AIBs if they choose to splash the GPU with a titanic amount of VRAM. The TITAN X will be powered by the GM200 chip, which is the full chip that we've all been waiting for. The GM204 powers the GeForce GTX 980 and GTX 970 cards while the GM206 reportedly powers the GTX 960 which is said to be imminent.
NVIDIA could slightly cut down the die to start with, but the fully unleashed GM200 core should arrive on a 384-bit memory bus, feature 12GB of VRAM and a huge 3072 CUDA cores. A cut down version of this could feature between 2560 and 2816 CUDA cores, but it would still perform like a champion. The price however, is what has people talking, and with the current GTX TITAN Black Edition still commanding a price of around $1000 on Amazon, the $1350 price isn't too bad. "Isn't too bad" is something worth talking about, because for around $350 more we should see a nice 30-50% performance increase given the specs. The 12GB of RAM is going to come in handy for future titles and super massive multi-monitor setups, which is something I will be getting into in March with these cards if they do materialize.
As usual, we can't confirm this as it's a rumor, but it looks like the upcoming next-gen GPU release from AMD could arrive as the Radeon R9 380X, and not the R9 390X like most would have expected. Sweclockers are behind the rumor this time, giving it a little more credit.
The Radeon R9 380X would feature 4096 cores, 4GB of the new, next-gen HBM (high bandwidth memory) and a Q2 2015 release window. The new card is known as the Fiji XT and will directly replace most of AMD's current high-end R9 200 series lineup. We should expect this card to be around 40-50% faster than the R9 290X, mostly thanks to architectural changes and the new, super-fast HBM RAM.
NVIDIA will be fighting back with another rumored product, the Titan II or Titan X. They could also battle it out with a refreshed GM200-based GeForce card in the form of the GeForce GTX 980 Ti, or if I had my way, the GeForce GTX 980 Ultra.
We know it's coming, but NVIDIA hasn't confirmed anything just yet - but here it is, the GM200. If you didn't know, this is the fully unleashed Maxwell GPU, which should materialize into the GeForce GTX Titan II.
The GM200 engineering board (180-1G600-1102-A04) will use a PG600 board according to VideoCardz, and features 24 of Hynix's H5GQ4H24MFR modules, clocked at 7GHz. We have a total of 12GB of VRAM, which indicates this will be a new Titan part, and not a normal GeForce GPU. We don't know if it will be the Titan Z II, Titan Ultra, Titan II, or something else entirely, but whatever it is... it's going to be awesome.
We can see the prototype board is missing the DVI port, but it features three DisplayPort outputs, and HDMI 2.0 just like the GTX 980 and GTX 970 cards. We should expect the full Maxwell card to feature 3072 CUDA cores, too. This card will be a monster, something I expect NVIDIA will unveil at GTC 2015 in March.
We heard that NVIDIA were preparing for the launch of its mid-range GeForce GTX 960 last month, and just recently the rumor of a buy-it-now price of $199 was teased. Now here we are, with actual box shots of various cards from NVIDIA's add-in-board (AIB) partners, such as GIGABYTE and Inno3D.
The first one, pictured above, is of the GIGABYTE G1 Gaming 'Super Overclock' GTX 960, which uses its impressive WindForce cooling setup and 2GB of RAM.
Next up, we have two cards from Inno3D; the iChill GeForce GTX 960 Ultra, and the GeForce GTX 960 OC. The Ultra variant rocks an insane cooling setup, featuring the HerculeZ X3 cooler. We have a clock speed of 1178MHz, while the memory is at 7010MHz effective. This provides memory bandwidth of 112GB/sec, on its 2GB of frame buffer.
The GTX 960 OC version features the smaller HerculeZ X2 cooler, but it'll also feature the same speeds as its Ultra brother. Two-way SLI support is provided on both cards, too.
If there anyone who likes GPU rumors, it's me. The latest reports are suggesting that AMD might make a huge leap with its new Radeon R9 300 series, which should make use of the new High Bandwidth Memory (HBM), but more interestingly, it is based on a 2.5D design, and will use up to 300W of power.
The news is coming from the LinkedIn profile of Linglan Zhang, who is currently employed by AMD as the System Architecture Manager. His profile lists that he is working on a new GPU SOC chip that uses the 2.5D design, and rocks a TDP of 300W. We should expect the new GPUs from AMD to be made on the 28nm architecture, since the 16nm and 20nm die shrinks are now at least six months or more away - from both AMD, and NVIDIA.
Comparing GDDR5 against HBM is something that people need to start really looking at, as the I/O per chip on GDDR5 is just 32-bit, while the 4-Hi HBM 'Stacked DRAM' pumps things up to a huge 1024-bit. Max bandwidth per min on GDDR5 is just 7Gbps, while HBM sits at 1GBps. The max bandwidth of GDDR5 sits at 28Gbps, while the HBM technology can scale between 64Gbps and 256Gbps.
When can we expect the new Radeon R9 390X and R9 380X? Well, the reports state that AMD is already taping out the new R9 380X, which means we could see GPUs in consumers' hands in the coming weeks, so we might see AMD unleash these new cards before NVIDIA kicks off its GPU Technology Conference in March, where we might see something unveiled - last year, we saw the GeForce GTX Titan Z for example. Whatever happens, it's exciting times in the world of GPUs, that's for sure.