Video Cards & GPUs News - Page 313
Hardware.info has leaked images of the forthcoming ASUS, XFX, and GIGABYTE versions of the R9 380X video card in preparing its reviews. The images, now pulled, show the R9 380X has 4GB GDDR5 memory, as previously rumored. As well, you can expect DVI-I, DVI-D, HDMI and DisplayPort for all display output configurations.
The ASUS card is known as the STRIX R9 380X and is said to come factory overclocked; the XFX card is known simply as the XFX Radeon R9 380X and looks similar to the 380, and the GIGABYTE card -- known as the GIGABYTE R9 380X G1 GAMING -- will employ dual-fan WindForce 2X cooling.
Last we checked in, other specification rumors for the R9 380X included pointed to a 256-bit memory bus, a core clock speed of 1000MHz, and a memory clock between 5500MHz and 6000MHz. We should have these confirmed one way or the other come November 20, if 4Gamer is to be believed.
The latest time we saw the discrete GPU market share numbers, NVIDIA was dominating with 82% of the market. We thought with the release of the Radeon R9 390X and the hyped to hell Radeon R9 Fury X (as well as the Fury, and R9 Nano) that AMD would win back a considerable chunk of the dGPU market in the last three months... well, they didn't.
AMD only took back 0.9% of the dGPU market, leaving NVIDIA with 81.1%, according to the latest data from analyst firm JPR. Back in February of this year, NVIDIA had 76% of the market leaving AMD with 24% of the market. But, we should hopefully begin to see a change in these numbers now that AMD is finally doing something about it, splitting its GPU division off recently into Radeon Technologies Group.
The biggest GPU fight is about to happen, with the shift from 28nm to 14/16nm set to happen next year. Not only that, but HBM2 will be used by both sides, versus just AMD using HBM right now. If that wasn't exciting enough, Windows 10 delivered DirectX 12 to the world, so we should begin to see some truly interesting things begin to happen early next year.
Following the launch of its take on the GTX 950, GIGABYTE is adding five more high-end video cards to its Xtreme Gaming lineup.
First is the Titan X (GV-NTITANXXTREME-12GD-B), which boasts 1165MHz/1266MHz clock speeds, a 7Gbps memory clock, 12GB VRAM, and a 384-bit memory bus.
The GTX 980 Ti comes in two flavours: Windforce Edition (GV-N98TXTREME-6GD) and Waterforce Edition (GV-N98TXTREME W-6GD). The Windforce Edition features the normal air cooling system; Waterforce features an all-in-one closed-loop water cooling system that houses FEP tubes, a 120mm silent fan, and a low-noise pump. GIGABYTE claims full coverage (GPU, VRAM, and MOSFET) and 38.8% better cooling over the reference design -- plenty enough that no additional fans are needed. Both include LN2 BIOS and an extra 6-pin PCI-E power connector for overclockers. As for specs, you get 1216MHz/1317MHz clock speeds, a 7.2Gbps memory clock, 6GB VRAM, and a 384-bit memory bus with both.
NVIDIA doesn't just release video cards for gaming you know? Yeah, we thought so - but check out their latest NVS 810 video card, which has an insane 8 x mini DisplayPort outputs - each capable of taking 4K displays at 30Hz each.
If the 8 x 4K @ 30Hz was a bit too much and you needed some 60Hz display action, you can knock it down to 4 x 4K at 60Hz. Also, we're not just talking about the normal 4K either - which is 3840x2160, we're talking full 4K - 4096x2160. Each of the miniDP ports on the NVIDIA NVS 810 can handle 4K @ 4096x2160. You can install four of the NVS 810 into a single system for 32 display support... yes, 32 x 4K displays!
The board takes just 68W of power, and is a single slot card for systems where you can have them nice and thin - think display-heavy and digital signage systems. As for what makes the NVS 810 tick, we have 1024 CUDA cores thanks to the use of two GM107 GPUs with 512 CUDA cores on each GPU. There's 4GB of RAM, but NVIDIA doesn't specify if its DDR3 or not.
AMD is expected to drop its mid-range Radeon R9 380X next month, based on the Antigua XT GPU, it should rock 4GB of GDDR5 spread out across a 256-bit memory bus.
Gamers have been expecting the release of the R9 380X for a while now, ever since the R9 285 was released last year. The Radeon R9 380X should feature 2048 stream processors, 128 TMUs, and 32 ROPS alongside the 4GB of GDDR5 on a 256-bit memory bus. Earlier rumors had the R9 380X pegged to include a 384-bit bus, but it looks like 256-bit is where we'll sit.
AMD is gracing the R9 380X with its GCN 1.2 architecture, providing it with the same iteration of architecture that the Fiji GPU includes - the chip that powers the HBM-powered Fury range of cards, as well as the R9 Nano. As for power consumption, we should expect the Radeon R9 380X to use between 200-220W, with aftermarket designs of the card to sport better coolers, lower temperatures and possibly higher power consumption - depending on the card, of course.
When we heard whispers of NVIDIA using GDDR5X on its upcoming Pascal GPUs, it made sense. HBM2 is going to be sparse, and with AMD reportedly having 'priority access' to HBM2, the use of GDDR5X is going to come in handy, big time. Not only is NVIDIA said to be set to reveal GDDR5X-based video cards, but AMD will, too.
According to a new presentation, GDDR5X is going to provide twice the data rate per memory access. We're going to see it jump from the current 32B data per memory access, to 64B data per memory access. GDDR5 has hit a ceiling at around 7Gbps, but GDDR5X is going to have a much higher I/O rates of around 10-12Gbps.
It looks as though GDDR5X is going to be used well into 2018, while GDDR5 will most likely be kept for lower-end cards. I think we're going to see $50 - $300 cards with GDDR5, $300 - $500 cards (and possibly higher) with GDDR5X, and $500+ left for the higher-end HBM2 technology. The use of GDDR5X by AMD and NVIDIA is going to save much more money on R&D, because instead of moving over their entire next-gen GPU product lines, they can continue to use GDDR5X, which is not all that different to the current GDDR5 technology.
One of the big deals about DirectX 12 is that you can use two different GPUs, from two different companies - such as an AMD Radeon card mixed with an NVIDIA GeForce card - in tandem, for more performance. AnandTech has gone ahead with quite an elaborate article, looking at Ashes of the Singularity with mixed GPUs running under Windows 10 and DirectX 12.
Ashes of the Singularity is the first game demo that makes use of DX12's multi-GPU Explicit Multi-Adapter feature, which has some surprising results in such an early game. The results using an R9 Fury X + GTX 980 Ti together are surprising, as it comes out on top - even compared to an R9 Fury X and R9 Fury combination at 2560x1440.
The same can be said for 4K, where the R9 Fury X + GTX 980 Ti comes out on top of even the GTX 980 Ti + GTX Titan X combo, as well as the R9 Fury X and R9 Fury combo. But where it gets exciting is with older cards, so for gamers who might own something like an AMD Radeon HD 7970, and have a GTX 680 laying around not being used (or you could pick one up cheap second hand) - the performance addition dropping the second GPU into the system is great. The HD 7970 on its own at 1440pis capable of 30FPS, but when mixed with the GTX 680, it gets bumped up to 46FPS average - not too bad at all, huh?
When AMD released the Radeon R9 Nano, we fell in love with not only its performance (for being such a tiny, tiny card) but its ultra awesome aesthetics. Well, it looks like we can expect ASUS to release a new R9 Nano, but the shroud on the cooler will be white.
According to ComputerBase.de, ASUS has used its own custom PCB (which isn't white) while dipping the shroud in white paint. The fan is still black, but you can't deny that it doesn't look slick. I would've expected something different from ASUS, maybe doing a black and red theme (in line with AMD's reference card) and in line with their Republic of Gamers brand.
But, you can't deny this R9 Nano doesn't look cool. Now I want to build a new black and white themed system with this new R9 Nano from ASUS.
It looks like NVIDIA is testing out a new GPU, with a new 'JM601 graphics processor' spotted on Zauba. The most mysterious part of this is that it's called 'JM601', which is completely out of place, and totally unexpected.
So what do we think the new JM601 GPU is? Well, for one it could be a dual GPU card based on NVIDIA's current Maxwell architecture - think GTX 990 (dual GTX 980s, or heaven forbid, dual GTX 980 Ti/Titan X GPUs). Alternatively, it could be there to throw us off - but the most exciting part is that it could be a huge Pascal GPU and the name used here - 'JM601' - is to throw us off.
The new JM601 GPU is sitting on Zauba where it was sent on October 14, priced at 73,917 INR - which converts to around $1136. This isn't a cheap chip, so we should expect a monster dual-GPU, or this is the first Pascal-based card, possibly with GDDR5X as we reported not too long ago.
It's getting closer to the end of the year, where we're learning more about what the New Year will bring in terms of new GPU technology. AMD has reportedly taped out two of its next-gen GPUs, with "Ellesmere" and "Baffin" both taping out - and both part of the upcoming Radeon 400 series of video cards.
AMD has also reportedly secured itself a major OEM design win, according to a "source with knowledge" close to rumor site WCCFTech. The Ellesmere and Baffin GPUs aren't high-end parts, but they will fill the shoes of the most important part of AMD's graphics portfolio, the performance, and mainstream markets.
These aren't the successors to the Fury range at all, but more the Radeon 300 series that was rebranded earlier this year from the Radeon 200 series. We should expect more news on AMD's new cards in the coming months.