TweakTown NewsRefine News by Category:
The AMD Radeon Pro Duo is quite the beast, rocking 2 x Fiji XT GPUs, a total of 8GB of HBM1, and some truly insane horsepower. According to leaked benchmarks, the Radeon Pro Duo is 1.3x faster than the R9 295X2 and 1.5x faster than the NVIDIA GeForce GTX 980 Ti.
The results are courtesy of Expreview, which ran a slew of benchmarks on the Radeon Pro Duo, comparing it against the Radeon R9 Fury X at 4K. Check out the benchmarks, below.
The Radeon Pro Duo versus the Radeon R9 Fury X.
We're only weeks away from the official unveiling of AMD's next-gen Polaris 10 GPU, with new reports from a recent event hosted by the company in Taiwan about Polaris.
The event doubled as a way of showcasing their next-gen Polaris architecture, as well as the new dual Fiji video card, the Radeon Pro Duo. The upcoming Polaris 10 has a TDP of 175W, but will "actually consume much less than that", reports WCCFTech.
The Polaris 10 reportedly hits around 4000 points in 3DMark FireStrike Ultra, compared to the R9 Fury X and GeForce GTX 980 Ti which also hit around 4000 points. All AMD will have to do is hit the $300 (or so) price point on the Polaris 10, and it would be a winner if it consumes less than 175W, and cranks out GTX 980 Ti like performance.
AMD is days away from pushing its Radeon Pro Duo into the wild, its first dual-GPU video card since the infamous Radeon R9 295X2. The press slides are now out - even before I received them (thanks, AMD...) and now we know the official specs - even though we kinda knew that from the constant flow of leaks on the Radeon Pro Duo.
The Radeon Pro Duo features 2 x Fiji XT cores built on the 28nm process, the same GPU powering the Radeon R9 Fury X. We have 4096 stream processors, so with two GPUs we have a total of 8192 stream processors on the Radeon Pro Duo. This is joined by 256 TMUs per GPU (512 TMUs in total) and 64 ROPs per GPU (128 ROPs total). This provides the dual-GPU video card with an insane 16.4 TFLOPs of performance, which is a huge jump on the 8.6 TFLOPs of compute performance from the R9 Fury X.
The GPUs are clocked at 1GHz, while the 4GB of HBM1 per GPU (8GB HBM1 total) is clocked at 500MHz, and with its 4096-bit memory bus, we have 1024GB/sec of memory bandwidth. There's a 350W TDP on the card, which is not too bad at all considering the R9 Fury X has a 275W TDP, while the price is a huge $1499. This isn't a card for gamers wanting to hit 1080p 60FPS, or even 4K gamers - this is a card for the serious, insane enthusiasts who want 4K 60FPS constant, or multi-monitor/VR gamers.
Project Quantum is one of the coolest things AMD has ever shown off, but it was never delivered to market. Project Quantum and an insane amount of power inside of its small, and super-unique chassis - but used an Intel CPU at the time, alongside the dual Fiji GPU when it was debuted at E3 2015.
But with AMD's next-gen Zen CPU architecture nearly here, the company can move away from using its competitors' faster CPUs, with one of its own. According to the latest rumors, AMD postponed the release of Project Quantum for two reasons.
First, an "internal decision to market the Radeon Duo Pro, not as a purely gaming product, but as a professional high performance solution". Second; "The lack of a high performance AMD CPU and Motherboard solution (the original Project Quantum uses an AsRock Motherboard) that could accompany the powerful dual GPU".
The last time we reported on the GPU market share, NVIDIA dominated with 82% of the dGPU market in November 2015.
During AMD's Q4 2015 earnings call, company CEO Lisa Su said: "We have clear opportunities to regain GPU share in 2016 based on the performance per watt of our new GPUs and software leadership". AMD has teased just a few hours ago that their Polaris 10 will be for the gaming PC and high-end notebook market, while the Polaris 11 will find its way into standard notebooks.
Now, for AMD to regain GPU market share, it would have to hit NVIDIA where it hurts: the mainstream video card market. This is the GeForce GTX 950/GTX 960 and GTX 970 markets - where AMD didn't do too much with the Radeon 300 series last year. The Radeon 300 series were mostly rebrands, but the new Polaris 10-based Radeon 400 series should be something completely different.
We now have a better idea of what AMD's next-gen Polaris architecture will be when it arrives on video cards next month, thanks to AMD's recent press release.
There's on specific paragraph that's worth discussing, where AMD said: "AMD demonstrated its "Polaris" 10 and 11 next-generation GPUs, with Polaris 11 targeting the notebook market and "Polaris" 10 aimed at the mainstream desktop and high-end gaming notebook segment. "Polaris" architecture-based GPUs are expected to deliver a 2x performance per watt improvement over current generation products and are designed for intensive workloads including 4K video playback and virtual reality (VR)".
But as VideoCardz notes, we need to focus on the word 'mainstream'. AMD has stamped the word 'mainstream' onto everything between casual and enthusiast segments, where back in 2014, mainstream was associated with the Radeon R7 260.
Every single day seems to bring something new to the table when it comes to next-gen GPUs, with today being no exception - in the last 14 hours or so, we've gotten our first look at the GP104.
NVIDIA's GP104 GPU is the Pascal-based consumer GeForce GPU, which will power the upcoming GeForce GTX 1080 and GTX 1070 video cards. The GP104 is made on the 16nm FinFET process, but it looks like the GP104 will power two very different video cards.
The GeForce GTX 1080 is expected to arrive with 8GB of GDDR5X, while its sibling in the GTX 1070 will reportedly sport 8GB of GDDR5 - both with 256-bit memory buses. The GTX 1080 with its 8GB of GDDR5X will have it clocked at 2GHz (8GHz effective), an increase from the 7GHz effective speeds on the current GDDR5-based offerings.
AMD is days away from the official launch of its Radeon Pro Duo, the company's dual-GPU based on the Fiji architecture and HBM1 technology. HWBattle has received their sample, and have uploaded pictures for our viewing pleasure.
The Radeon Pro Duo features two Fiji XT GPUs with 4096 stream processors each, for a total of 8192 stream processors. Each GPU features 64 ROPs and 256 TMUs, with each GPU clocked at 1GHz. Each GPU also has 4GB of HBM1, for 8GB HBM1 total. We have 16 TFLOPs of 32-bit single precision compute performance, which makes the Radeon Pro Duo the fastest video card on the market.
The 4GB of HBM1 per GPU is clocked at 500MHz, resulting in 512GB/sec over the hugely wide 4096-bit memory bus. This means that applications designs for specific workloads on the Radeon Pro Duo can hit over 1TB/sec bandwidth - which is simply put, insane. The TDP of the Radeon Pro Duo is 350W, with 3 x 8-pin PCIe power connectors required. There's 3 x DP outputs and 1 x HDMI, too. Watercooling is standard, as the Fiji XT GPU runs quite hot, but AMD's engineering department took care of that with the Radeon Pro Duo.
It looks like AMD's next-gen Polaris flagship GPU has been spotted again, with the mysterious 'C99' receiving its RRA certification. This certification puts it on a path towards being consumer ready, which is exciting. These numbers might not look like much, but the nomenclature is used for test boards. All of the boards start with the letter C, with the number proceeding the letter C being the generation of video card. For example:
- Hawaii XT had the C67101 code name
- Tonga had the C76501 code name
- Fiji XT had the C88001 code name
This means that Polaris-based boards, alongside other GCN 4.0-based, should have the C9XXXX nomenclature. The Baffin XT-based GPU that was on Zauba recently had the C98101 codename, which tells us it's a Polaris-based product. But now the C99305, C99304, C99303, and C99302 have been spotted getting their RRA certification, back on April 8.
The RRA certification is South Korea's National Radio Research Agency, with any silicon-based electronic needing this certification before it can be a consumer product. The certification doubles as a final stage of design of the GPU, as no more changes can be made after the RRA certification has been approved.
What makes this new entry interesting, is that there's also the C99398 board - which could be a higher-end Polaris 10 offering or even our first tease of Vega 10. Vega will be powered by HBM2, which is something that won't grace Polaris. We don't know much more than this, but as soon as more news is here, you can be sure we'll be all over it.
We keep reporting on the GeForce GTX 1080 and GTX 1070, based on the GP104 GPU, but what about the successor to the insanely popular GTX 950 and GTX 960? The GP106 will be released sometime later this year, and it might be a surprisingly powerful, yet power efficient video card.
According to the new reports, GP106 cards won't require any additional PCIe power connectors, so it'll be a slot-and-forget deal. We should expect the GeForce GTX 1060 to use less than 75W of power, which will be insane given that it should deliver a nice 30-50% bump in performance over the current GM206-based GTX 960. The 16nm FinFET process at its finest.
Right now, the GTX 950 and GTX 960 are still doing incredibly well in the market, so there's no need for NVIDIA to pump out a GM206-based successor just yet. But, the GTX 970 and GTX 980 cards have been out for nearly two years now, so bring on the GTX 1070/GTX 1080!