maxresdefault

Motivated by the motto “Gaming Perfected”, NVIDIA has recently launched not only its Pascal architecture but also appended a new family of consumer based gaming cards to its popular Geforce lineup, the GTX 10 series.

The successor the Geforce 900 series the Geforce GTX 10 series has just cast out and NVIDIA is making a huge deal out it but is it really that big or just a promotional stunt. Let’s check out what they have to offer.

nvidia_gtx_1080_jpg_w_590_h_295_l_50_t_40

The Geforce GTX 10 series right now comprises of only two cards the Geforce GTX 1080 and GTX 1070 and that too are founders edition available right now that means all we have found right now is reference cards made by NVIDIA itself which mean poor cooling and low Overclocking. The Geforce GTX 1080 is marked around 699$ and GTX 1070 for I guess a 100$ less bot of the cards sport a 16nm finFET Pascal GP 104 GPU(Graphics Processing Unit). The specifications for respective cards are as follows.

                                                                       GTX 1080                                                            GTX 10710

                              CUDA Core                                     2560                                                                     1920

                              Core Clock                                   1607MHz                                                                1506MHz

                              Boost Clock                                  1733MHz                                                                1683MHz

                              VRAM                                         8GB DDR5X                                                            8GB DDR5

                              Tera FLOPS                                        9                                                                            6.5

                              Memory Speed                               10Gb/s                                                                    8Gb/s

                              Memory Bus Width                       256-bit                                                                    256-bit

                              Memory Band Width                    320GB/s                                                                 256GB/s

                              Transistor Density                       7.2 Billion                                                                 7.2 Billion

                              Manufacturing Process                   TSMC 16nm finFET                                       TSMC 16nm finFET

                                                 TDP                                 150W                                                                    180W

Supported technologies for both the cards are as follows :

                                   Multi-Projection(new)                                          YES

                                   VR Ready                                                                 YES

                                   NVIDIA Ansel (new)                                             YES

                                   NVIDIA Gsync Ready                                           YES

                                   NVIDIA Game Stream Ready                              YES

                                   NVIDIA GPU Boost                                               3.0

                                   Microsoft Directx                                                   12 with feature level 12_1

                                   Vulkan API                                                             YES

                                   Open GL                                                                  4.5

                                   Bus Support                                                            PCIe 3.0

Let me break that down for you.  The GTX 1080 and 1070 being newer generation cards doesn’t actually include all the improvements that we normally see when generations advance.  As have seen in previous generation cards or what we ascertain from its competitor AMD, and why this is so let me tell you.

Comparing the above-mentioned specifications with their respective predecessors. GTX Titan X had 8 Billion 28nm transistors, clearly as the size of transistors has decreased thanks to new finFET technology. NVIDIA doesn’t seem to be interested in increasing their tally. In spite of less transistor count, GTX 1080 outperforms the GTX Titan X.

Is it all that? NO.  GTX 1080 not only has small transistors, but also sports a new memory controller, which facilitates to gain more bandwidth on same bus size that’s why it sports GDDR5x (DEVELOPED BY MICRON SAME AS GDDR5 BUT HAS to DOUBLE THE BANDWIDTH DUE TO DOUBLE DATA FETCHING IN SINGLE CYCLE) but this is only limited to 1080.  Both the cards now support High-Bandwidth SLI due to which the performance increases a great deal when running in SLI mode, but that doesn’t mean SLI is 100% efficient it has just improved the overall efficiency of previous generation SLI.

2016_NVIDIA_Pascal_FINAL_published-page-026-635x357

Concurrent multi-projection: was talked a lot about at NVIDIA’s launch event.  What it causes that it renders different displays differently.  That means in the case of VR you’re each eye takes an explicitly rendered image depending upon view angle and head position for greater depth of field.

Talking about NVIDIA Ansel it’s nothing, just a camera technique which allows you to capture 360-degree footage in the game and allows you to see it on your VR.

The major gap that NVIDIA faced due to thlaunchingon of new Microsoft Directx12 was A-syncomputes. .
Actually earlier till DX 11 the major performance of your game was dependent upon how powerful your single core cluster of GPU is there the codes were executed sequentially so total number of cores didn’t affect the overall performance a lot but scenario changed with AMD’s API Mantle followed by Microsoft Directx 12 where the codes can be executed A-synchronously utilizing most of the hardware delivering massive boost in performance but NVIDIA hardware was limited to sequential execution which did not help a lot and although API supports parallel processing but the basic underlying hardware did not unlike AMD which in its GCN architecture included Async shaders and implemented Async compete on hardware. With their new lineup NVIDIA sort of addressed this issue but not completely the included what’s called Dynamic Load Balancing and fast Pre-emption.

Dynamic Load Balancing :

Your GPU apart from just filling the colors in textures and making textures does a batch of things such as physics processing & post processing these tasks are computing tasks so earlier in Maxwell architecture software developer used to decide how much resources of GPU should he allocate for computing and how much for graphics this method was called static partitioning the main drawback of this was maybe your computing requires fewer resources than graphics and so it won’t utilize all of the given resources due to which some part of GPU will sit idly by dynamic load balancing the partitioning task is handled by the GPU itself so the programmers won’t have to bother himself wit the perfect ratio and idle parts of your GPU can be balanced in order to get most out of your GPU.

Fast Pre-emption :

Load Balancing is an important factor, but when it comes to time-critical operations, it is necessary to cover them first. Think of it is this way, you are standing in a security check queue in front of a theatre. Your movie is just about to start and you are at position number 5, the security guard finds out everyone one by one and then allows to pass. If it takes 2 min to check every person it would then require you 10 mins to fall through. But if the security guard is nice and you tell them it’s highly important for you to go through now, because your movie is just about to start whereas others can wait for a while, and he permits you to pass through so that you reach on time. We say that he preempted the normal sequence in order for a high priority task to complete IE for you to communicate. Similarly, at that place are various tasks that GPU faces that have higher priority than others, so it performs preemption, coming onto the point, preemption was only allowed on the draw call (THE CALL THAT GPU ANSWERS FOR DRAWING OBJECTS ON SCREEN CONTAINING HUNDREDS OF POLYGONS OR THOUSANDS OF PIXELS) level until Maxwell. Now due to change in architecture Maxwell can implement preemption at pixel level , but what actual difference it makes well consider this a draw call was made which for simplicity let’s consider consists of 1 polygon with 100 pixels now as soon as GPU started execution it encountered a Async time-warp call in VR environment which if not processed withing given time will cause a major issue such as frame drop, that is the worst thing to happen in VR environment, if there would be no preemption allowed at pixel level it cannot process the call and will for sure lead to frame drop but if it can pause pixel processing instead process Async time wrap call first and then again start where it left hurry GPU save the day.

Wellnowadaysow you might say that OK so many technologi, , faster than previous generation cards and VR features with top of the line performance all this for 699$ there’s no doubt that this is something big well technically NO.

2016-01-12-image

As of now the Pascal series consists of 2 GPU’s namely GP104 and GP 100 . GP 100 is the real breakthrough of Pascal architecture with 15.2 billion transistor density , HBM 2 memory , NVIDIA NVLink memory coupling etc.  The sadness that occurs only with Tesla P100 server grade cards which cost more than a hundred thousand dollars for a set of I guess 4 – 7 GPUs. As far as gaming is concerned GP104 is only what we have got till now.  According to our sources, we might see GP102 in GTX 1080 TI or Pascal based Titan cards which might release soon in the upcoming months but it’s not supported yet.

The core cluster of GP104 was analyzed with regard to its predecessor GM200-310 what we found was not major architectural changes which were expected.

Conclusion: The new GTX family is good GTX 1080 is best NVIDIA can offer now at this price point we are also happy to see the VR features offered, but cannot state that it will be best in the market as its competitor AMD has Today unveiled it’s Polaris architecture based cards.  With the price of Mid-range segment, but performance equivalent to high goal. Unveiled at Computex by Raja Koduri Sr.  Vice President and Chief Architect of Radeon Technologies and soon will reach the market.  AMD will also strike with Vega series that is destined for high-end market.  Now GTX 10 series is best what you can get by NVIDIA in and is best in the market but will it stay that way?. We will it find out real soon.

 Do not miss AMD event at Computex Taiwan for some real groundbreaking action LINK (skip until  16:00 if you don’t want to gaze at AMD logo with some old school music in background)

LEAVE A REPLY

Please enter your comment!
Please enter your name here