Performance not two-die four.
When designing an integrated circuit, you are attempting to fit as much complexity as possible within your budget of space, power, and so forth. One harsh limitation for GPUs is that, while your workloads could theoretically benefit from more and more processing units, the number of usable chips from a batch shrinks as designs grow, and the reticle limit of a fab’s manufacturing node is basically a brick wall.
What’s one way around it? Split your design across multiple dies!
NVIDIA published a research paper discussing just that. In their diagram, they show two examples. In the first diagram, the GPU is a single, typical die that’s surrounded by four stacks of HBM, like GP100; the second configuration breaks the GPU into five dies, four GPU modules and an I/O controller, with each GPU module attached to a pair of HBM stacks.
NVIDIA ran simulations to determine how this chip would perform, and, in various workloads, they found that it out-performed the largest possible single-chip GPU by about 45.5%. They scaled up the single-chip design until it had the same amount of compute units as the multi-die design, even though this wouldn’t work in the real world because no fab could actual lithograph it. Regardless, that hypothetical, impossible design was only ~10% faster than the actually-possible multi-chip one, showing that the overhead of splitting the design is only around that much, according to their simulation. It was also faster than the multi-card equivalent by 26.8%.
While NVIDIA’s simulations, run on 48 different benchmarks, have accounted for this, I still can’t visualize how this would work in an automated way. I don’t know how the design would automatically account for fetching data that’s associated with other GPU modules, as this would probably be a huge stall. That said, they spent quite a bit of time discussing how much bandwidth is required within the package, and figures of 768 GB/s to 3TB/s were mentioned, so it’s possible that it’s just the same tricks as fetching from global memory. The paper touches on the topic several times, but I didn’t really see anything explicit about what they were doing.
If you’ve been following the site over the last couple of months, you’ll note that this is basically the same as AMD is doing with Threadripper and EPYC. The main difference is that CPU cores are isolated, so sharing data between them is explicit. In fact, when that product was announced, I thought, “Huh, that would be cool for GPUs. I wonder if it’s possible, or if it would just end up being Crossfire / SLI.”
Apparently not? It should be possible?
I should note that I doubt this will be relevant for consumers. The GPU is the most expensive part of a graphics card. While the thought of four GP102-level chips working together sounds great for 4K (which is 4x1080p in resolution) gaming, quadrupling the expensive part sounds like a giant price-tag. That said, the market of GP100 (and the upcoming GV100) would pay five-plus digits for the absolute fastest compute device for deep-learning, scientific research, and so forth.
The only way I could see this working for gamers is if NVIDIA finds the sweet-spot for performance-to-yield (for a given node and time) and they scale their product stack with multiples of that. In that case, it might be cost-advantageous to hit some level of performance, versus trying to do it with a single, giant chip.
This is just my speculation, however. It’ll be interesting to see where this goes, whenever it does.
Subject: Graphics Cards | June 30, 2017 - 05:26 PM | Scott Michaud
Tagged: nvidia, graphics drivers
Aligning with the LawBreakers “Rise Up” open beta, as well as the Spider-Man: Homecoming VR Experience VR experience, intentionally written twice, NVIDIA has released new graphics drivers!
The GeForce Game Ready 384.76 WHQL drivers were published yesterday on GeForce Experience and their website. Apart from game-specific optimizations, the driver also fixes a bunch of issues, many of which seem very important. First, if you are a fan of Firefall, and your system was unable to launch the game, this driver should remedy that. The driver also claims to remove some or all of the stuttering experienced by GTX 1080, GTX 1070, and GTX 1060 GPUs on Prey 2. Texture corruption in No Man’s Sky, for those who still play the game in an SLI configuration, should be fixed as well, which I believe was a long standing issue, although I could be wrong (as I haven’t been following that game). Vulkan support on Doom (2016) has also been improved.
I should note that, when I tried to custom install the driver through GeForce Experience, the install “failed” three times -- as in, the installed wouldn’t even draw the install button. Eventually, it gave me an install button, and it installed just fine. Not sure what’s going on with that, but I thought you all should know.
Subject: Graphics Cards | June 28, 2017 - 11:00 PM | Scott Michaud
Tagged: epic games, ue4, nvidia, geforce, giveaway
If you are an indie game developer, and you could use a little more GPU performance, NVIDIA is hosting a hardware giveaway. Starting at the end of July, and ongoing until Summer 2018, NVIDIA and Epic Games will be giving away GeForce GTX 1080 and GeForce GTX 1080 Ti cards to batches of Unreal Engine 4 projects.
To enter, you need to share screenshots and videos of your game on Twitter, Facebook, and Instagram, tagging both UnrealEngine and NVIDIA. (The specific accounts are listed on the Unreal Engine blog post that announces this initiative.) They will also feature these projects on both the Unreal Engine and the NVIDIA blog, which is just as valuable for indie projects.
So... hey! Several chances at free hardware!
Subject: General Tech | June 28, 2017 - 06:24 PM | Scott Michaud
Tagged: solidworks, ray tracing, radeon, prorender, nvidia, mental ray, Blender, amd
AMD has released a free ray-tracing engine for Blender, as well as Maya, 3D Studio Max, and SolidWorks, called Radeon ProRender. It uses a physically-based workflow, which allows multiple materials to be expressed in a single, lighting-independent shader, making it easy to color objects and have them usable in any sensible environment.
Image Credit: Mike Pan (via Twitter)
I haven’t used it yet, and I definitely haven’t tested how it stacks up against Cycles, but we’re beginning to see some test renders from Blender folks. It looks pretty good, as you can see with the water-filled Cornell box (above). Moreover, it’s rendered on an NVIDIA GPU, which I’m guessing they had because of Cycles, but that also shows that AMD is being inclusive with their software.
Radeon ProRender puts more than a little pressure on Mental Ray, which is owned by NVIDIA and licensed on annual subscriptions. We’ll need to see how quality evolves, but, as you see in the test render above, it looks pretty good so far... and the price can’t be beat.
Subject: Graphics Cards | June 26, 2017 - 11:29 PM | Tim Verry
Tagged: pascal, nvidia, nicehash, mining, gp106-100, gp104-100, cryptocurrency
In addion to the AMD-based mining graphics cards based on the RX 470 Polaris silicon that have appeared online, NVIDIA and its partners are launching cryptocurrency mining cards based on GP106 and GP104 GPUs. Devoid of any GeForce or GTX branding, these cost controlled cards focused on mining lack the usual array of display outputs and have much shorter warranties (rumors point at a 3 month warranty restriction imposed by NVIDIA). So far Asus, Colorful, EVGA, Inno3D, MSI, and Zotac "P106-100" cards based on GP106 (GTX 1060 equivalent) silicon have been spotted online with Manli and Palit reportedly also working on cards. Many of these manufacturers are also also planning "P104-100" cards based on GP104 or the GTX 1070 though much less information is available at the moment. Pricing is still up in the air but pre-orders are starting to pop up overseas so release dates and prices will hopefully become official soon.
These mining oriented cards appear to be equipped with heatsinks similar to their gaming oriented siblings, but have fans rated for 24/7 operation. Further, while the cards can be overclocked they are clocked out of the box at reference clock speeds and allegedly have bolstered power delivery hardware to keep the cards mining smoothly under 24/7 operation. The majority of cards from NVIDIA partners lack any display outputs (the Colorful card has a single DVI out) which helps a bit with ventilation by leaving both slots vented. These cards are intended to be run in headless system or with systems that also have graphics integrated into the CPU (miners not wanting to waste a PCI-E slot!).
|Base Clock||Boost Clock||Memory (Type)||Pricing|
|ASUS MINING-P106-6G||1506 MHz||1708 MHz||6 GB (GDDR5) @ 8 GHz||$226|
|Colorful P106-100 WK1/WK2||1506 MHz||1708 MHz||6GB (GDDR5) @ 8 GHz||?|
|EVGA GTX1060 6G P106||1506 MHz||1708 MHz||6GB (GDDR5) @ 8 GHz||$284?|
|Inno3D P106-100 Compact||1506 Mhz||1708 MHz||6GB (GDDR5) @ 8 GHz||?|
|Inno3D P106-100 Twin||1506 MHz||1708 MHz||6GB (GDDR5) @ 8 GHz||?|
|MSI P106-100 MINER||1506 MHz||1708 MHz||6GB (GDDR5) @ 8 GHz||$224|
|MSI P104-100 MINER||TDB||TBD||6GB (GDDR5X) @ ?||?|
|ZOTAC P106-100||1506 MHz||1708 MHz||6GB (GDDR5) @ 8 GHz||?|
Looking at the Nicehash Profitability Calculator, the GTX 1060 and GTX 1070 are rated at 20.13 MH/s and 28.69 MH/s at DaggerHashimoto (Etherium) mining respectively with many users able to get a good bit higher hash rates with a bit of overclocking (and in the case of AMD undervolting to optimize power efficiency). NVIDIA cards tend to be good for other algorithms as well such as ZCash and Libry and Equihash (at least those were the majority of coins my 750 Ti mined likely due to it not having the memory to attempt ETH mining heh). The calculator estimates these GPUs at 0.00098942 BTC per day and 0.00145567 BTC per day respectivey. If difficulty and exchange rate were to remains constant that amounts to an income of $1197.95 per year for a GP106 and $1791.73 per year for a GP104 GPU and ROI in under 3 months. Of course cryptocurrency to USD exchange rates will not remain constant, there are transactions and mining fees, and mining difficulty will rise as more hardware is added to the network as miners so these estimated numbers will be lower in reality. Also, these numbers are before electricity, maintainence time, and failed hardware costs, but currently mining alt coins is still very much profitable using graphics cards.
AMD and NVIDIA (and their AIB partners) are hoping to get in on this action with cards binned and tuned for mining and at their rumored prices placing them cheaper than their gaming focused RX and GTX variants miners are sure to scoop these cards up in huge batches (some of the above cards are only availabe in large orders). Hopefully this will alleviate the strain on the gaming graphics card market and bring prices back down closer to their original MSRPs for gamers!
- Mining specific cards are real - ASUS and Sapphire GP106 and RX 470 show up
- First look at Pascal-based GPU cryptocurrency mining station @ Videocardz
- ASUS, COLORFUL and MSI showcase their mining graphics cards @ Videocardz
- Riding the Crypto wave @ TechPowerUP Forums (links/info on mining cards collected here)
- Donate to the PC Perspective Mining Pool! A NiceHash How-to
- Let's Talk About Mining - Cryptocurrency Revisited
- Computex 2017: ASRock Launching H110 Pro BTC+ Motherboard With 13 PCI-E Slots
What are your thoughts on all this GPU mining and cryptocurrency / blockchain technology stuff?
Subject: Graphics Cards | June 26, 2017 - 12:21 PM | Ryan Shrout
Tagged: radeon, nvidia, mining, geforce, cryptocurrency, amd
It appears that the prediction of mining-specific graphics cards was spot on and we are beginning to see the release of them from various AMD and NVIDIA board partners. ASUS has launched both a GP106-based solution and an RX 470 offering, labeled as being built exclusively for mining. And Sapphire has tossed it's hat into the ring with RX 470 options as well.
The most interesting release is the ASUS MINING-P106-6G, a card that takes no official NVIDIA or GeForce branding, but is clearly based on the GP106 GPU that powers the GeForce GTX 1060. It has no display outputs, so you won't be able to use this as a primary graphics card down the road. It is very likely that these GPUs have bad display controllers on the chip, allowing NVIDIA to make use of an otherwise unusable product.
The specifications on the ASUS page list this product as having 1280 CUDA cores, a base clock of 1506 MHz, a Boost clock of 1708 MHz, and 6GB of GDDR5 running at 8.0 GHz. Those are identical specs to the reference GeForce GTX 1060 product.
The ASUS MINING-RX470-4G is a similar build but using the somewhat older, but very efficient for mining, Radeon RX 470 GPU.
Interestingly, the ASUS RX 470 mining card has openings for a DisplayPort and HDMI connection, but they are both empty, leaving the single DVI connection as the only display option.
The Mining RX 470 has 4GB of GDDR5, 2048 stream processors, a base clock of 926 MHz and a boost clock of 1206 MHz, again, the same as the reference RX 470 product.
We have also seen Sapphire versions of the RX 470 for mining show up on Overclockers UK with no display outputs and very similar specifications.
In fact, based on the listings at Overclockers UK, Sapphire has four total SKUs, half with 4GB and half with 8GB, binned by clocks and by listing the expected MH/s (megahash per second) performance for Ethereum mining.
These releases show both NVIDIA and AMD (and its partners) desire to continue cashing in on the rising coin mining and cryptocurrency craze. For AMD, this allows them to find an outlet for the RX 470 GPU that might have otherwise sat in inventory with the upgraded RX 500-series out on the market. For NVIDIA, using GPUs that have faulty display controllers for mining-specific purposes allows it to be better utilize production and gain some additional profit with very little effort.
Those of you still looking to buy GPUs at reasonable prices for GAMING...you remember, what these products were built for...are still going to have trouble finding stock on virtual or physical shelves. Though the value of compute power has been dropping over the past week or so (an expected result of increase interesting in the process), I feel we are still on the rising side of this current cryptocurrency trend.
Subject: General Tech | June 23, 2017 - 05:13 PM | Ryan Shrout
Tagged: nvidia, gtx, geforce gtx usb drive, geforce
What started as merely an April Fool's prank by NVIDIA has now turned into one of the cutest little promotions I've ever seen. Originally "launched" as part of the GeForce G-ASSIST technology that purported to offer AI-enabled gaming if you were away from your keyboard, NVIDIA actually built the tiny, adorable, GeForce GTX USB Key.
This drive was made to look like the GeForce GTX 1080 Founders Edition graphics card and was only produced in a quantity of 1080. I happen to find a 64GB option in a Fedex box this morning when I cam into the office.
Performance on this USB 3.0 based drive is pretty solid, peaking at 111 MB/s on reads and 43 MB/s on writes.
If you want of these for yourself, you need to be signed up through GeForce Experience and opting in to the GeForce newsletter. Do that, and you're entered.
We have some more pictures of the USB drive below (including the surprising interior shot!), so click this link to see them.
Subject: General Tech, Graphics Cards | June 13, 2017 - 01:39 PM | Jeremy Hellstrom
Tagged: nvidia, free games, evga, destiny 2
Were you a fan of the original Destiny or simply a fan of free games and happen to be shopping for a new NVIDIA GPU? EVGA have just launched a new giveaway, if you pick up one of their GTX 1080 or 1080 Ti's they will provide you with a code that not only provides you with a free copy of Destiny 2 but also allows you access to the beta.
As usual you need to have an EVGA account so you can register your GPU and so the code can be provided to your account. From there head on over to NVIDIA to redeem the code and patiently await the start of the beta and final release of the game.
June 13th, 2017 - Get Game Ready with EVGA GeForce GTX 10 Series and experience Destiny 2 on PC. For a limited time, buy a select EVGA GeForce GTX 1080 Ti or EVGA GeForce GTX 1080 graphics card and get Destiny 2 at PC Launch and [Early] Access to the PC Beta!
GeForce GTX 10 Series GPUs brings the beautiful world of Destiny 2 to life in stunning 4K. Experience incredibly smooth, tear-free gameplay with NVIDIA G-SYNC™ and share your greatest gameplay moments with NVIDIA ShadowPlay using GeForce Experience.
About Destiny 2:
Humanity's last safe city has fallen to an overwhelming invasion force, led by Ghaul, the imposing commander of the brutal Red Legion. He has stripped the city's Guardians of their power, and forced the survivors to flee. You will venture to mysterious, unexplored worlds of our solar system to discover an arsenal of weapons and devastating new combat abilities. To defeat the Red Legion and confront Ghaul, you must reunite humanity's scattered heroes, stand together, and fight back to reclaim our home.
Learn more and see qualifying EVGA cards at https://www.evga.com/articles/01112/destiny-2-game-ready/
Subject: Graphics Cards | June 13, 2017 - 01:17 PM | Jeremy Hellstrom
Tagged: nvidia, gtx 1080 ti, GTX 1080 Ti GAMING X, msi, Twin Frozr VI, 4k
MSI's latest version of the GeForce GTX 1080 Ti is their GAMING X 4K and has the design features you would expect, Twin Frozr VI, Hi-C CAPs, Super Ferrite Chokes and Japanese Solid Caps. When benchmarking the card [H]ard|OCP saw performance significantly higher than the quoted 1657MHz boost speed, the average was 1935MHz before they overclocked and an impressive 2038MHz for the highest stable in game frequency. They tested both the default and overclocked frequencies against a battery of benchmarks, including the newly released Prey. The card performed admirably at 4k, with many games still performing will with all graphics options at maximum, drop by for a look.
"We review a custom GeForce GTX 1080 Ti based video card with custom cooling and a factory overclock built for overclocking. Can the MSI GeForce GTX 1080 Ti GAMING X truly deliver a consistent enjoyable high-end graphics setting gameplay experience in games at 4K finally? Is a single card viable for current generation gaming at 4K?"
Here are some more Graphics Card articles from around the web:
- Asus ROG Strix GeForce GTX 1080 OC Edition 8GB 11Gbps Video Card Review @ Bjorn3d
- 15-Way NVIDIA/AMD OpenCL GPU Linux Benchmarks Of Ethereum Ethminer @ Phoronix
- XFX RX 460 4GB Heatsink Edition Review @ Bjorn3d
- XFX Rs XXX Edition Rx 570 4GB OC Review @ Bjorn3d
Subject: Systems | June 12, 2017 - 07:00 PM | Sebastian Peak
Tagged: radeon, PC, Optane, nvidia, Intel, geforce, gaming, desktop, dell, Core X-Series, Core i9, Area-51, amd, alienware
Dell has announced upcoming Alienware Area-51 gaming desktops featuring Intel's new Core X-Series processors, with CPU options up to the 10-core Intel Core i9 7900X and GPU configurations up to dual GeForce GTX 1080 Ti or triple Radeon RX 580 graphics.
"The Alienware Area-51 is our flagship gaming desktop, in this next generation, a new Intel architecture based on ‘Skylake-X’ technology has come to the high end desktop arena; Intel introduces the new Intel Core XSeries processors with a new level of Intel Core i9 options.
Gamers looking for the best that Intel has to offer that love gaming and have creative hobbies that employ resource intensive applications should anticipate the new Area-51 with Intel Core X-series processors. Geared to deliver the best gaming experiences in 4K, 8K and in VR environments, this new rig is powered for gamers running applications that prioritize clock with the 10-core option running at speeds of up to 4.5GHz using stock settings.
The Area-51 featuring Intel Core X-Series is ideal for customers who explore the world of megatasking, doing many system demanding tasks at the same time, and are looking for a complete, reliable solution from a trusted brand."
The Area-51 desktops feature (from Dell):
- Iconic triad high quality, uniquely engineered chassis built to deliver exceptional airflow, thermal management, and user ergonomics for daily use and future upgrades
- Supports NVIDIA SLI and AMD Crossfire graphics technology, with dual and triple GPU options
- Introduces Intel Optane Memory technology and M.2 SSD storage options to Area-51
- Built for gaming enthusiast wanting the absolute best gaming performance played with a VR, 4k or 8k display
- Designed with power supplies that provide modular cabling and a 1500W option with 80 Plus Gold efficiency for clean and efficient power
- Alienware Command Center includes AlienFX, AlienAdrenaline, AlienFusion, Thermal and Overclocking Controls
- Intel X299 w/unlocked BIOS for overclocking, CPU Socket R4 (2066 pins)
- Processor Options:
- Intel Core i7 7800X (6-core, 8.25MB Cache, up to 4.0GHz with Intel Turbo Boost Max 3.0 Technology)
- Intel Core i7 7820X (8-core, 11MB Cache, up to 4.5GHz with Intel Turbo Boost Max 3.0 Technology)
- Intel Core i9 7900X (10-core, 13.75MB Cache, up to 4.5GHz with Intel Turbo Boost Max 3.0 Technology)
- Single Video Card Options
- NVIDIA GeForce GTX 1050 Ti, GTX 1060, GTX 1070, GTX 1080, or GTX 1080 Ti
- Liquid Cooled NVIDIA GeForce GTX 1080
- AMD Radeon RX 570 or RX 580
- Multi GPU Options
- Dual NVIDIA GeForce GTX 1070, GTX 1080, or GTX 1080 Ti (NVIDIA SLI Enabled)
- Triple AMD Radeon RX 570 or RX 580 (AMD Crossfire Enabled)
- Memory Support
- 4x 288-Pin DDR4 UDIMM Slots
- 8GB DDR4 at 2667MHz standard, additional memory available up to 64GB of quad-channel 2667MHz or 2933MHz (HyperX)
- Storage Options
- Single drive: 2TB 7200RPM SATA 6Gb/s or 256GB - 1TB M.2 PCIe SSD
- Dual drive: 128GB - 1TB M.2 SATA SSD (Boot) + 2TB 7200RPM SATA 6Gb/s (Storage)
- Intel Optane Accelerated Options
- 16GB Intel Optane memory accelerated 1TB 7200RPM HDD
- 32GB Intel Optane memory accelerated 1TB - 2TB 7200RPM HDD
- Slot-Loading Dual-Layer DVD Burner (DVD±RW) (Standard)
- Slot-Loading Dual Layer Blu-ray Disc Reader (BD-ROM, DVD±RW, CD-RW)
- Internal High-Definition 7.1 Audio (Standard)
- Dual Killer E2500 Intelligent Networking (Gigabit Ethernet NIC)
- Dell 1820 802.11ac 2x2 WiFi/Bluetooth 4.1 or Killer 1535 802.11ac 2x2 WiFi/Bluetooth 4.1
- Front Ports
- 2x SuperSpeed USB 3.1 Gen 1 Type-A
- 3.5 mm headphone and 3.5 mm Mic Port
- Media Card Reader
- Rear Ports
- 2x RJ-45 Killer Networks E2500 Gigabit Ethernet Port
- 2x Hi-Speed USB 2.0
- 6x SuperSpeed USB 3.1 Gen 1 Type-A
- 1x SuperSpeed USB 3.1 Gen 2 Type-A
- 1x SuperSpeed USB 3.1 Gen 2 Type-C w/ 15W PowerShare technology
- 1x SPDIF Digital Output (TOSLINK)
- 1x Line-In (blue port)
- 1x Front L/R / Headphone (green port)
- 1x Center Channel / Subwoofer (orange port)
- 1x L/R Rear Surround (black port)
- 1x L/R Side Surround (white port)
- Operating System:
- Windows 10 Home (64-bit) (Standard)
- Windows 10 Pro (64-bit)
The release date and pricing have not been announced, but Dell states these Intel Core X-series desktops "will be available late summer" with pricing information soon to come.