Eight-GPU SLI in Unreal Engine 4 (Yes There Is a Catch)

Subject: Graphics Cards | March 29, 2018 - 09:52 PM |
Tagged: nvidia, GTC, gp102, quadro p6000

At GTC 2018, Walt Disney Imagineering unveiled a work-in-progress clip of their upcoming Star Wars: Galaxy’s Edge attraction, which is expected to launch next year at Disneyland and Walt Disney World Resort. The cool part about this ride is that it will be using Unreal Engine 4 with eight, GP102-based Quadro P6000 graphics cards. NVIDIA also reports that Disney has donated the code back to Epic Games to help them with their multi-GPU scaling in general – a win for us consumers… in a more limited fashion.

nvidia-2018-GTC-starwars-8-way-sli.jpg

See? SLI doesn’t need to be limited to two cards if you have a market cap of $100 billion USD.

Another interesting angle to this story is how typical PC components are contributing to these large experiences. Sure, Quadro hardware isn’t exactly cheap, but it can be purchased through typical retail channels and it allows the company to focus their engineering time elsewhere.

Ironically, this also comes about two decades after location-based entertainment started to decline… but, you know, it’s Disneyland and Disney World. They’re fine.

Source: NVIDIA
Author:
Manufacturer: ASUS

Specifications and Design

With all of the activity in both the GPU and CPU markets this year, it's hard to remember some of the launches in the first half of the year—including NVIDIA's GTX 1080 Ti. Maintaining the rank of fastest gaming GPU for the majority of the year, little has challenged NVIDIA's GP102-based offering, making it the defacto choice for high-end gamers.

Even though we've been giving a lot of attention to NVIDIA's new flagship TITAN V graphics card, the $3000 puts it out of the range of almost every gamer who doesn't have a day job involving deep learning.

IMG_5011.JPG

Today, we're taking a look back to the (slightly) more reasonable GP102 and the one of the most premiere offerings to feature it, the ASUS ROG Strix GTX 1080 Ti.

Hardware Specifications

While the actual specifications of the GP102 GPU onboard the ASUS Strix GTX 1080 Ti hasn't changed at all, let's take a moment to refresh ourselves on where it sits in regards to the rest of the market.

  RX Vega 64 Liquid RX Vega 56 GTX 1080 Ti GTX 1080 GTX 1070 Ti GTX 1070
GPU Cores 4096 3584 3584 2560 2432 1920
Base Clock 1406 MHz 1156 MHz 1480 MHz 1607 MHz 1607 MHz 1506 MHz
Boost Clock 1677 MHz 1471 MHz 1582 MHz 1733 MHz 1683 MHz 1683 MHz
Texture Units 256 256 224 160 152 120
ROP Units 64 64 88 64 64 64
Memory 8GB 8GB 11GB 8GB 8GB 8GB
Memory Clock 1890 MHz 1600 MHz 11000 MHz 10000 MHz 8000 MHz 8000 MHz
Memory Interface 2048-bit HBM2 2048-bit HBM2 352-bit G5X 256-bit G5X 256-bit 256-bit
Memory Bandwidth 484 GB/s 410 GB/s 484 GB/s 320 GB/s 256 GB/s 256 GB/s
TDP 345 watts 210 watts 250 watts 180 watts 180 watts 150 watts
Peak Compute 13.7 TFLOPS 10.5 TFLOPS 11.3 TFLOPS 8.2 TFLOPS 7.8 TFLOPS 5.7 TFLOPS
MSRP (current) $699 $399 $699 $499 $449 $399

If you'd like some additional details on the NVIDIA GTX 1080 Ti, or it's GP102 GPU, take a look at our review of the reference Founder's edition.

The GTX 1000 series of products from NVIDIA has marked a consolidation in ASUS's GPU offerings. Instead of having both Strix and Matrix products available, the Strix has supplanted everything to be the most premium option from ASUS for any given GPU, and the Strix GTX 1080 Ti doesn't disappoint.

IMG_5018.JPG

While it might not be the largest graphics card we've ever seen, the ASUS Strix GTX 1080 Ti is more massive in all dimensions compared to both the NVIDIA Founder's Edition card, as well as the EVGA ICX option we took a look at earlier this year. Compared to the Founder's Edition, the Strix GTX 1080 Ti is 1.23-in longer, 0.9-in taller, and takes up an extra PCIe slot in width.

Continue reading our review of the ASUS ROG Strix GTX 1080 Ti!!

Zotac Announces Factory Overclocked GTX 1080 Ti Mini

Subject: Graphics Cards | June 14, 2017 - 08:42 PM |
Tagged: zotac, gtx 1080 ti, factory overclocked, gp102, SFF

Zotac recently unveiled a slimmed down GTX 1080 Ti graphics card that uses a dual slot and dual fan cooler with a short PCB. The aptly named Zotac GTX 1080 Ti Mini measures 8.3” (211mm) long and will be the smallest GTX 1080 Ti on the market. Despite the miniaturization, Zotac is still offering a decent factory overclock on the Pascal GPU (but not the memory) with a boost clock of 1620 MHz versus the reference boost clock of 1582.

Zotac GP102 GTX 1080 Ti Graphics Card SFF.jpg

Zotac uses two 8-pin PCI-E power connectors to drive the card with its GTX 1080 Ti GPU (3584 CUDA cores) and 11GB of GDDR5X memory clocked at 11 GHz. The slimmed down graphics card features a metal backplate, dual shrouded fans, and a heatsink with aluminum fins and five 6mm heat pipes. The card has three DisplayPort 1.4 ports, one HDMI 2.0b port, and one DL-DVI output with the card supporting up to four simultaneous displays.

The Zotac GTX 1080 Ti Mini should enable quite a bit of horsepower in small form factor systems. The graphics card is model number ZT-P10810G-10P and Zotac has it listed on its website. Unfortunately, Zotac is not yet talking pricing or availability for the shortened card.

It appears that overclocking is not out of the question, but I am curious just how far it could be pushed especially in a small case with tight quarters and less airflow.

Source: Zotac

GTC 17: NVIDIA Demos (Professional) Multi-User VR

Subject: Graphics Cards | May 9, 2017 - 07:01 AM |
Tagged: VR, quadro, nvidia, gp102

Four Quadro P6000s installed in a single server, which looks like a 4U rack-mounted box, are shown running four HTC Vive Business Edition VR systems through virtual machines. It isn’t designed to be a shipping product, just a demo for NVIDIA’s GPU Technology Conference that was developed by their engineers, but that should get the attention of this trade show’s attendees, who are mostly enterprise-focused.

nvidia-2017-fouruservrserver.png

For context, this system has roughly equivalent GPU horsepower to four Titan Xps, albeit with twice the RAM and slightly different clocks; there’s plenty of power per headset to harness. Still, running this level of high-performance application on a virtual machine could be useful in a variety of business applications, from architectural visualization to, as NVIDIA notes, amusement parks.

Given that it’s just a proof-of-concept demo, you’ll need to build it yourself to get one. They didn’t mention using any special software, though.

Source: NVIDIA
Author:
Manufacturer: EVGA

Specifications and Design

When the GeForce GTX 1080 Ti launched last month it became the fastest consumer graphics card on the market, taking over a spot that NVIDIA had already laid claim to since the launch of the GTX 1080, and arguably before that with the GTX 980 Ti. Passing on the notion that the newly released Titan Xp is a graphics cards gamers should actually consider for their cash, the 1080 Ti continues to stand alone at the top. That is until NVIDIA comes up another new architecture or AMD surprises us all with the release of the Vega chip this summer.

NVIDIA board partners have the flexibility to build custom hardware around the GTX 1080 Ti design and the EVGA GeForce GTX 1080 Ti SC2 sporting iCX Technology is one of those new models. Today’s story is going to give you my thoughts and impressions on this card in a review – one with fewer benchmarks than you are used to see but one that covers all the primary differentiation points to consider over the reference/Founders Edition options.

Specifications and Design

The EVGA GTX 1080 Ti SC2 with iCX Technology takes the same GPU and memory technology shown off with the GTX 1080 Ti launch and gussies it up with higher clocks, a custom PCB with thermal sensors in 9 different locations, LEDs for externally monitoring the health of your card and a skeleton-like cooler design that is both effective and aggressive.

  EVGA 1080 Ti SC2 GTX 1080 Ti Titan X (Pascal) GTX 1080 GTX 980 Ti TITAN X GTX 980 R9 Fury X R9 Fury
GPU GP102 GP102 GP102 GP104 GM200 GM200 GM204 Fiji XT Fiji Pro
GPU Cores 3584 3584 3584 2560 2816 3072 2048 4096 3584
Base Clock 1557 MHz 1480 MHz 1417 MHz 1607 MHz 1000 MHz 1000 MHz 1126 MHz 1050 MHz 1000 MHz
Boost Clock 1671 MHz 1582 MHz 1480 MHz 1733 MHz 1076 MHz 1089 MHz 1216 MHz - -
Texture Units 224 224 224 160 176 192 128 256 224
ROP Units 88 88 96 64 96 96 64 64 64
Memory 11GB 11GB 12GB 8GB 6GB 12GB 4GB 4GB 4GB
Memory Clock 11000 MHz 11000 MHz 10000 MHz 10000 MHz 7000 MHz 7000 MHz 7000 MHz 500 MHz 500 MHz
Memory Interface 352-bit 352-bit 384-bit G5X 256-bit G5X 384-bit 384-bit 256-bit 4096-bit (HBM) 4096-bit (HBM)
Memory Bandwidth 484 GB/s 484 GB/s 480 GB/s 320 GB/s 336 GB/s 336 GB/s 224 GB/s 512 GB/s 512 GB/s
TDP 250 watts 250 watts 250 watts 180 watts 250 watts 250 watts 165 watts 275 watts 275 watts
Peak Compute 11.1 TFLOPS 10.6 TFLOPS 10.1 TFLOPS 8.2 TFLOPS 5.63 TFLOPS 6.14 TFLOPS 4.61 TFLOPS 8.60 TFLOPS 7.20 TFLOPS
Transistor Count 12.0B 12.0B 12.0B 7.2B 8.0B 8.0B 5.2B 8.9B 8.9B
Process Tech 16nm 16nm 16nm 16nm 28nm 28nm 28nm 28nm 28nm
MSRP (current) $719 $699 $1,200 $599 $649 $999 $499 $649 $549

Out of the box EVGA has overclocked the GTX 1080 Ti SC2 above reference specs. With a base clock of 1557 MHz and a GPU Boost clock of 1671 MHz, it has a 77 MHz jump on base and an 89 MHz jump on boost. Though moderate by some overclockers’ standards, that’s a healthy increase of 5.3% on the typical boost clock rate. The memory speed remains the same at 11.0 Gbps on 11GB, unchanged from the Founders Edition.

IMG_7629.jpg

I’m not going to walk through the other specifications of the GeForce GTX 1080 Ti GPU in general – I assume if you are looking at this story you are already well aware of it features and capabilities. If you need a refresh on this oddly-designed 352-bit memory bus behemoth, just read over the first page of my GeForce GTX 1080 Ti launch review.

DSC02824.JPG

Continue reading our review of the EVGA GeForce GTX 1080 Ti SC2!!

PCPer Live! GeForce GTX 1080 Ti Live Stream with Tom Petersen

Subject: General Tech, Graphics Cards | March 10, 2017 - 11:15 AM |
Tagged: video, tom petersen, pascal, nvidia, live, gtx 1080 ti, gtx, gp102, geforce

Our review of the GeForce GTX 1080 Ti 11GB graphics card is live and ready for consumption! Make sure you check it out before this afternoon's live stream!

Did you miss our GTX 1080 Ti Live Stream? Catch the reply below!

Ready your mind and body, it’s time for another GeForce GTX live stream hosted by PC Perspective’s Ryan Shrout and NVIDIA’s Tom Petersen. The general details about the GeForce GTX 1080 Ti graphics card are already official and based on the hype train and the response on social media, there is more than a little excitement.

box1.jpg

On hand to talk about the new graphics card will be Tom Petersen, well known in our community. While the GTX 1080 Ti will be the flagship part of our live stream we will also be diving into the world of VR performance evaluation and how the new FCAT VR tool will help reviewers and standard enthusiast see where their systems stand in producing smooth, effective virtual reality gaming. We have done quite a few awesome live steams with Tom in the past, check them out if you haven't already.

pcperlive.png

NVIDIA GeForce GTX 1080 Ti and FCAT VR Live Stream

1pm PT / 4pm ET - March 9th

PC Perspective Live! Page

Need a reminder? Join our live mailing list!

The event will take place Thursday, March 9th at 4pm ET / 1pm PT at http://www.pcper.com/live. There you’ll be able to catch the live video stream as well as use our chat room to interact with the audience, asking questions for me and Tom to answer live. 

Tom has a history of being both informative and entertaining and these live streaming events are always full of fun and technical information that you can get literally nowhere else. Previous streams have produced news as well – including statements on support for Adaptive Sync, release dates for displays and first-ever demos of triple display G-Sync functionality. You never know what’s going to happen or what will be said!

This just in fellow gamers: Tom is going to be providing a GeForce GTX 1080 Ti graphics card to give away during the live stream! We won't be able to ship it until the end of next week, but one lucky viewer of the live stream will be able to get their paws on the fastest graphics card we have ever tested!! Make sure you are scheduled to be here on March 9th at 1pm PT / 4pm ET!!

icon2.jpg

Win this beauty.

If you have questions, please leave them in the comments below and we'll look through them just before the start of the live stream. Of course you'll be able to tweet us questions @pcper and we'll be keeping an eye on the IRC chat as well for more inquiries. What do you want to know and hear from Tom or I?

So join us! Set your calendar for this coming Thursday at 4pm ET / 1pm PT and be here at PC Perspective to catch it. If you are a forgetful type of person, sign up for the PC Perspective Live mailing list that we use exclusively to notify users of upcoming live streaming events including these types of specials and our regular live podcast. I promise, no spam will be had!

The GTX 1080 Ti reviews are here; the card not so much

Subject: Graphics Cards | March 9, 2017 - 01:53 PM |
Tagged: 1080 ti, geforce, gp102, gtx 1080 ti, nvidia, pascal

As you have probably noticed from our front page, today is the day we can see how the GTX 1080 Ti performs in reviewers systems.  The unfortunate news is that you can't buy one yet nor do we know when you will be able to spend the $699 it will cost to order one.  We can share the performance with you, once again NVIDIA's Ti model takes the top spot out performing even the $1200 TITAN X.  As for overclocking the reference model, as we have not had a chance to test any cards with third party cooler on them, [H]ard|OCP were able to increase the GPU frequency over 200MHz to 1967-1987MHz in game and push the memory to 12GHz, somewhat better than what Ryan was able to.  Check out their full review here, with many more just below.

1489035168S7z42o2d6c_1_11_l.png

"NVIDIA is launching the fastest video card it offers for gaming today in the new $699 GeForce GTX 1080 Ti. We will take this video card and test it against the GeForce GTX 1080 and GeForce GTX TITAN X at 1440p and 4K resolutions to find out how it compares. Is it really faster than a $1200 GeForce GTX TITAN X Pascal?"

Here are some more Graphics Card articles from around the web:

Graphics Cards

Source: [H]ard|OCP
Author:
Manufacturer: NVIDIA

Flagship Performance Gets Cheaper

UPDATE! If you missed our launch day live stream, you can find the replay below:

It’s a very interesting time in the world of PC gaming hardware. We just saw the release of AMD’s Ryzen processor platform that shook up the processor market for the first time in a decade, AMD’s Vega architecture has been given the brand name “Vega”, and the anticipation for the first high-end competitive part from AMD since Hawaii grows as well. AMD was seemingly able to take advantage of Intel’s slow innovation pace on the processor and it was hoping to do the same to NVIDIA on the GPU. NVIDIA’s product line has been dominant in the mid and high-end gaming market since the 900-series with the 10-series products further cementing the lead.

box1.jpg

The most recent high end graphics card release came in the form of the updated Titan X based on the Pascal architecture. That was WAY back in August of 2016 – a full seven months ago! Since then we have seen very little change at the top end of the product lines and what little change we did see came from board vendors adding in technology and variation on the GTX 10-series.

Today we see the release of the new GeForce GTX 1080 Ti, a card that offers only a handful of noteworthy technological changes but instead is able to shake up the market by instigating pricing adjustments to make the performance offers more appealing, and lowering the price of everything else.

The GTX 1080 Ti GP102 GPU

I already wrote about the specifications of the GPU in the GTX 1080 Ti when it was announced last week, so here’s a simple recap.

  GTX 1080 Ti Titan X (Pascal) GTX 1080 GTX 980 Ti TITAN X GTX 980 R9 Fury X R9 Fury R9 Nano
GPU GP102 GP102 GP104 GM200 GM200 GM204 Fiji XT Fiji Pro Fiji XT
GPU Cores 3584 3584 2560 2816 3072 2048 4096 3584 4096
Base Clock 1480 MHz 1417 MHz 1607 MHz 1000 MHz 1000 MHz 1126 MHz 1050 MHz 1000 MHz up to 1000 MHz
Boost Clock 1582 MHz 1480 MHz 1733 MHz 1076 MHz 1089 MHz 1216 MHz - - -
Texture Units 224 224 160 176 192 128 256 224 256
ROP Units 88 96 64 96 96 64 64 64 64
Memory 11GB 12GB 8GB 6GB 12GB 4GB 4GB 4GB 4GB
Memory Clock 11000 MHz 10000 MHz 10000 MHz 7000 MHz 7000 MHz 7000 MHz 500 MHz 500 MHz 500 MHz
Memory Interface 352-bit 384-bit G5X 256-bit G5X 384-bit 384-bit 256-bit 4096-bit (HBM) 4096-bit (HBM) 4096-bit (HBM)
Memory Bandwidth 484 GB/s 480 GB/s 320 GB/s 336 GB/s 336 GB/s 224 GB/s 512 GB/s 512 GB/s 512 GB/s
TDP 250 watts 250 watts 180 watts 250 watts 250 watts 165 watts 275 watts 275 watts 175 watts
Peak Compute 10.6 TFLOPS 10.1 TFLOPS 8.2 TFLOPS 5.63 TFLOPS 6.14 TFLOPS 4.61 TFLOPS 8.60 TFLOPS 7.20 TFLOPS 8.19 TFLOPS
Transistor Count 12.0B 12.0B 7.2B 8.0B 8.0B 5.2B 8.9B 8.9B 8.9B
Process Tech 16nm 16nm 16nm 28nm 28nm 28nm 28nm 28nm 28nm
MSRP (current) $699 $1,200 $599 $649 $999 $499 $649 $549 $499

The GTX 1080 Ti looks a whole lot like the TITAN X launched in August of last year. Based on the 12B transistor GP102 chip, the new GTX 1080 Ti will have 3,584 CUDA core with a 1.60 GHz Boost clock. That gives it the same processor count as Titan X but with a slightly higher clock speed which should make the new GTX 1080 Ti slightly faster by at least a few percentage points and has a 4.7% edge in base clock compute capability. It has 28 SMs, 28 geometry units, 224 texture units.

GeForce_GTX_1080_Ti_Block_Diagram.png

Interestingly, the memory system on the GTX 1080 Ti gets adjusted – NVIDIA has disabled a single 32-bit memory controller to give the card a total of 352-bit wide bus and an odd-sounding 11GB memory capacity. The ROP count also drops to 88 units. Speaking of 11, the memory clock on the G5X implementation on GTX 1080 Ti will now run at 11 Gbps, a boost available to NVIDIA thanks to a chip revision from Micron and improvements to equalization and reverse signal distortion.

The move from 12GB of memory on the GP102-based Titan X to 11GB on the GTX 1080 Ti is an interesting move, and evokes memories of the GTX 970 fiasco where NVIDIA disabled a portion of that memory controller but left the memory that would have resided on it ON the board. At that point, what behaved as 3.5GB of memory at one speed and 500 MB at another speed, was the wrong move to make. But releasing the GTX 970 with "3.5GB" of memory would have seemed odd too. NVIDIA is not making the same mistake, instead building the GTX 1080 Ti with 11GB out the gate.

Continue reading our review of the NVIDIA GeForce GTX 1080 Ti 11GB graphics card!

NVIDIA Announces GeForce GTX 1080 Ti 11GB Graphics Card, $699, Available Next Week

Subject: Graphics Cards | February 28, 2017 - 10:59 PM |
Tagged: pascal, nvidia, gtx 1080 ti, gp102, geforce

Tonight at a GDC party hosted by CEO Jen-Hsun Huang, NVIDIA announced the GeForce GTX 1080 Ti graphics card, coming next week for $699. Let’s dive right into the specifications!

card1.jpg

  GTX 1080 Ti Titan X (Pascal) GTX 1080 GTX 980 Ti TITAN X GTX 980 R9 Fury X R9 Fury R9 Nano
GPU GP102 GP102 GP104 GM200 GM200 GM204 Fiji XT Fiji Pro Fiji XT
GPU Cores 3584 3584 2560 2816 3072 2048 4096 3584 4096
Base Clock 1480 MHz 1417 MHz 1607 MHz 1000 MHz 1000 MHz 1126 MHz 1050 MHz 1000 MHz up to 1000 MHz
Boost Clock 1600 MHz 1480 MHz 1733 MHz 1076 MHz 1089 MHz 1216 MHz - - -
Texture Units 224 224 160 176 192 128 256 224 256
ROP Units 88 96 64 96 96 64 64 64 64
Memory 11GB 12GB 8GB 6GB 12GB 4GB 4GB 4GB 4GB
Memory Clock 11000 MHz 10000 MHz 10000 MHz 7000 MHz 7000 MHz 7000 MHz 500 MHz 500 MHz 500 MHz
Memory Interface 352-bit 384-bit G5X 256-bit G5X 384-bit 384-bit 256-bit 4096-bit (HBM) 4096-bit (HBM) 4096-bit (HBM)
Memory Bandwidth 484 GB/s 480 GB/s 320 GB/s 336 GB/s 336 GB/s 224 GB/s 512 GB/s 512 GB/s 512 GB/s
TDP 250 watts 250 watts 180 watts 250 watts 250 watts 165 watts 275 watts 275 watts 175 watts
Peak Compute 10.6 TFLOPS 10.1 TFLOPS 8.2 TFLOPS 5.63 TFLOPS 6.14 TFLOPS 4.61 TFLOPS 8.60 TFLOPS 7.20 TFLOPS 8.19 TFLOPS
Transistor Count 12.0B 12.0B 7.2B 8.0B 8.0B 5.2B 8.9B 8.9B 8.9B
Process Tech 16nm 16nm 16nm 28nm 28nm 28nm 28nm 28nm 28nm
MSRP (current) $699 $1,200 $599 $649 $999 $499 $649 $549 $499

The GTX 1080 Ti looks a whole lot like the TITAN X launched in August of last year. Based on the 12B transistor GP102 chip, the new GTX 1080 Ti will have 3,584 CUDA core with a 1.60 GHz Boost clock. That gives it the same processor count as Titan X but with a slightly higher clock speed which should make the new GTX 1080 Ti slightly faster by at least a few percentage points and has a 4.7% edge in base clock compute capability. It has 28 SMs, 28 geometry units, 224 texture units.

archoverview.jpg

Interestingly, the memory system on the GTX 1080 Ti gets adjusted – NVIDIA has disabled a single 32-bit memory controller to give the card a total of 352-bit wide bus and an odd-sounding 11GB memory capacity. The ROP count also drops to 88 units. Speaking of 11, the memory clock on the G5X implementation on GTX 1080 Ti will now run at 11 Gbps, a boost available to NVIDIA thanks to a chip revision from Micron and improvements to equalization and reverse signal distortion.

memoryeye.jpg

The TDP of the new part is 250 watts, falling between the Titan X and the GTX 1080. That’s an interesting move considering that the GP102 was running at 250 watts with identical to the Titan product. The cooler has been improved compared to the GTX 1080, offering quieter fan speeds and lower temperatures when operating at the same power envelope.

coolerperf.jpg

Performance estimates from NVIDIA put the GTX 1080 Ti about 35% faster than the GTX 1080, the largest “kicker performance increase” that we have seen from a flagship Ti launch.

perf.jpg

Pricing is going to be set at $699 so don't expect to find this in any budget builds. But for the top performing GeForce card on the market, it's what we expect. It should be on virtual shelves starting next week.

(Side note, with the GTX 1080 getting a $100 price drop tonight, I think we'll find this new lineup very compelling to enthusiasts.)

card2.jpg

card3.jpg

NVIDIA did finally detail its tiled caching rendering technique. We'll be diving more into that in a separate article with a little more time for research.

One more thing…

In another interesting move, NVIDIA is going to be offering “overclocked” versions of the GTX 1080 and GTX 1060 with +1 Gbps memory speeds. Partners will be offering them with some undisclosed price premium.

1080oc.jpg

I don’t know how much performance this will give us but it’s clear that NVIDIA is preparing its lineup for the upcoming AMD Vega release.

GeForce_GTX_1080ti_3qtr_Front_Left_1488313915.jpg

We’ll have more news from NVIDIA and GDC as it comes!

Source: NVIDIA
Manufacturer: NVIDIA

Is Enterprise Ascending Outside of Consumer Viability?

So a couple of weeks have gone by since the Quadro P6000 (update: was announced) and the new Titan X launched. With them, we received a new chip: GP102. Since Fermi, NVIDIA has labeled their GPU designs with a G, followed by a single letter for the architecture (F, K, M, or P for Fermi, Kepler, Maxwell, and Pascal, respectively), which is then followed by a three digit number. The last digit is the most relevant one, however, as it separates designs by their intended size.

nvidia-2016-Quadro_P6000_7440.jpg

Typically, 0 corresponds to a ~550-600mm2 design, which is about as larger of a design that fabrication labs can create without error-prone techniques, like multiple exposures (update for clarity: trying to precisely overlap multiple designs to form a larger integrated circuit). 4 corresponds to ~300mm2, although GM204 was pretty large at 398mm2, which was likely to increase the core count while remaining on a 28nm process. Higher numbers, like 6 or 7, fill back the lower-end SKUs until NVIDIA essentially stops caring for that generation. So when we moved to Pascal, jumping two whole process nodes, NVIDIA looked at their wristwatches and said “about time to make another 300mm2 part, I guess?”

The GTX 1080 and the GTX 1070 (GP104, 314mm2) were born.

nvidia-2016-gtc-pascal-banner.png

NVIDIA already announced a 600mm2 part, though. The GP100 had 3840 CUDA cores, HBM2 memory, and an ideal ratio of 1:2:4 between FP64:FP32:FP16 performance. (A 64-bit chunk of memory can store one 64-bit value, two 32-bit values, or four 16-bit values, unless the register is attached to logic circuits that, while smaller, don't know how to operate on the data.) This increased ratio, even over Kepler's 1:6 FP64:FP32, is great for GPU compute, but wasted die area for today's (and tomorrow's) games. I'm predicting that it takes the wind out of Intel's sales, as Xeon Phi's 1:2 FP64:FP32 performance ratio is one of its major selling points, leading to its inclusion in many supercomputers.

Despite the HBM2 memory controller supposedly being actually smaller than GDDR5(X), NVIDIA could still save die space while still providing 3840 CUDA cores (despite disabling a few on Titan X). The trade-off is that FP64 and FP16 performance had to decrease dramatically, from 1:2 and 2:1 relative to FP32, all the way down to 1:32 and 1:64. This new design comes in at 471mm2, although it's $200 more expensive than what the 600mm2 products, GK110 and GM200, launched at. Smaller dies provide more products per wafer, and, better, the number of defective chips should be relatively constant.

Anyway, that aside, it puts NVIDIA in an interesting position. Splitting the xx0-class chip into xx0 and xx2 designs allows NVIDIA to lower the cost of their high-end gaming parts, although it cuts out hobbyists who buy a Titan for double-precision compute. More interestingly, it leaves around 150mm2 for AMD to sneak in a design that's FP32-centric, leaving them a potential performance crown.

nvidia-2016-pascal-volta-roadmap-extremetech.png

Image Credit: ExtremeTech

On the other hand, as fabrication node changes are becoming less frequent, it's possible that NVIDIA could be leaving itself room for Volta, too. Last month, it was rumored that NVIDIA would release two architectures at 16nm, in the same way that Maxwell shared 28nm with Kepler. In this case, Volta, on top of whatever other architectural advancements NVIDIA rolls into that design, can also grow a little in size. At that time, TSMC would have better yields, making a 600mm2 design less costly in terms of waste and recovery.

If this is the case, we could see the GPGPU folks receiving a new architecture once every second gaming (and professional graphics) architecture. That is, unless you are a hobbyist. If you are? I would need to be wrong, or NVIDIA would need to somehow bring their enterprise SKU into an affordable price point. The xx0 class seems to have been pushed up and out of viability for consumers.

Or, again, I could just be wrong.