Today we find the GPU Flashback Archive delving into the not so distant past to focus on the NVIDIA 900 series of graphics cards, the first to use NVIDIA’s new Maxwell architecture which had already seen the light day in mobile GPU solutions, an indication of the direction that the company were taking at the time. Let’s take a look at the cards that were launched as part of the 900 Series, the improvements and changes that Maxwell brought and some of the more memorable scores that have been posted on HWBOT.
NVIDIA GeForce 900: Overview
The first question one may well have regarding the NVIDIA 900 series is simple – what happened to the 800 series? To answer the question fully, you must first look at the direction that NVIDIA was moving at the time. A movement to expand its product offerings in order to compete in the quickly expanding mobile SoC market. The suddenly ubiquity of Android-based smartphones around the globe was fuelled in part by the development of mobile SoCs from Qualcomm, Samsung, Mediatek, Marvell, Allwinner and others. The traditional feature phone was quickly being replaced by smartphones that now required improved multi-core CPU performance, HD display support and, importantly from NVIDIA’s perspective, decent enough graphics processing to actually play 3D games. Intel and NVIDIA were two companies with plenty of R&D and marketing budget who sought to enter a new market to help bolster revenues during an inevitable slow down of desktop PC sales, a traditional cash cow for both.
NVIDIA knew that it had to improve its game in terms of performance per watt and overall efficiency. NVIDIA had been developing SoC parts under the Tegra moniker since 2008, but despite some notable design wins, the Zune HD perhaps being a highlight, Tegra had never challenged the big guns of Qualcomm and Samsung. The Tegra K1 launched in Q2 2014 and featured a Kepler graphics architecture, an architecture design for PC gaming and Supercomputer applications. It was not ideal for mobile. NVIDIA’s solution to this was to put all of their efforts into creating a new and more efficient architecture. It was codenamed Maxwell.
To return to the question posed earlier, the NVIDIA 8 Series does exist as a set of mobile PC GPUs for notebooks and ultrabooks, but it never showed its face as a desktop part. Maxwell part one arrived as GMxx parts in early 2014 and finally arrived (perhaps too late to make impact) on the Tegra X1 SoC that powered the Nvidia Shield, Google Pixel C and Nintendo Switch. However, by this stage in the game, it was getting more and more difficult for NVIDIA and Intel to penetrate a market dominated by four or five players. The rest is history.
Maxwell finally arrived as a full blown desktop part in September 2014 with the launch of the GeForce GTX 980 and GTX 970, both based on the NVIDIA GM204 GPU, the second coming of Maxwell. The Maxwell Part I series had totally focused on reducing power draw to create a GPU design that could compete with competing offerings such as Qualcomm’s Krait and PowerVR from Imagine Technologies. Maxwell Part II would be a matter of building up a much bigger version that could advance the company’s desktop GPU lineup. The result is a leaner, meaner GPU that beats the previous generation in terms of performance, but actually not by much.
Here’s a very Titan-esque GTX 980 reference card with distinctive NVIDIA branding.
The NVIDIA GTX 980 used a Maxwell-based GM204 GPU, physically, a much smaller GPU than the GK110 that adorned the GTX 780. The GM204 had a die that was 398 mm², considerably reduced compared to the GK110 which was 561 mm² in size. The Maxwell chip featured 5.2 billion transistors compared to it’s Kelper predecessor with more than 7.2 billion. What we are looking at is essentially ‘Big Kepler’ vs ‘Small Maxwell’. Have no fear, ‘Big maxwell’ would arrive six months later powering a new Titan card, the GTX Titan X. This would soon be joined by the GTX 980 Ti card, the true flagship desktop card in most people’s minds. Both GTX 980 Ti and the Titan X used the GM200 GPU, a ‘Big Maxwell’ monster at 601 mm² and 8 billion transistors.
The Most Popular NVIDIA GeForce 900 Card: The GeForce GTX 970
Let’s take a look at the most popular NVIDIA 900 series cards in terms of submission numbers to the HWBOT database. Although not technically a member of the GeForce 900 series family, we have included the GTX Titan X in the data along with 8 Series mobile GPUs. After all they are all Maxwell architecture chips.
- -GeForce GTX 970 – 43.30%
- -GeForce GTX 980 Ti – 16.70%
- -GeForce GTX 980 – 16%
- -GeForce GTX 960 – 12.8%
- -GeForce GTX 980M – 4.1%
- -GeForce GTX 970M – 3.6%
- -GeForce GTX 950 – 2.7%
- -GeForce GTX Titan X – 2.6%
- -GeForce 960M – 2.4%
- -GeForce GTX 860M -2%
As noted earlier, for most gamers the GeForce GTX 980 card was a hard sell at $549 USD seeing as the GTX 780 Ti had dropped from its original price of $699 and could be found at very attractive prices around this time. However, at $329 USD the GTX 970 was actually quite a decent value proposition. The most powerful, non-Titan, consumer grade single GPU 900 series card is of the GTX 980 Ti, a card that was breaking records on its first day and never looked back. This will be the focus later on when we look at some of the more outstanding scores in the HWBOT database.
We can also take a look at the GTX 980 Ti now. It arrived on the scene on June 1st 2015, being launched during the Computex 2015 trade show in Taipei, Taiwan. Fitted with a larger GM200 GPU it featured 2,816 CUDA Cores, 176 TMUs and 96 ROPs. This is slightly fewer than you find on the Titan X, but substantially more than what we got with the GTX 980. In terms of comparison, this chart from Anandtech puts most of the data into perspective..
It’s clear that the GTX Titan X remains top dog in terms of CUDA Cores, texture units and other specifications, but in truth the arrival of the GTX 980 Ti made it irrelevant to most enthusiasts, gamers and indeed overclockers. The GTX 980 Ti was fitted with 6GB of GDDR5 tuned to 1,753 MHz (7,012 MHz effective) and had a default GPU boost clock of 1,076MHz. Other specs include support for PCIe 3.0, a 250W TDP, 1x 6-pin and 1x 8-pin power connectors all wrapped in a cooling shroud that was very similar to previous Titan cards. The card measures 267mm in length, took up two slots and featured 1x DVI, 1x HDMI and 3x DisplayPort outputs.
Here’s a shot of the reference design GTX 980 Ti as presented at launch:
AIB vendors were soon pumping out their own designs. Here’s an example from GALAX who had refined their distinctive white Hall of Fame offering:
This is an ASUS STRIX take on a GTX 1080 Ti card:
NVIDIA GeForce 900 Series: Record Scores
We can now take a look at some of the highest scores posted on HWBOT using an NVIDIA GeForce 980 Ti card, the fastest single GPU (non-Titan) card in the 900 Series lineup.
Highest GPU Frequency
Although technically speaking, GPU frequency (as with CPU frequency) is not a true benchmark, it does remain an important metric for many overclockers. Looking through the database, we find that the submission with the highest GPU core frequency using a GeForce GTX 980 Ti card comes from Xtreme Addict (Poland). He pushed the GPU of a GALAX GeForce GTX 980 Ti Hall of Fame card to 2,220MHz, which is an impressive +106.32% beyond stock settings. The rig used also included an Intel Core i7 6950X ‘Broadwell-E’ processor clocked at 5,300MHz (+76.67%)).
You can find the submission from XA here on HWBOT: http://hwbot.org/submission/3233730_xtreme_addict_3dmark___fire_strike_geforce_gtx_980_ti_27452_marks
3DMark Fire Strike Extreme
The highest 3DMark Fire Strike Extreme score submitted to HWBOT using a single NVIDIA GeForce 780 Ti card comes from legendary GPU pusher k|ngp|n (US). He pushed a EVGA GeForce GTX 980 Ti Kingpin Edition to 2,203MHz (+104.74%) on the GPU and 2,170MHz (+23.79%) on the memory to hit a score of 14,432 marks. The rig also used a very nicely pushed Intel Core i7 5960X ‘Haswell-E’ processor clocked at 5,765MHz (+92.17%).
You can find the submission from k|ngp|n here on HWBOT: http://hwbot.org/submission/3051525_kingpin_3dmark___fire_strike_extreme_geforce_gtx_980_ti_14432_marks
Here’s a shot of the EVGA GeForce GTX 980 Ti Kingpin Edition card in question:.
In the classic Aquamark benchmark we find that HWBOT No.1 Dancop (Germany) is the highest scorer with a single GeForce GTX 980 Ti card. He holds the Hardware First Place record and Global First Place record with an impressive score of 694,174 marks. The score was made with the Maxwell GM200 GPU clocked at 1,537.7MHz (+42.91%) with an Intel Core i7 8700K ‘Coffee Lake’ chip clocked at 7,245.62MHz (+95.83%). As a matter of fact, the score was made earlier this month, underlining the fact that in some benchmarks, the GTX 980 Ti remains the card of choice.
You can find the submission from Dancop here on HWBOT: http://hwbot.org/submission/3724124_dancop_aquamark_geforce_gtx_980_ti_694174_marks
Here’s a shot of the rig in action:
Thanks for joining us for this week’s episode of the GPU Flashback Archive series. Come back next week and check out the current NVIDIA GeForce 1000 series of graphics processors and cards.