GTX 1060 3GB Turns 10: How NVIDIA Cut A 1280-Core Pascal GPU Down To 1152

By Indie Kings | October 2, 2026

Updated October 2, 2026: The GeForce GTX 1060 3GB turns ten years old on August 18, 2026, ten years after NVIDIA launched a card that was a 6GB GTX 1060 with one of its ten Streaming Multiprocessors physically disabled. The NVIDIA TITAN X (Pascal) reaches the same anniversary on August 2, 2026, and its ten-year legacy is the opposite problem: it was the first $1,200 consumer card NVIDIA built, and NVIDIA's own GeForce GTX 1080 Ti beat it seven months later at $699.

Two Pascal cards hit their tenth anniversaries within the same month of 2026, and the two cuts NVIDIA made to build them still define what each card was. The GeForce GTX 1060 3GB launched on August 18, 2016, and it was a 6GB card with one of its ten Streaming Multiprocessors switched off. The NVIDIA TITAN X (Pascal) launched on August 2, 2016, and it was a consumer graphics card that cost $1,200 and was beaten in games, by NVIDIA's own claim, by a $699 card seven months later.

VideoCardz, by WhyCry, marked the 1060 3GB anniversary on August 29, 2026, in a piece titled "NVIDIA launched GeForce GTX 1060 3GB 10 years ago, but it wasn't the last 3GB GeForce." VideoCardz, by the same author, marked the TITAN X anniversary on August 2, 2026, in "NVIDIA TITAN X (Pascal) turns 10." Both pieces are retrospective rather than prescriptive, and neither makes a buying case for anything.

Two design decisions, ten years apart from each other, sit on the same August calendar. One put a disabled compute unit on a cheap retail shelf. The other put a $1,200 price tag on the most expensive consumer card NVIDIA had yet built. Both decisions were reversed in a way that is easier to see now than it was at the time.

Two Pascal cards, ten years, sixteen days apart

The NVIDIA TITAN X (Pascal) was released on August 2, 2016. It had been announced on July 21, 2016 at $1,200, with direct availability through NVIDIA in North America and Europe.

The GeForce GTX 1060 3GB launched on August 18, 2016, about one month after the original GTX 1060 6GB. NVIDIA had originally planned an August 24, 2016 launch and then moved it forward to August 18.

Officially the flagship was called the NVIDIA TITAN X. The "Pascal" suffix was adopted by buyers and reviewers rather than by NVIDIA, and the reason was the existence of an earlier card with the same name. The GeForce GTX TITAN X based on Maxwell GM200 had launched in March 2015, so two different products in the TITAN X line would otherwise be indistinguishable in a search result.

The naming collision is itself a small piece of the decade. NVIDIA shipped a product name that it had already used, and the industry spent the following months appending the architecture name to keep the two apart. Pascal itself refers to Blaise Pascal, Maxwell refers to James Clerk Maxwell, and NVIDIA's architecture names in that era were drawn from scientist and mathematician surnames.

Per the VideoCardz coverage, both cards were built on 16nm FinFET silicon. The 1060 3GB is a cut-down GP106 part, and the TITAN X is the first consumer product built on GP102.

The 3GB card that was never just a halved card

The GTX 1060 3GB was announced as the cheaper version of a card that had launched at $249. The 3GB model launched at $199 MSRP, and the memory reduction was the part NVIDIA put in the marketing.

What the marketing did not put front and center was that the 3GB model was not only a memory cut. NVIDIA also disabled one of the ten Streaming Multiprocessors on the GP106 GPU. A Streaming Multiprocessor is the unit that holds CUDA cores and texture units, so removing one of ten removed a whole slice of the chip's throughput rather than only its storage.

CUDA cores went from 1,280 on the 6GB card to 1,152 on the 3GB card. Texture units went from 80 to 72. ROPs were unchanged at 48 on both. The silicon designations differed too, with GP106-400 on the 6GB model and GP106-300 on the 3GB model, both on 16nm FinFET.

The consequence was that the 3GB card was slower in workloads that did not exceed its smaller framebuffer. That is the part that separates this from a memory downgrade. A pure VRAM cut is invisible until a game or application runs out of framebuffer. The disabled Streaming Multiprocessor was felt below that limit, because there were fewer compute units doing the work.

Several headline specifications were deliberately left untouched. Both versions kept a 192-bit GDDR5 memory interface, memory stayed at 8 Gbps, and bandwidth stayed at 192 GB/s on both. Reference clocks were unchanged at 1506 MHz base and 1708 MHz boost. Board power was unchanged at 120W. Max compute performance was 4.4 TFLOPs on the 6GB card and 3.9 TFLOPs on the 3GB card.

The 3GB model also did not get the Founders Edition design. NVIDIA reserved its own cooler and shroud for the $249 6GB card, which launched in July 2016 at $249 MSRP and $299 as a Founders Edition, and the $199 card shipped through board partners instead. Those partners moved quickly, and custom models built on the 1,152-core configuration followed the reference launch.

The pulled-forward date is the small detail worth sitting with. A launch moved six days earlier than planned, and the card that arrived was slower than the one above it in every configuration NVIDIA had deliberately preserved except memory capacity and one disabled Streaming Multiprocessor.

The full GTX 1060 3GB spec table

The 1060 3GB and 6GB cards share a memory interface, a memory speed, a clock setup and a power limit. They do not share a GPU configuration. The table below places the two NVIDIA cards side by side.

The Radeon RX 480 column is present because the 1060 3GB was positioned against it in 2016, but our sources do not state its specifications, so every entry there is marked for verification rather than filled in.

Image: ASUS Dual GeForce GTX 1060 3GB OC Edition. Credit: Wikimedia Commons, CC BY-SA 4.0, ITEagle Europe.

Read down the two NVIDIA columns and the pattern is clear. Every number that could have been trimmed to justify a cheaper bin was trimmed, and every number that would have made the card obviously compromised was preserved. The GPU lost one of ten Streaming Multiprocessors, 128 CUDA cores and eight texture units. Everything else stayed identical.

That is why the 1060 3GB is still a useful case study in what a memory cut does not tell you. A shopper comparing 3GB against 6GB would reasonably assume the cheaper card is the same GPU with less storage. In this case it was the same GPU with less compute as well.

The last 3GB GeForce

The 1060 3GB was not NVIDIA's final desktop GeForce card with 3GB of memory. That is the explicit framing of the VideoCardz anniversary piece, whose title states that the card was not the last 3GB GeForce.

In May 2018 NVIDIA introduced the GeForce GTX 1050 3GB. That card used the full 768-core GP107 configuration, so no Streaming Multiprocessor was removed. What it narrowed instead was the memory interface, which dropped to a 96-bit bus.

The contrast between the two cards is the interesting part. In 2016 NVIDIA built a 3GB card by disabling compute and keeping the memory bus. In 2018 NVIDIA built a 3GB card by keeping all of the compute and narrowing the bus. The same memory capacity was reached through two different cuts, and the second cut preserved the part of the chip that determines how fast the card can work when the framebuffer is not full.

That reversal says something about where NVIDIA's 3GB floor eventually sat. A memory bus is a capacity limit, and a disabled Streaming Multiprocessor is a speed limit. The 2018 card kept the speed and accepted the capacity constraint, which is a different engineering priority than the one NVIDIA applied to the 2016 card.

Neither the 1060 3GB nor the 1050 3GB is being recommended here, and our sources do not state current street prices for either. This is a retrospective on two launch decisions, not a buying guide, and what these cards do in games today is not stated in our sources.

The $1,200 card

The NVIDIA TITAN X (Pascal) was announced on July 21, 2016 at $1,200, with direct availability through NVIDIA in North America and Europe. It released on August 2, 2016. Per the VideoCardz piece, that price reads differently in 2026.

The hardware was GP102 on 16nm FinFET, at about 12 billion transistors. The card carried 3,584 CUDA cores with a 1,417 MHz base clock and a 1,531 MHz boost clock. Memory was 12GB of GDDR5X on a 384-bit bus running at 10 Gbps effective, giving about 480 GB/s of bandwidth. The power envelope was 250W, supplied through one 6-pin and one 8-pin PCIe power connector.

The chassis was the angular, tessellated Founders Edition shroud used across the GeForce GTX 10-series era. It was not the first product to use that design language, because NVIDIA had already introduced the same construction approach on the GeForce GTX 1080 Founders Edition earlier in 2016. Reviewers later noted that the TITAN X used the same style of blower-based assembly.

GP102 had more to give than the TITAN X shipped with. NVIDIA announced the full Quadro P6000, a workstation card on the same GPU, four days after the TITAN X release. The Quadro P6000 enabled all 3,840 CUDA cores and doubled memory to 24GB. The TITAN X was the first GP102 product to reach buyers, but it was not the first GP102 product to be announced.

The sequence is the point. A consumer card went on sale first, a workstation card with the complete die followed four days later, and the full die came back to the consumer market seven months afterward as the TITAN Xp.

Seven months later, the GTX 1080 Ti

The GeForce GTX 1080 Ti released on March 10, 2017, seven months after the TITAN X. It used the same GP102 silicon and carried the same 3,584 CUDA cores. It launched at $699.

Memory went down and bandwidth went up. The GTX 1080 Ti reduced memory to 11GB and used a 352-bit interface, but faster 11 Gbps GDDR5X produced 484 GB/s of bandwidth, slightly more than the 480 GB/s in the TITAN X.

NVIDIA said the GTX 1080 Ti was faster in games than the $1,200 TITAN X. That claim is the single most consequential line in the VideoCardz TITAN X retrospective, because it is the manufacturer making the comparison rather than a reviewer.

The same period also closed the gap on the disabled silicon. The fully enabled GP102 returned to the consumer market as the TITAN Xp in April 2017, one month after the GTX 1080 Ti and eight months after the TITAN X. That card restored the 3,840 CUDA cores that the Quadro P6000 had shown off four days after launch.

What happened between August 2016 and March 2017 is a pricing structure unwinding. A $1,200 consumer flagship was followed by a $699 consumer flagship with the same core count and marginally more memory bandwidth, and then by a corrected version of the $1,200 card with the complete die enabled. Per the VideoCardz piece, we do not get new TITANS now, and the source observes that they seem to have been replaced by x090 and x090Ti class GPUs.

The TITAN line, as VideoCardz lists it

The VideoCardz TITAN X retrospective includes a table of the TITAN line, reproduced here in the source's own order and with its own figures. The list runs from a cancelled Ada Lovelace part back to the first Kepler TITAN.

Three rows in that table share the same GPU and the same power figure. The TITAN X, the TITAN Xp and the TITAN Xp SW variants are all GP102 at 250W with 12GB of GDDR5X, separated only by year. The line also includes two Star Wars themed TITAN Xp variants in 2017, both built on the same GP102 configuration as the standard TITAN Xp.

The table also shows how short the top of the line was. The only Ada Lovelace entry is cancelled, and the highest board power listed anywhere in the series is 800W on that cancelled card. The listed high-water mark for shipping consumer silicon is the TITAN RTX at 280W in 2018.

Board power for the 2016 TITAN X and the 2017 TITAN Xp is identical at 250W, which is what you would expect when a product line changes its bin counts rather than its power target.

What ten years actually changed

The most concrete thing that changed for both of these cards is support. Pascal is a generation for which NVIDIA has ended driver support, alongside Maxwell and Volta. Our sources state that driver support has ended for those three architectures, and they do not state a specific driver version or end date, so any such version number is marked [NEEDS VERIFICATION] here. No download links are included in this article.

That changes what a ten-year-old Pascal card is in 2026. A driver-supported card can receive new features with new games. A card past the end of its driver support cannot, and the practical consequence over a ten-year span is much larger for a 2016 part than it would have been in 2018.

What has not changed is the hardware story. The 1060 3GB still has one of its ten Streaming Multiprocessors disabled and still carries 1,152 CUDA cores where the 6GB card carries 1,280. The TITAN X (Pascal) still had 3,584 of GP102's 3,840 CUDA cores enabled, and the full die reached consumers four months after it did, as the TITAN Xp in April 2017.

The two cards bracket the range of what a launch decision can do. On one side, a $50 price difference was achieved by removing compute as well as memory from a chip, producing a card that was slower before it ever ran out of framebuffer. On the other side, a $1,200 price was attached to the top of a stack that NVIDIA undercut with a consumer card of the same core count seven months later.

Both decisions were visible at the time and both were reversed. The 1060 3GB's approach to a 3GB card was replaced in May 2018 by the GTX 1050 3GB, which kept all 768 GP107 cores and narrowed the bus to 96 bits instead. The TITAN X's approach to a $1,200 card was replaced in March 2017 by a $699 GTX 1080 Ti on the same GPU, with the full die returning to the consumer market as the TITAN Xp in April 2017.

None of this is a recommendation, and no prices are offered for any card in this article beyond the launch MSRP figures stated in our sources. What these cards do in current games is not stated in our sources, and any current street price is not stated in our sources either.

FAQ

When did the GeForce GTX 1060 3GB launch?
The GeForce GTX 1060 3GB launched on August 18, 2016, at $199 MSRP, about one month after the $249 6GB model. NVIDIA had originally planned an August 24, 2016 launch and moved it forward by six days.

Why was the GTX 1060 3GB slower than the 6GB version?
NVIDIA disabled one of the ten Streaming Multiprocessors on the GP106 GPU, which cut CUDA cores from 1,280 to 1,152 and texture units from 80 to 72. Because compute was reduced along with memory, the 3GB card was slower even in workloads that did not exceed its smaller framebuffer.

What were the full specs of the NVIDIA TITAN X (Pascal)?
The TITAN X (Pascal) used GP102 on 16nm FinFET at about 12 billion transistors, with 3,584 CUDA cores, a 1,417 MHz base clock, a 1,531 MHz boost clock, and 12GB of GDDR5X on a 384-bit bus at 10 Gbps for about 480 GB/s of bandwidth. Its power envelope was 250W with one 6-pin and one 8-pin PCIe power connector.

Was the TITAN X beaten by the GeForce GTX 1080 Ti?
NVIDIA said yes. The GeForce GTX 1080 Ti released on March 10, 2017 with the same GP102 silicon and the same 3,584 CUDA cores, at a $699 launch price, and NVIDIA stated that it was faster in games than the $1,200 TITAN X.

Was the GTX 1060 3GB the last 3GB GeForce card?
No. In May 2018 NVIDIA introduced the GeForce GTX 1050 3GB, which used the full 768-core GP107 configuration but paired it with a narrower 96-bit memory bus. That card kept all of the compute and cut capacity instead, the opposite approach from the 2016 3GB card.

Can Pascal cards still receive driver updates?
Pascal is a generation for which NVIDIA has ended driver support, alongside Maxwell and Volta. Our sources do not state a specific driver version or end date, so any such number is marked [NEEDS VERIFICATION] in this article, and no download links are provided.

Bottom Line

The GTX 1060 3GB and the TITAN X (Pascal) both turn ten in August 2026, and both mark a decision NVIDIA later undid. The 1060 3GB showed that a cheaper card can be made by disabling compute rather than only memory, and NVIDIA took the opposite approach two years later with the GTX 1050 3GB. The TITAN X showed that a $1,200 consumer flagship can be undercut by the manufacturer's own $699 card seven months later, and NVIDIA corrected the bin with the TITAN Xp in April 2017. Pascal is also a generation for which NVIDIA has ended driver support, which is the most practical fact about owning either card today.

Sources: VideoCardz by WhyCry, "NVIDIA launched GeForce GTX 1060 3GB 10 years ago, but it wasn't the last 3GB GeForce," published August 29, 2026, relaying NVIDIA's launch specifications and the GP106-400 versus GP106-300 configuration split. And VideoCardz by WhyCry, "NVIDIA TITAN X (Pascal) turns 10," published August 2, 2026, relaying NVIDIA's specifications and the TITAN series lineup. No purchase links, product links, or driver download links are included in this article. Items marked [NEEDS VERIFICATION] are listed in the text above.