Video Cards & GPUs News - Page 8

"NVIDIA's GeForce GTX 1080 Ti is the most amazing graphics card to come in a very long time," Anthony wrote in his review of the NVIDIA GeForce GTX 1080 Ti in 2017 - seven years ago (and three NVIDIA GPU generations). The flagship GeForce GTX 1080 Ti was a game changer, offering a 35% performance uplift compared to the baseline GeForce GTX 1080 for just $699 USD.

It was expensive at the time, but hindsight is everything. Inflation aside, in 2024, the idea of a flagship GPU priced at $699 is wild-complex hardware, and advanced nodes have steadily driven prices up since the GeForce GTX 1080 Ti hit the scene.

The crown jewel of the Pascal generation, the 1080 Ti is powered by the GP102 chip with its 16nm process, 3584 CUDA Cores, Boost Clock speed of 1582 MHz, and 11 GB or GDDR5X memory.

NVIDIA's next-gen GeForce RTX 50 Series of GPUs, the 'Blackwell' generation, is on track to launch either later this year or early next year, with the flagship GeForce RTX 5090 and RTX 5080 leading the pack.

Rumors point to impressive performance gains, part of which could come from NVIDIA's use of next-gen GDDR7 memory. We've been following this closely here at TweakTown, with recent word from a well-known industry insider and leaker "kopite7kimi" confirming on X that the flagship Blackwell GPUs in the upcoming GeForce RTX 50 Series would be using a similar memory interface as the existing GeForce RTX 40 Series.

That would mean the same 384-bit for the GeForce RTX 5090 found in the GeForce RTX 4090. So, where does GDDR7 fit into this picture? Especially now, with new info emerging about its speed.

With AMD's recent global launch (or re-launch) of the Radeon RX 7900 GRE for $549, the mid-range discrete GPU market got upsized with another great value option. The $400 to $600 USD price range for new graphics cards is pretty stacked at the moment, and with AMD offering the Radeon RX 7900 GRE for the same price as the GeForce RTX 4070, the company also officially dropped the price of the Radeon RX 7700 XT.

At the time, AMD noted that pricing for Radeon RX 7700 XT models would drop to as low as $419; however, over at US retailer Newegg, you can pick up the SAPPHIRE PULSE Radeon RX 7700 XT 12GB reference model for $399 USD after entering a promo code at checkout.

This puts its price on par with the GeForce RTX 4060 Ti, a card it outperforms by 16.7% in 1440p gaming workloads. In our full review of the Sapphire PULSE AMD Radeon RX 7700 XT, we noted that it was an excellent alternative to NVIDIA's option-however, the price was too close to the Radeon RX 7800 XT.

NVIDIA is preparing its next-generation Blackwell GPU architecture as you're reading this, but now NVIDIA CEO Jensen Huang has said that the company's next DGX AI system will be liquid-cooled.

This might not seem like much, but data centers aren't easy to liquid cool... it will be a huge infrastructure investment to continue down this path, but the insatiable demand for AI is driving it down this road. NVIDIA's next-gen Blackwell B100 AI GPU is expected to be unveiled at NVIDIA's upcoming GPU Technology Conference (GTC) event, while Dell recently teased the beefed-up B200 for 2025.

NVIDIA's next-gen B200 AI GPU is expected to draw 1000W+ of power when it drops in 2025, so hearing Jensen talk about the new DGX AI system being liquid-cooled makes sense if you consider those power numbers.

ZOTAC's new GeForce RTX 4090 D PGF OC Edition graphics card has been reviewed, and it is the largest graphics card on the planet... measuring in at 381mm long (!!!).

Expreview has tested the new ZOTAC RTX 4090 D PGF graphics card, with the downgraded internals meeting the China-specific US export rules. ZOTAC revised its GeForce RTX 4090 design with the RTX 4090 D PGF, with 3 x 11 fans and a vapor chamber cooler that features "high-frequency stability... comparable to that of water-cooled solutions."

The review of ZOTAC's new China-specific GeForce RTX 4090 D PGF graphics card from Expreview explained: "(The PGF) has high maximum TGP (530 W) and a powerful 28-phase power PCB design. Despite the technological headroom, the card struggles to offer much of the overclocking potential. The team from Expreview only managed to squeeze 3.7% more performance from this card. That's despite 24.7% more power theoretically available."

NVIDIA's next-generation GeForce RTX 50 series "Blackwell" gaming GPUs will feature a memory interface configuration that's not a big departure from the current-gen RTX 50 "Ada Lovelace" GPUs.

In a new post on X, leaker "kopite7kimi" said that "although I still have fantasies about 512-bit, the memory interface configuration of GB20x is not much different from that of AD10x." To give some context there, "GB20x" are the GPU SKUs of the next-gen Blackwell GPUs, while AD10x are the SKUs for Ada Lovelace. AD102 powers the GeForce RTX 4090, for example.

NVIDIA will be upgrading to next-gen GDDR7 memory on its GeForce RTX 50 series graphics cards, where if we see the same 384-bit memory bus on the next-gen RTX 5090, we should expect a 50% upgrade in memory bandwidth with the slowest GDDR7 memory providing the RTX 5090 with 1.5TB/sec of memory bandwidth (the RTX 4090 has 1TB/sec memory bandwidth).

NVIDIA will be the first to use next-gen GDDR7 memory on its next-gen GeForce RTX 50 series "Blackwell" GPUs, where we could see the first-generation GPUs using GDDR7 featuring 16Gb GDDR7 modules, which means the minimum VRAM capacity for next-gen graphics card will begin at 2GB.

GDDR6 started off with 8Gb dies and 1GB capacity per module, but now, hardware leaker "kopite7kimi" has posted on X, saying, "We will use 16Gbit first." I don't think we're going to get some beast GeForce RTX 5090 with double the VRAM -- so, 48GB of GDDR7, at first -- but it means that we'll get similar, or the same VRAM capacity but at wicked bandwidth levels compared to current-gen cards.

For example, lower-end graphics cards using 8GB of 32Gbps GDDR7 memory on a 192-bit memory bus will feature 512GB/sec of memory bandwidth. Meanwhile, a 192-bit memory bus ramps up to 768GB/sec, and a mid-range 256-bit memory bus will feature 1TB/sec of memory bandwidth (that's as much as the RTX 4090).

Samsung will reportedly show off its new HBM3E 12-Hi stack memory at NVIDIA's upcoming GPU Technology Conference (GTC) between March 18 and 21.

NVIDIA is a major partner for Samsung's bleeding-edge HBM memory, part of the heart and soul of its AI GPUs coming soon in the form of the Hopper H200 (a beefed-up HBM3E-based version of H100) and its next-gen Blackwell B100 AI GPU that are both featuring HBM3E memory.

Samsung will exhibit its new high-capacity 36GB HBM3E 12-Hi stack memory in physical form at NVIDIA GTC 2024. Samsung is expected to begin mass production in the coming months, with NVIDIA as a major customer for its H200 and B100 AI GPUs.

SK hynix reportedly provided NVIDIA with its latest 12-layer HBM3E memory modules in Universal Test Vehicle (UTV) form for its upcoming next-generation AI GPU hardware.

The news is coming from ZDNet Korea, which reports that SK hynix provided initial samples of its new 12-layer DRAM stacked HBM3E memory to NVIDIA "last month." The site reports that the 12-layer HBM3E memory samples provided to NVIDIA are an "extremely early version and is mainly used to establish standards and characteristics of new products. SK hynix calls it UTV (Universal Test Vehicle)."'

SK hynix recently completed the performance validation required for its 8-layer HBM3E memory, with the same performance verification on the new 12-layer HBM3E memory to "not take much more time," reports ZDNet Korea.

It seems that NVIDIA is responding to mid-range pricing pressure from AMD, and has taken the hatchet to the price of the RTX 4060 Ti 16GB, with this graphics card coming down close to the $400 mark in the US (or at least one model has).

If you've been following the discounts in this GPU bracket recently, you probably saw that AMD's RX 7700 XT has dropped down to $409 over at Newegg (following an official price cut to $419). There are two models at that cheaper price, from Sapphire (Pulse) and ASRock (Challenger).

That's a fair drop from the MSRP of $449 that the RX 7700 XT launched at, which at the time made the GPU look distinctly unfavorable compared to the more powerful 7800 XT which was only just over 10% more expensive.