Graphics Cards - Page 128

The totally next-gen PCI Express (PCIe) 7.0 specification has been announced today at the PCI-SIG Developers Conference 2022, ushering in yet another round of doubling the data rate: now up to 128GB/sec.

The new PCIe 7.0 x16 standard will offer up to 512GB/sec bi-directionally, and while your GPU would be sitting on its hands at that point -- graphics cards of today don't need the wicked fast upgraded speeds we're getting with PCIe 5.0, PCIe 6.0, and the just-announced PCIe 7.0 -- the new speeds will be a boon for HPC, AI, and supercomputers.

This means we're going to see next-gen PCIe 7.0 x4 NVMe M.2 SSD of the future pushing a huge 128GB/sec -- up from the 64GB/sec on PCIe 6.0 x4, the 32GB/sec on PCIe 5.0 x4, the 16GB/sec on PCIe 4.0 x4, and 8GB/sec on PCIe 3.0 x4. 128 gigabytes per second, transferring on that tiny M.2 SSD on your motherboard.

The first custom Intel Arc A380 desktop GPU has launched... well, in China at least... with GUNNIR's new Arc A380 Photon listed for $599.

GUNNIR's new custom Arc A380 Photon GPU is listed for 3999 RMB (around $595) on the official JD webstore, but it's unavailable because it's sold out (obviously). JD is waiting for its shipment, with all of the new Intel Arc A380 graphics cards shipping with slower GDDR6.

Intel updated its own Arc A380 website, detailing the 6GB of GDDR6 memory at 15.5Gbps bandwidth -- down from 16Gbps -- reducing the memory bandwidth down to 186GB/sec, down from 192GB/sec. Intel has also confirmed that its Arc A380 desktop GPUs launching in June are limited to pre-built PCs only, so the Chinese JD.com website is the only place allowed to sell the Arc A380 GPUs in China.

Intel's new Arc A380 desktop GPU has finally launched, at least in China, with its 6GB of GDDR6 memory being reduced in bandwidth silently.

GUNNIR launched its new custom Arc A380 Photon OC 6G graphics card last week with its 6GB of GDDR6 memory clocked at 15.5Gbps, down from 16Gbps. The 6GB of GDDR6 memory is on a 96-bit memory bus, with the 15.5Gbps bandwidth packing 186GB/sec bandwidth -- down from the 192GB/sec on the 16Gbps GDDR6.

The official Intel website and its Arc A380 GPU listing have 6GB of GDDR6 at 15.5Gbps bandwidth, with the company also clarifying some of the issues with what was happening with HDMI 2.1 on its Arc A380 graphics cards. Intel Arc A380 officially supports HDMI 2.1 but it is up to AIB partners to add it, partners... please use HDMI 2.1 and not HDMI 2.0 on ALL of your Arc GPUs.

I'm sure you've heard by now that the cryptocurrency market has had the arse fall out of it, with Bitcoin circling $20,000 or so -- leaving crypto miners, small, and big -- shuffling on the spot.

In a new report from Bloomberg, we're being told crypto miners across the world spent around $15 billion in 2021 alone... taking up more GPUs than gamers. Yes, crypto miners purchased more graphics cards than gamers, which should shock some, and not shock others that know crypto miners are buying 10s, 100s, or 1000s of graphics cards at once.

Bloomberg reported: "Ethereum miners have spent approximately $15 billion on graphics processing units (GPUs), according to Bitpro Consulting, and that doesn't include ancillary costs like wiring and transformers".

GALAX has just unveiled its new GALAX GeForce RTX 3060 Vegetables Fairy 'Cabbage Dog' special edition graphics card... which... well... I'll let you see it first:

What is this thing? The new card is based on the Vegetables Fairy concept by Japanese creator "PonkichiM" who envisions animals as foods and vegetables. Each of the animals are based on various foods, including: Garlic Cat, Shiitake Mushroom, Lotus Root, Vegetables Red Fox Radish, Vegetables Squirrel, and Cabbage Dog.

The GPU isn't any different, with GALAX using a custom PCB and shroud design for its new GALAX GeForce RTX 3060 Vegetables Fairy 'Cabbage Dog' special edition graphics card, but inside we have the GA106-302 GPU.

NVIDIA's next-gen AD102 GPU has been rumored to be a power-hungry monster since day one, with leaks suggesting between 600W and 900W of power consumption.

The new Ada Lovelace GPU architecture and its flagship AD102 GPU have a power limit of 800W according to leaker "kopite7kimi" but adds at the end of his tweet "I don't think we need to use the full power cap". I've said in previous articles that I think the 600-800W monster AD102 GPU we keep hearing about has to be a new TITAN RTX graphics card.

NVIDIA launching a new GeForce RTX 4090 Ti and or RTX 4090 with 800W of power sounds insane, but 600W doesn't... because the not-so-old GeForce RTX 3090 Ti uses 450W out of the gate. We keep hearing 600W+ but then these 800W rumors are just as common in the rumor mill. The leak includes that the AD102 GPU will use 450W, AD104 with 400W, and AD106 with 260W.

NVIDIA's new GeForce Game Ready 516.40 WHQL drivers are here, offering the usual best day-0 gaming experience for Fall Guy: Free for All. Not only that but there's the usual support for new and updated titles on the market.

This includes NVIDIA Reflex technology for ICARUS, as well as ray tracing for Jurassic World Evolution 2, Resident Evil 2, Resident Evil 3, and Resident Evil 7. Jurassic World Evolution 2: Dominion Biosyn Expansion is now here, which introduces Ray-Traced Ambient Occlusion (RTAO) effects, "making dinosaurs and your carefully-crafted dino parks look even better".

Resident Evil fans with GeForce RTX-powered GPUs on the PC will enjoy Resident Evil 2, Resident Evil 3, and Resident Evil 7 Biohazard PC releases with Ray-Traced Global Illumination (RTGI) and Ray-Traced Reflections. The new updates are available now, so if you haven't updated your game yet, once you do you'll get the new ray tracing effects ready with the new GeForce Game Ready 516.40 WHQL drivers.

GUNNIR has just unveiled the world's first custom Intel Arc A380 graphics card, with the introduction of the new GUNNIR Arc A380 Photon 6G OC graphics card. Check it out in all its custom Arc glory:

The new GUNNIR Arc A380 Photon graphics card features a tight dual-slot design, a single 8-pin power connector, and 6GB of GDDR6 @ 15.5Gbps (slower than the reference 16Gbps GDDR6). There's an aluminum block as a heat sink, keeping the Intel ACM-G11 GPU cool.

Intel's new Arc A380 desktop GPU has a 2000MHz base GPU clock and 75W TBP, with the custom offering from GUNNIR in the new Arc A380 Photon 6G OC rolling out with a maximum GPU clock of 2450MHz and higher 92W TBP. We don't know how the base + maximum GPU clocks work just yet, but with the "OC" branding on the card it'll be interesting to see what sub 100W can do.

AMD has been detailing more and more about its next-gen RDNA 3 GPU architecture, but now leaker "Greymon55" has said when the launch will take place.

The next-gen RDNA 3-based AMD Radeon RX 7000 series GPUs will reportedly be launching in "late October, no later than mid-November" with a nice 'wink' emoji at the end. The leaker quickly followed it up with a tweet, starting with a 'whisper' emoji and then '3rd gen DNA'.

Before AMD unravels and releases its new Radeon RX 7000 series CPUs we'll see the company push out its next-gen Zen 4-based Ryzen 7000 series CPUs. The new Zen 4 processors will be on a new AM5 socket, with the upcoming X670E + X670 motherboards packing both next-gen DDR5 and PCIe 5.0 technologies. The new RDNA 3-based GPUs are rumored to be new PCIe 5.0 x16 on the flagship Navi 31-based graphics cards.

AMD confirmed a few tidbits for its next-gen RDNA 3 and RDNA 4 GPU architectures during its recent Financial Analyst Day 2022 where Senior Vice President of Radeon Technologies Group, David Wang, showed off the roadmap for next-gen RDNA and CDNA GPUs.

In the update, AMD confirmed that its next-gen GPU architecture after RDNA 3 debuts later this year, is RDNA 4 which is expected in 2024. In the slide, we can see that AMD shows us that RDNA was made on 7nm, RDNA 2 was also made on 7nm, next-gen RDNA 3 is made on 5nm, and the next-next-gen RDNA 4 is made on an "Advanced Node".

This isn't new for anyone who keeps up with GPU rumors, where I reported on rumors about RDNA 4 using 3nm and 5nm nodes for RDNA 4 MCM GPUs (multi-chip module, or chiplet design as AMD just confirmed for its RDNA 3 GPUs). Back in August 2020, I was reporting from leaks that said the RDNA 4 is a "full product line with a common architecture, and RDNA 4 MCM will use 3nm+5nm". It seems RDNA 3 will be made on 5nm, and RDNA 4 to be made on 3nm+5nm (with GPU tiles on 3nm + I/O tile on 5nm).

AMD has provided more details for its next-gen RDNA 3 GPU architecture, confirming that RDNA 3 features a chiplet design -- compared to the monolithic GPU we've gotten used to.

The news on RDNA 3 is coming from AMD itself during its recent Financial Analyst Day 2022 where Senior Vice President of Radeon Technologies Group, David Wang, showed off the roadmap for next-gen RDNA and CDNA GPUs. We now know that AMD is promising over 50% more performance-per-watt with RDNA 3 over RDNA 2, cementing the rumors that NVIDIA's next-gen GPUs will be using FAR more power than RDNA 3 GPUs.

AMD says that the 50% performance-per-watt uplift is achieved through the shift to the new 5nm process node by TSMC, advanced chiplet packaging, rearchitected compute unit, optimized graphics pipeline, and next-gen AMD Infinity Cache technology. This means that we can expect a significant performance jump from new Radeon RX 7000 series cards, that won't require 1200W+ power supplies and 4 x 8-pin power connectors.

SK hynix announced its super-fast next-gen HBM3 memory in October 2021, and now it's here: powering NVIDIA's new H100 Tensor Core GPU, SK Hynix has kicked off mass production of HBM3 DRAM.

The announcement comes in on schedule to allow NVIDIA to unleash its new H100 Tensor Core GPUs, with NVIDIA recently completing its performance evaluation of SK hynix's HBM3 samples. SK hynix will be providing HBM3 for NVIDIA H100 GPUs starting in Q3 2022, with the South Korean giant to "expand HBM3 volume in the first half in accordance with NVIDIA's schedule".

SK hynix will be providing HBM3 memory that will be pushing 819GB/sec memory bandwidth, which is the equivalent to the transmission of 163 x 1080p movies (5GB in size) every, single, second. Kevin (Jongwon) Noh, president, and chief marketing officer at SK hynix said: "We aim to become a solution provider that deeply understands and addresses our customers' needs through continuous open collaboration".

NVIDIA's new GeForce RTX 40 series GPUs are still a few months away, but now we're hearing more about a chunky reference cooler with a triple-fan design.

According to leaker "kopite7kimi" who teases said: "at least, they designed a triple-fan cooler for reference board of AD102". If you look at all of the custom GeForce RTX 3090 Ti graphics cards -- where I've benched both the MSI GeForce RTX 3090 Ti SUPRIM X and ASUS ROG Strix LC GeForce RTX 3090 Ti OC Edition -- both with chunky triple-slot-and-beyond coolers.

This isn't really news, as we already heard that AIB partners were "preparing 4.5-slot coolers" for an "undisclosed" NVIDIA GPU from leaker Tom @ Moore's Law is Dead. Now, leaker "kopite7kimi" says that NVIDIA has a beefy reference cooler with a triple-fan cooler. We might not see this turn into a reality, but it wouldn't surprise me whatsoever.

Intel's new Arc Alchemist A730M graphics card has been tested in the flesh, inside of the new Machenike Discovery Edition 2022 laptop which is available for sale in China.

We've now got a couple of gaming benchmarks with the Intel ACM-G10 GPU inside of the Arc Alchemist A730M graphics card, with both Metro Exodus Enhanced Edition and Hitman 2. The new Arc Alchemist A730M with the ACM-G10 GPU was tested against the NVIDIA GeForce RTX 3060 Laptop GPU by Golden Pig Upgrade.

Both of the laptops used were running an Intel Core i7-12700H processor, where in Metro Exodus Enhanced Edition at 1080p on High quality the Arc a730M outperforms the RTX 3060M by 6.5%. But then Arc a730M gets its ass whooped by RTX 3060M in Hitman 2 with 76FPS versus 100FPS.

NVIDIA's new GeForce RTX 4090 is expected to be twice as fast as the second-fastest Ampere flagship -- the GeForce RTX 3090 -- but now we're hearing rumors it'll be OVER twice as fast.

In a new tweet, leaker "kopite7kimi" says that "nothing changed yet" with the GeForce RTX 4090 and that the "performance will be slightly higher than twice the RTX 3090". To clarify, he said "I mean RTX 4090 > 2x RTX 3090". Yeah, it can't get any clearer than that.

We already knew the GeForce RTX 4090 would be twice as fast, but OVER twice as fast isn't too much of a stretch. Not when we're hearing rumors that NVIDIA is testing an even faster AD102-based GPU with up to 800-900W of power. This is a TITAN-class Ada Lovelace GPU... so having headroom between "over twice as fast as the RTX 3090" and a new up-to-900W graphics card is making much, much more sense.

AMD's new Radeon RX 7000 series GPUs aren't official yet, but Seasonic is jumping the gun with its new wattage calculator... accidentally pre-announcing some of the cards.

The new Radeon RX 7900 XT, Radeon RX 7800 XT, and Radeon RX 7700 XT were all teased in the wattage calculator -- which lets users estimate their PC power requirements with various parts -- and includes AMD's new Radeon RX 7000 series graphics cards.

In these estimates, which we're not holding much up to... the Radeon RX 7900 XT requires a 750W PSU for the GPU, and an 850W PSU if you include an 8-core AMD Ryzen CPU. The slightly cut-down Radeon RX 7800 XT will reportedly have the same power supply requirements, while there's 100W less power required for the Radeon RX 7700 XT.

NVIDIA's next-gen GeForce RTX 4080 is going to be using a decent amount more power over the current-gen GPUs, with the latest rumor on the RTX 4080 is that it will have a 420W TGP.

The rumor is coming from leaker "kopite7kimi" who tweeted that a possible GeForce RTX 4080 SKU with the PG139-SKU360 board is on its way, with a 420W TGP. Kopite7kimi says that the GeForce RTX 4080 "still uses AD103" which lines up with previous rumors, but the 420W TGP is new: 100W more than the RTX 3080, and this makes sense.

We just heard that the mid-range GeForce RTX 4060 will use 220W of power, but I explained in that post yesterday that the RTX 4060 should beat (or at least get very close to) the performance of the higher-end, but current-gen GeForce RTX 3080.

Intel has shown off its Arc Alchemist desktop GPU at its recent Intel Extreme Masters (IEM) 2022 event in Dallas, Texas, USA. Check it out:

I don't know if I'd call this a "real" card as it would be a dud that wouldn't work, wouldn't have final hardware, and probably wouldn't even turn on. We just get a look at what the card might look like if it's ever released because right now it's the Duke Nukem Forever of graphics cards.

Intel had its dual-fan Alchemist ACM-G10 GPU "on display" with the full 32 Xe-Core setup, which should have 16GB of GDDR6 memory when it launches. You can also see that it has an 8+6-pin power connector setup, and I'm sure this part won't change when Intel eventually launches its Arc Alchemist desktop GPUs.

NVIDIA's next-gen GeForce RTX 4060 is going to be a powerhouse compared to the regular x060 series GPU from NVIDIA, but I thought that was obvious by now.

If the new GeForce RTX 4090 graphics card is going to be twice as fast as the GeForce RTX 3090, then we'd expect the GeForce RTX 4060 to be twice as fast as the GeForce RTX 3060... right? Architecture improvements, faster VRAM, more power, tweaks everywhere else... the benefits would be there for all SKUs.

But in a new tweet, leaker "kopite7kimi" says that the GeForce RTX 4060 would be using more power than the RTX 3070, which isn't so shocking -- given I prefaced this article with the fact that we are to expect the GeForce RTX 4060 to be twice as fast as the GeForce RTX 3060 -- so surely it'll use more power, and tip over the RTX 3070 TDP.

EVGA's new monster GeForce RTX 3090 Ti KINGPIN graphics card has broken a bunch of world records, with the card being pushed by South Korean overclocker Biso Biso.

The new EVGA GeForce RTX 3090 Ti KINGPIN graphics card was chilled with LN2 cooling, with overclocker Biso Biso using an EVGA Z690 KINGPIN DARK motherboard and Intel Core i9-12900K processor at up to 5.4GHz in an 8+2 configuration. G.SKILL Trident Z5 memory was used, with DDR5-6400 speeds (CL32) in a very extreme custom-loop setup.

Diving right into the GPU goodness: Biso Biso overclocked the EVGA GeForce RTX 3090 Ti KINGPIN to 2925MHz, very close to the 3GHz mark. LN2 cooling had the GA102 GPU at -40C, with the 24GB of GDDR6X memory overclocked to 24Gbps -- nothing surprising there, I had already achieved that on my MSI GeForce RTX 3090 Ti SUPRIM X in an article about that here.