Recent studies have found that GDDR6 memory, despite its impressive bandwidth, suffers from significantly higher latency compared to DDR4 and DDR5. The studies were conducted using an AMD 4800S desktop kit, which stands out for its use of an Xbox Series X System on Chip (SoC) and GDDR6 memory.
During testing, AMD's console-specific CPUs did not perform as expected, falling short of the Ryzen 5 3600, even with the additional memory bandwidth provided by GDDR6 and the inclusion of two extra cores in the Xbox chip.
Tests performed on games such as Call of Duty: Black Ops: Cold War, Cyberpunk 2077, Crysis 3 Remastered, and others revealed that cache and memory latency play a critical role in increasing game performance on the CPU side.
Therefore, despite GDDR6's high bandwidth capabilities, the significant memory latency can make it less attractive for use in modern desktop CPUs.
The supposed prototype of the NVIDIA GeForce RTX 40 series' quad-slot GPU cooler has leaked yet again, showcasing the superior engineering design of this unrealized technology. This cooler, rumored to be an early model for the reportedly cancelled GeForce RTX 4090 Ti and Titan RTX Ada GPUs, reveals astonishing innovations.
The cooler design involves a vertical PCB connection for the graphics card, diverging from traditional horizontal connections. Two hefty copper wires lead from the 16-pin power plug straight into the PCB, suggesting the card was meant to exceed 400W of power, possibly tapping into the full 600W limit.
The cooling solution incorporated a total of 22 heat pipes and a vapor chamber similar to those employed in the RTX 4090 and RTX 4080 Founders Edition graphics cards. An additional third fan was included to ensure airflow within the card, especially targeting the DRAM dies on the backside of the PCB.
AMD's new RX 7900 GRE has been unveiled, revealing an unexpected redesign in its GPU lineup. The new model deviates from the established form with a physically smaller package and a unique PCB. Most notably, the memory bus width has been trimmed down, transitioning from the 384-bit found in the RX 7900 XTX to a leaner 256-bit bus in the GRE.
This restructuring has resulted in a decrease in the number of memory packages installed on the PCB, facilitating a more compact layout. While it's uncommon for such a drastic redesign to occur for just one SKU, this move could signal AMD's forward-looking strategy for its future models. Based on the alterations in the 7900 GRE, similar transformations might be expected in the forthcoming RX 7800 series or the equivalent mobile GPUs. Though it represents a considerable shift, AMD could be ushering in a new era of compact and efficient designs for its future lineups.
AMD is readying a budget-friendly workstation solution, the Radeon Pro W7600, based on the RDNA 3 GPU architecture. The card was spotted within the Geekbench database, being tested on an Intel Core i9-12900K PC.
The AMD Radeon Pro W7600 will feature a Navi 33 chip with 32 compute units (16 WGPs), which means we're looking at 2048 cores, a similar configuration to the Radeon RX 7600 (gaming GPU). The card has a clock speed of 1940 MHz, providing about 16 TFLOPs of compute power, slightly lower than the RX 7600 which features a higher clock speed.
The AMD Radeon Pro W7600 will have the same memory capacity as the RX 7600 - 8 GB based on GDDR6 with a 128-bit bus interface. This also means there's 32 MB of Infinity Cache onboard. The power consumption is likely to be around 150W.
AMD has not specified a release date for the Radeon PRO W7600, but it could be expected by Q4 2023. It can be expected that the Radeon PRO W7600 will be cheaper than its counterparts like the Radeon Pro W7900.
AMD's latest Adrenalin 23.7.2 graphics driver update has caused significant performance issues. Twitter user @carygolomb reported a drastic decrease in the performance of his handheld gaming device after installing the update, with his integrated Radeon 780M GPU being rebranded to a slower 760M model.
In addition, the user experienced the malfunctioning of certain internal and external devices when the new driver was activated, including a fingerprint reader, sound card, and an Xbox 360 game controller.
Digital Foundry confirmed the performance problems, noting a 33% decrease in the performance of their Aya Neo handheld gaming console equipped with a Ryzen 7 7840U processor while running the new driver.
The official list of known issues with Adrenalin 23.7.2 includes high idle power consumption on RX 7000 series GPUs, faulty performance metrics overlay in some games, and widely known ray tracing crashes in Rachet and Clank: Rift Apart.
At this point, it's hard to determine the extent of the performance issues, but they seem particularly relevant for the mobile Ryzen 7000 series processors.
As the demand for AI GPUs such as NVIDIA's H800 model drastically increases in China, there has been a significant hike in their prices. Major companies like Alibaba and ByteDance are actively purchasing these GPUs for their server applications, leading to a market shortage. Rumor has it that ByteDance has acquired NVIDIA's AI GPUs worth about $1 billion.
NVIDIA has been selling cut-down versions of its H100 and A100 GPUs in China as a workaround to US sanctions. Despite their lower performance, these GPUs are extremely popular in the Chinese market. The prices for the A800 and H800 models have reached record levels of 500,000 yuan, nearing $69,000.
But that's not all. As it turns out, NVIDIA also sets its product prices based on "relationships" with clients. Having "good relations" with NVIDIA's CEO Jensen Huang can play a role in making substantial orders.
All of this has been made possible thanks to NVIDIA's monopoly in the field of AI GPUs. However, in light of the rapidly evolving industry and the emergence of new competitors, such as Intel with their "Gaudi2 AI Accelerators", this "monopoly bubble" may burst at any moment.
AMD has officially announced the release of the new Radeon RX 7900 GRE, also known as the "Golden Rabbit Edition". This new RDNA 3 product is intended to compete with the RTX 4070 from NVIDIA.
The Radeon RX 7900 GRE is equipped with a trimmed-down Navi 31 GPU featuring 80 RDNA 3 compute units or 5120 stream processors, slightly less than the 7900 XT, which comes with 84 compute units or 5376 stream processors. The graphics card also boasts 16 GB of memory.
The frequencies of the AMD Radeon RX 7900 GRE are maintained at up to 2245 MHz, but with a base frequency of just 1270 MHz, 730 MHz less than the XT variant. The G6 memory operates at 18 Gbps speeds, delivering 576 GB/s of bandwidth.
In China, the card will retail for 5299 RMB ($740 USD), while in North America it will be priced at $649 USD. This makes it $200 USD cheaper than the RX 7900 XT, and 400 RMB cheaper than the Chinese retail price of the 7900 XT.
Rumors suggest that AMD is also preparing to launch the Radeon RX 7800 XT and RX 7700 XT graphics cards in September.
Despite Seagate starting commercial shipments of new hard drives based on Heat-Assisted Magnetic Recording (HAMR) technology, it will continue to produce models based on Perpendicular Magnetic Recording (PMR) and Shingled Magnetic Recording (SMR). Dave Mosley, the head of Seagate, confirmed that the development of the last PMR product is nearing completion and will allow the drive capacities to extend into the upper 20TB range.
"Our efforts on developing what may be our last PMR product are nearing completion and will extend drive capacities into the mid- to upper 20TB range," Mosley stated.
It's worth noting that new HAMR drives have capacities exceeding 30TB and are already being shipped to select hyperscale clients for quality verification. Such clients need every bit of available space, so they appreciate such capacity. Although HAMR drives consume slightly more power than the typical PMR and SMR drives, their usage is justified by the increased capacity.
NVIDIA is rumored to be planning a significant upgrade for their next flagship graphics card. Insider Kopite7kimi has shared information that NVIDIA's next-generation "Ada-Next" GPUs will utilize a massive 512-bit bus interface.
This is a major change, as the last GPU from NVIDIA to feature a memory configuration above 384-bit was the GeForce GTX 285, which was released back in 2009. Such a formidable memory configuration would allow NVIDIA to experiment with large memory capacities. Current cards with a 384-bit interface, such as the RTX 4090 and RTX 6000 Ada, feature memory capacities up to 24 GB and 48 GB respectively. With a 512-bit bus interface, graphics cards could have memory capacities of up to 32 GB, and even up to 64 GB on a single card, which is a 33% increase from the current options.
This change is likely to come along with the introduction of the new GDDR7 memory standard being developed by companies like Micron and Samsung. The new "Ada-Next" based cards are expected to hit the market in 2025.
According to recent rumors, NVIDIA might cancel plans to launch the GeForce RTX 4090 Ti. Insider Kopite7kimi, known for accurate leaks in the past, reported on the possible cancellation of this GPU, which was supposed to use the more powerful AD102 "Ada" chip.
It was also noted that NVIDIA is working on two new GPUs within the updated GeForce RTX 4070 and RTX 4060 series. Presumably, these new models will be based on modified AD103 and AD106 chips.
Nevertheless, keep in mind that NVIDIA's plans can change.
