Intel / AMD & AMD / NVIDIA
125 members have voted
Recently Browsing 0 members
No registered users viewing this page.
The Liquid Cooled reference RX 6900 XT card has an unexpected trick up its sleeve
by Sayan Sen
AMD launched the Radeon RX 6900 XT, its flagship product this GPU generation, in December last year. The company it seems has moved on from its blower-style designs and instead the reference RX 6900 XT brought on a cooler with triple axial fans. However, that apparently is not the only reference 6900 XT in existence as a liquid-cooled (LC) version of the reference GPU has also been spotted a few times since.
The reference RX 6900 XT LC allegedly packs the binned XTXH variant of the Navi 21 chip, the same chip that is also present in PowerColor's Liquid Devil Ultimate as well as in Sapphire's TOXIC Extreme Edition. However, and rather bizarrely, it seems the reference 6900 XT LC is the one that actually features the fastest memory clock among the three, even though the other two models are aftermarket versions and are supposed to have the better specs.
The LC 6900 XT model was tested today by a Baidu Tieba user who spotted this deviation. The memory on this new card is apparently clocked at 18.48Gbps whereas all other 6000 series Radeon cards are locked at 17.2 Gbps. That's an increase of close to 7.5%.
Below is an image showing the specs of the card. While the language may be Chinese, the specification details are fairly easy to figure out.
As expected, the card, being liquid-cooled and also packing a binned chip, is faster and cooler than the air-cooled reference 6900 XT. The faster memory though doesn't seem to be helping the reference 6900 XT LC much in terms of performance against the other liquid-cooled 6900 XT models.
It is alleged that AMD originally planned to launch these chips to DIY gamers worldwide but due to the shortage has changed tactic. The Radeon RX 6900 XT LC, for now, is only available to Chinese OEMs and system integrators.
Source and images: 寒山虹光 (Baidu Tieba)
AMD's Zen 4 could be a behemoth with up to 128 cores in a single socket
by Sayan Sen
AMD's Zen 4 is the next big revision of the company's Zen CPU micro-architecture and lately, information related to the upcoming platform has been spilling out fast. According to the latest rumor today, each next-gen EPYC server processor based on Zen 4 (codenamed 'Genoa') will pack up to 128 cores, which is double that of what AMD offers in its current EPYC 7003 lineup.
The image below shows the 64-core layout of an EPYC 7703 (Milan) processor:
The rumor isn't completely new as earlier there were reports of Zen 4 allegedly having more than 64 cores with new instructions like AVX-512, BFloat16, and more. These new instructions are helpful for high-performance computing (HPC) and server workloads so the alleged addition of them definitely makes sense.
Intel added AVX-512 instructions to its CPUs with the Rocket Lake architecture and the gains in compatible workloads are truly impressive. According to AnandTech, even an 8-core Rocket Lake-S part was able to win against a 64-core Zen 2 EPYC processor in 3D Particle Movement AVX-enabled benchmark.
While Zen 4 and Genoa are still a while away, it is already known that fourth-gen EPYC processors will be deployed inside an upcoming exascale supercomputer dubbed "El Capitan". El Capitan is expected to release in 2023 and it will be used for overlooking U.S. nuclear research and operations.
Source: Vegeta (Twitter) | Image via zhangzhonghao (Chiphell forum)
Here are more details on AMD's big.LITTLE CPU architecture leak
by Sayan Sen
Pretty much like Intel is doing, AMD too has been working on its own hybrid processor architecture consisting of big and little cores. We came to know about this from a leaked patent last year (via @Underfox3). Today we have new information on the development as Twitter user @Kepler_L2 has spotted one of AMD's new patents related to big.LITTLE published a few days back.
The patent outlines how task processing between the two types of cores would be handled in this hybrid approach.
According to this patent, the little cores will have a time threshold built-in and sensors will monitor the length of time it runs at its full clock speed. Once the threshold is crossed, the task will be handed over to the big core. A similar process would be carried out for memory-intense workloads if it runs at its highest frequency state for higher than the threshold time.
That's because the idea behind the use of the little cores is to save power and running them at full speeds for long durations defeats that purpose.
For the big cores, the implementation is exactly the opposite. In essence, if a workload running on the bigger core does not cross the threshold, the task is sent to the little cores since clearly so much processing power does not seem to be necessary for the workload.
Going back to the patent from last year, the architectural block diagram of the big.LITTLE design approach was described in it.
Both the cores will have their own dedicated L1 cache but they will share the pool of L2 between them.
Source: @Kepler_L2 (Twitter) | Images via FreePatentsOnline (1), (2)
AMD's next-gen RDNA 3 performance jump rumored to be absolutely insane
by Sayan Sen
AMD's second-generation RDNA architecture (RDNA 2) was generally praised by reviewers for the performance and power efficiency gains it was able to achieve despite being on the same 7nm node as RDNA. However, this was AMD's first time implementing hardware-accelerated ray tracing (RT), and the results for this, compared to Nvidia's RT capabilities, were far less impressive. That is all set to change according to a report by RedGamingTech (RGT).
The report claims that AMD's RDNA 3 ray tracing performance will get a significant uplift and will be very competitive with what Nvidia offers. It also adds that RDNA 3 will be utilizing a next-gen, "smarter" Ray Tracing IP 2 that could enable it to even leapfrog Nvidia's RT performance. The architecture will also feature new Machine Learning instructions.
AMD hasn't forgotten about the rasterization performance of RDNA 3 either as a leakster on Twitter alleges that Navi 31, Navi 32, and Navi 33 will respectively offer 2.8x, 2.2x, and 1.5x times the performance of AMD's current best, the Radeon RX 6900 XT.
Unknown at this point is how exactly AMD could be achieving this uplift. Whether the performance claims purported here mean the company will be adding more compute units (CUs) to the 80 units on the 6900 XT, or if the improvement is purely based on per CU architectural and clock gains. Or perhaps it's a combination of all of them.
That said, it is important to note that this is all based on speculation and unconfirmed reports for now, so it is advisable to take these rumors with a grain of salt.
Source: vegeta (Twitter) via RGT (YouTube)
By Abhay V
Nvidia to drop Game Ready Driver updates for Windows 7, 8, and 8.1 starting this October
by Abhay Venkatesh
Nvidia today detailed its plans for Game Ready Drivers upgrade support for Windows 7, Windows 8, and Windows 8.1. The company posted a support article that states that it will cease to provide Game Ready Driver updates for its graphic cards for the mentioned versions starting October 2021. However, it does note that it will continue to serve “critical security updates” for systems running those operating systems until September 2024.
Microsoft ended support for Windows 7 in January 2020, while Windows 8 lost its support in January 2016 – a short life span for the OS thanks to Windows 8.1 and the debacle that Windows 8 was. However, while Windows 8.1 reached the end of mainstream support back in 2018, the OS is still being serviced with security updates and will be till January 2023.
Nvidia, says that a “vast majority” of its GeForce customers have migrated to Windows 10 and that it aims to provide the “best possible security, support, and functionality” for those users, which is why it is focusing on Windows 10 alone. In the FAQ section, it adds that it will ship the last Game Ready Driver that supports the three operating system on August 31, with the first drivers to drop support for the versions completely expected to ship in October.
The change might not be a major one considering that most users are running the latest offering from the Redmond giant. However, for those that are still on older versions, they can rest assured that their GPUs will be served with updates to address any critical vulnerabilities. However, they will lose out on upgrades with performance enhancements, new features, and bug fixes, Nvidia says.