AMD
AMD has published a document relating to a potential security vulnerability that can be exploited on processors based on the Zen 3 microarchitecture. The side-channel exploit is similar to Specter, which is known to target large numbers of Intel CPUs and more. With Zen 3, AMD introduced a new technology called Predictive Store Forwarding (PSF), which helps improve code execution performance by predicting the relationship between loading and storing data. PSF's predictions are accurate in most cases, but there is still a small chance that the prediction is not accurate, which results in speculation about the incorrect CPU.AMD engineers have found that a erroneous speculation PSF is equivalent to Specter v4. Software that relies on isolation or "sandboxing" is highly at risk when it comes to bad speculation. AMD has provided two scenarios in which such a circumstance can occur:
First, it is possible that the store / load pair have had a dependency for a while, but subsequently stop having it. This can occur if the address of the "store" or "load" operation changes during the execution of the program. Second, erroneous PSF predictions can occur if there is an alias in the PSF predictor structure. The PSF predictor is designed to track store / load pairs based on parts of their RIP. It is possible that a pair that does not have a dependency may have an alias in the predictor with another store / load pair that does not. This could cause erroneous speculation when the second store / load pair is executed.
AMD concludes that Predictive Store Forwarding helps improve application performance, but also brings security complications. However, the chip maker has not spotted any code considered vulnerable to PSF's erroneous predictions nor have there been any reported cases of such exploit, and therefore, the security risk remains low for most applications.
Consumers working with software that uses sandboxing and are alarmed by this discovery may choose to disable the Predictive Store Forwarding feature. AMD recently proposed patches for Linux that would also disable PSF. Phoronix colleagues have carried out some preliminary tests on the impact of disabling the feature, revealing a negligible drop in performance.
Are you looking for a good motherboard to pair with the new Ryzen processors? ASUS ROG Strix X570-F with 14 power phases might be a good choice. You can find it on Amazon at a discounted price.
AMD May Have Inadvertently Revealed Some Specifications For Its Next Mainstream Radeon GPU
The Recently Released AMD Radeon RX 6700 XT.
Marco Chiappetta - HotHardwareAMD recently released a patch for Linux, that may have revealed some key specifications of an as yet unannounced Radeon GPU. In an update to the AMD Kernel Fusion Driver, or AMDKFD, the company detailed L1 cache information and added L2/3 cache information for its Vega 10 and new ASICs. Amidst the release, there are details regarding a GPU codenamed “dimgrey_cavefish”, which reveal that GPU’s AMD Infinity Cache configuration of 32MB. To date, AMD hasn’t announced any GPUs with such a configuration.
AMD introduced Infinity Cache with its RDNA 2-based Radeon RX 6000 series GPUs. The top-end Radeon RX 6800 and RX 6900 series cards, based on the Navi 21 GPU, have 128MB of Infinity Cache. According to AMD, when using is Infinity Cache in conjunction with a 256-bit GDDR6 memory interface, bandwidth delivered to the GPU is effectively more than doubled. Infinity Cache improves absolute performance and bandwidth per watt, because it can feed data to the graphics pipeline at lower latencies and with a much higher cache hit rate, which enhances overall efficiency. The 256-bit memory interface on Navi 21, when coupled to 16Gbps of GDDR6 memory, offers up to 512GB/s of native bandwidth. But when the Infinity Cache is being fully leveraged, effective bandwidth jumps to 1,664GB/s. On the recently released Radeon RX 6700 XT, AMD scaled the Infinity Cache on its Navi 22 GPU down to 96MB.
AMD 'dimgrey_cavefish' GPU reference.
AMD / Freedesktop.orgThe internal codenames for Navi 21 (Radeon RX 6800/6900) and Navi 22 (Radeon RX 6700 XT) were “Sienna Cichlid” and “Navy Flounder”, respectively, which has some speculating that the “Dimgrey Cavefish” reference in the AMDKFD patch is an upcoming, scaled-down, more mainstream member of the RDNA 2 family, most likely Navi 23. By all accounts, Navi 23 will probably debut as a member of the Radeon RX 6600 series, but it could also be a low-power mobile Radeon GPU. AMD CEO Dr. Lisa Su teased a discrete mobile RDNA 2-based Radeon back during her CES 2021 keynote, and that GPU is due to debut soon.
How the smaller 32MB Infinity Cache affects performance remains to be seen. The larger 128MB cache on the Radeon RX 6800/6900 series helps offset the limitations of those cards’ 256-bit memory interfaces, versus NVIDIA’s wider 320-bit / 384-bit memory interfaces on the GeForce RTX 3080 and RTX 3090. The disparity in interface width on more mainstream cards probably won’t be as significant though.