With NVIDIA’s Turing architecture turning six years old this year, the company has been retiring many of the remaining Turing products from its video card lineup. And today that spirit of spring cleaning is coming to the entry-level segment of NVIDIA’s professional visualization lineup, where NVIDIA is introducing a pair of new desktop cards based on their low-end Ampere hardware.

The new RTX A1000 and RTX A400 cards will be replacing the T1000/T600/T400 lineup, which was released three years ago in 2021. The new cards slot into the same entry-level category and finally finish fleshing out the RTX A series of proviz cards, offering NVIDIA’s Ampere-generation professional graphics technologies in the lowest-power, lowest-performance, lowest-cost configuration possible.

Notably, since the entry-level T-series were based on NVIDIA’s feature-limited TU11x silicon, which lacked ray tracing and tensor core support – the basis of NVIDIA’s RTX technologies and associated branding – this marks the first time these technologies will be available in NVIDIA’s entry-level desktop proviz cards. And accordingly, these are being promoted to RTX-branded video cards, ending the odd overlap with NVIDIA’s compute cards, which never carry RTX branding.

It goes without saying that as low-end cards, the ray tracing performance of either part is nothing to write home about, but it gives NVIDIA’s current proviz lineup a consistent set of graphics features from top to bottom.

Both the A1000 and A400 are based on the same board design, with NVIDIA doing away with any pretense of physical feature differentiation this time around (T400 was missing its 4^th Mini DisplayPort). This means both cards are based on the GA107 GPU, sporting different core and memory configurations.

RTX A1000 is a not-quite-complete configuration of GA107, with 2304 CUDA cores and 72 tensor cores. This is paired with 8GB of GDDR6, which runs at 12Gbps, for a total of 192GB/second of memory bandwidth. The TDP of the card is 50 Watts, matching its predecessor.

Meanwhile RTX A400 is far more cut down, offering about a third of the active hardware on the GPU itself, and half the memory bandwidth. On paper this gives it around 40% of T1000’s performance, and half the memory bandwidth – or 96GB/second. Notably, despite the hardware cut-down, the official TDP is still 50 Watts, versus the 30 Watts of its predecessor. So at this point NVIDIA will soon cease offering a desktop proviz card lower than 50 Watts.

As noted before, both cards otherwise feature the same physical design, with a half-height half-length (HHHL) board with active cooling. As you’d expect from such low-TDP cards, these are single-slot cooler designs. Both cards feature a quartet of Mini DisplayPorts, with the same DP 1.4a functionality that we’ve seen across all of NVIDIA’s products for the last several years.

Finally, video-focused users will want to make note that the A1000/A400 have slightly different video capabilities. While A1000 gets access to both of GA107’s NVDEC video decode blocks, A400 only gets access to a single block – one more cutback to differentiate the two cards. Otherwise, both video cards get access to the GPU’s sole NVENC block.

According to NVIDIA, the RTX A1000 will be available starting today through its distribution partners. Meanwhile the RTX A400 will hit distribution channels in May, and with OEMs expected to begin offering the cards as part of their pre-built systems this summer.

Intel this morning is kicking off the second day of their Vision 2024 conference, the company’s annual closed-door business and customer-focused get-together. While Vision is not typically a hotbed for new silicon announcements from Intel – that’s more of an Innovation thing in the fall – attendees of this year’s show are not coming away empty handed. With a heavy focus on AI going on across the industry, Intel is using this year’s event to formally introduce the Gaudi 3 accelerator, the next-generation of Gaudi high-performance AI accelerators from Intel’s Habana Labs subsidiary.

The latest iteration of Gaudi will be launching in the third quarter of 2024, and Intel is already shipping samples to customers now. The hardware itself is something of a mixed bag in some respects (more on that in a second), but with 1835 TFLOPS of FP8 compute throughput, Intel believes it’s going to be more than enough to carve off a piece of the expansive (and expensive) AI market for themselves. Based on their internal benchmarks, the company expects to be able beat NVIDIA’s flagship Hx00 Hopper architecture accelerators in at least some critical large language models, which will open the door to Intel grabbing a larger piece of the AI accelerator market at a critical time in the industry, and a moment when there simply isn’t enough NVIDIA hardware to go around.

Introspect this week introduced its M5512 GDDR7 memory test system, which is designed for testing GDDR7 memory controllers, physical interface, and GDDR7 SGRAM chips. The tool will enable memory and processor manufacturers to verify that their products perform as specified by the standard.

One of the crucial phases of a processor design bring up is testing its standard interfaces, such as PCIe, DisplayPort, or GDDR is to ensure that they behave as specified both logically and electrically and achieve designated performance. Introspect's M5512 GDDR7 memory test system is designed to do just that: test new GDDR7 memory devices, troubleshoot protocol issues, assess signal integrity, and conduct comprehensive memory read/write stress tests.

The product will be quite useful for designers of GPUs/SoCs, graphics cards, PCs, network equipment and memory chips, which will speed up development of actual products that rely on GDDR7 memory. For now, GPU and SoC designers as well as memory makers use highly-custom setups consisting of many tools to characterize signal integrity as well as conduct detailed memory read/write functional stress testing, which are important things at this phase of development. But usage of a single tool greatly speeds up all the processes and gives a more comprehensive picture to specialists.

The M5512 GDDR7 Memory Test System is a desktop testing and measurement device that is equippped with 72 pins capable of functioning at up to 40 Gbps in PAM3 mode, as well as offering a virtual GDDR7 memory controller. The device features bidirectional circuitry for executing read and write operations, and every pin is equipped with an extensive range of analog characterization features, such as skew injection with femto-second resolution, voltage control with millivolt resolution, programmable jitter injection, and various eye margining features critical for AC characterization and conformance testing. Furthermore, the system integrates device power supplies with precise power sequencing and ramping controls, providing a comprehensive solution for both AC characterization and memory functional stress testing on any GDDR7 device.

Introspects M5512 has been designed in close collaboration with JEDEC members working on the GDDR7 specification, so it promises to meet all of their requirements for compliance testing. Notably, however, the device does not eliminate need for interoperability tests and still requires companies to develop their own test algorithms, but it's still a significant tool for bootstrapping device development and getting it to the point where chips can begin interop testing.

“In its quest to support the industry on GDDR7 deployment, Introspect Technology has worked tirelessly in the last few years with JEDEC members to develop the M5512 GDDR7 Memory Test System,” said Dr. Mohamed Hafed, CEO at Introspect Technology.

AMD's FidelityFX Super Resolution 3 technology package introduced a plethora of enhancements to the FSR technology on Radeon RX 6000 and 7000-series graphics cards last September. But perfection has no limits, so this week, the company is rolling out its FSR 3.1 technology, which improves upscaling quality, decouples frame generation from AMD's upscaling, and makes it easier for developers to work with FSR.

Arguably, AMD's FSR 3.1's primary enhancement is its improved temporal upscaling image quality: compared to FSR 2.2, the image flickers less at rest and no longer ghosts when in movement. This is a significant improvement, as flickering and ghosting artifacts are particularly annoying. Meanwhile, FSR 3.1 has to be implemented by the game developer itself, and the first title to support this new technology sometime later this year is Ratchet & Clank: Rift Apart.

Another significant development brought by FSR 3.1 is its decoupling from the Frame Generation feature introduced by FSR 3. This capability relies on a form of AMD's Fluid Motion Frames (AFMF) optical flow interpolation. It uses temporal game data like motion vectors to add an additional frame between existing ones. This ability can lead to a performance boost of up to two times in compatible games, but it was initially tied to FSR 3 upscaling, which is a limitation. Starting from FSR 3.1, it will work with other upscaling methods, though AMD refrains from saying which methods and on which hardware for now. Also, the company does not disclose when it is expected to be implemented by game developers.

In addition, AMD is bringing support for FSR3 to Vulkan and Xbox Game Development Kit, enabling game developers on these platforms to use it. It also adds FSR 3.1 to the FidelityFX API, which simplifies debugging and enables forward compatibility with updated versions of FSR. 

Upon its release in September 2023, AMD FSR 3 was initially supported by two titles, Forspoken and Immortals of Aveum, with ten more games poised to join them back then. Fast forward to six months later, the lineup has expanded to an impressive roster of 40 games either currently supporting or set to incorporate FSR 3 shortly. As of March 2024, FSR is supported by games like Avatar: Frontiers of Pandora, Starfield, The Last of Us Part I. Shortly, Cyberpunk 2077, Dying Light 2 Stay Human, Frostpunk 2, and Ratchet & Clank: Rift Apart will support FSR shortly.

Source: AMD

Already solidly in the driver’s seat of the generative AI accelerator market at this time, NVIDIA has long made it clear that the company isn’t about to slow down and check out the view. Instead, NVIDIA intends to continue iterating along its multi-generational product roadmap for GPUs and accelerators, to leverage its early advantage and stay ahead of its ever-growing coterie of competitors in the accelerator market. So while NVIDIA’s ridiculously popular H100/H200/GH200 series of accelerators are already the hottest ticket in Silicon Valley, it’s already time to talk about the next generation accelerator architecture to feed NVIDIA’s AI ambitions: Blackwell.

Asus this week has become the latest PC video card manufacturer to announce a sub-75W video card based on NVIDIA's recently-released low-power GeForce RTX 3050 6GB design. And going one step further for small form factor PC owners, Asus has used NVIDIA's low-power GPU configuration to produce a half-height video card that can fit into low-profile systems.

As Asus puts it, the GeForce RTX 3050 LP BRK 6GB GDDR6 is a 'big productivity in a small package' and for a low-profile dual-slot graphics board, it indeed is. The unit has three display outputs, including a DVI-D, HDMI 2.1, and DisplayPort 1.4a with HDCP 2.3 support, which makes the graphics card s viable option both for a a dual-display desktop and a home theater PC (Nvidia's GA107 graphics processor supports all popular codecs except AV1). Furthermore, a DVI-D output enables the card to drive outdated displays, which even over half a decade after DVI-D was retired, still hang around as spare parts. Meanwhile, because the card only consumes around 70W, it does not require any auxiliary PCIe power connectors, which are at times not available in cheap systems from big PC makers.

Underlying this card is the aforementioned GeForce RTX 3050 6 GB, which uses the GA107 GPU with 2304 CUDA cores, and it comes with 6GB of GDDR6 memory connected to a narrower 96-bit memory bus (down from 128-bits for the full 8GB version. With a lower boost clock of 1470 MHz (1500 MHz in OC mode), the RTX 3050 6GB has reduced computing performance, delivering 6.77 FP32 TFLOPS versus 9.1 FP32 TFLOPS of the full-fledged RTX 3050.

As a result, the low-profile GeForce RTX 3050 6 GB is very much an entry-level card, though the low power requirements for such a card are also what make it special. This should be plenty for low-end gaming – beating out integrated GPUs – though suffice it to say, it's not going to compete with high-end, power-hungry cards either.

With its diminutive size, the Asus GeForce RTX 3050 LP BRK 6 GB GDDR6 looks to be a nice candidate for upgrading cheap systems from OEMs as well as fixing outdated PCs. What remains to be seen is how price competitive it is going to be. The graphics board already has one low-profile rival from MSI — which costs $185 — so Asus is not the only vendor competing here.

JEDEC on Tuesday published the official specifications for GDDR7 DRAM, the latest iteration of the long-standing memory standard for graphics cards and other GPU-powered devices. The newest generation of GDDR brings a combination of memory capacity and memory bandwidth gains, with the later being driven primarily by the switch to PAM3 signaling on the memory bus. The latest graphics RAM standard also boosts the number of channels per DRAM chip, adds new interface training patterns, and brings in on-die ECC to maintain the effective reliability of the memory.

“JESD239 GDDR7 marks a substantial advancement in high-speed memory design,” said Mian Quddus, JEDEC Board of Directors Chairman. “With the shift to PAM3 signaling, the memory industry has a new path to extend the performance of GDDR devices and drive the ongoing evolution of graphics and various high-performance applications.”

GDDR7 has been in development for a few years now, with JEDEC members making the first disclosures around the memory technology about a year ago, when Cadence revealed the use of PAM3 encoding as part of their validation tools. Since then we've heard from multiple memory manufacturers that we should expect the final version of the memory to launch in 2024, with JEDEC's announcement essentially coming right on schedule.

As previously revealed, the biggest technical change with GDDR7 comes with the switch from two-bit non-return-to-zero (NRZ) encoding on the memory bus to three-bit pulse amplitude modulating (PAM3) encoding. This change allows GDDR7 to transmit 3 bits over two cycles, 50% more data than GDDR6 operating at an identical clockspeed. As a result, GDDR7 can support higher overall data transfer rates, the critical component to making each generation of GDDR successively faster than its predecessor.

The first generation of GDDR7 is expected to run at data rates around 32 Gbps per pin, and memory manufacturers have previously talked about rates up to 36 Gbps/pin as being easily attainable. However the GDDR7 standard itself leaves room for even higher data rates – up to 48 Gbps/pin – with JEDEC going so far as touting GDDR7 memory chips "reaching up to 192 GB/s [32b @ 48Gbps] per device" in their press release. Notably, this is a significantly higher increase in bandwidth than what PAM3 signaling brings on its own, which means there are multiple levels of enhancements within GDDR7's design.

Digging deeper into the specification, JEDEC has also once again subdivided a single 32-bit GDDR memory chip into a larger number of channels. Whereas GDDR6 offered two 16-bit channels, GDDR7 expands this to four 8-bit channels. The distinction is somewhat arbitrary from an end-user's point of view – it's still a 32-bit chip operating at 32Gbps/pin regardless – but it has a great deal of impact on how the chip works internally. Especially as JEDEC has kept the 256-bit per channel prefetch of GDDR5 and GDDR6, making GDDR7 a 32n prefetch design.

GDDR Channel Architecture. Original GDDR6-era Diagram Courtesy Micron

The net impact of all of this is that, by halving the channel width but keeping the prefetch size the same, JEDEC has effectively doubled the amount of data that is prefetched per cycle of the DRAM cells. This is a pretty standard trick to extend the bandwidth of DRAM memory, and is essentially the same thing JEDEC did with GDDR6 in 2018. But it serves as a reminder that DRAM cells are still very slow (on the order of hundreds of MHz) and aren't getting any faster. So the only way to feed faster memory buses is by fetching ever-larger amounts of data in a single go.

The change in the number of channels per memory chip also has a minor impact on how multi-channel "clamshell" mode works for higher capacity memory configurations. Whereas GDDR6 accessed a single memory channel from each chip in a clamshell configuration, GDDR7 will access two channels – what JEDEC is calling two-channel mode. Specifically, this mode reads channels A and C from each chip. It is effectively identical to how clamshell mode behaved with GDDR6, and it means that while clamshell configurations remain supported in this latest generation of memory, there aren't any other tricks being employed to improve memory capacity beyond ever-increasing memory chip densities.

On that note, the GDDR7 standard officially adds support for 64Gbit DRAM devices, twice the 32Gbit max capacity of GDDR6/GDDR6X. Non-power-of-two capacities continue to be supported as well, allowing for 24Gbit and 48Gbit chips. Support for larger memory chips further pushes the maximum memory capacity of a theoretical high-end video card with a 384-bit memory bus to as high as 192GB of memory – a development that would no doubt be welcomed by datacenter operators in the era of large language AI models. With that said, however, we're still regularly seeing 16Gbit memory chips used on today's memory cards, even though GDDR6 supports 32Gbit chips. Coupled with the fact that Samsung and Micron have already disclosed that their first generation of GDDR7 chips will also top out at 16Gbit/24Gbit respectively, it's safe to say that 64Gbit chips are pretty far off in the future right now (so don't sell off your 48GB cards quite yet).

For their latest generation of memory technology, JEDEC is also including several new-to-GDDR memory reliability features. Most notably, on-die ECC capabilities, similar to what we saw with the introduction of DDR5. And while we haven't been able to get an official comment from JEDEC on why they've opted to include ECC support now, its inclusion is not surprising given the reliability requirements for DDR5. In short, as memory chip densities have increased, it has become increasingly hard to yield a "perfect" die with no flaws; so adding on-chip ECC allows memory manufacturers to keep their chips operating reliably in the face of unavoidable errors.

This figure is reproduced, with permission, from JEDEC document JESD239, figure 124

Internally, the GDDR7 spec requires a minimum of 16 bits of parity data per 256 bits of user data (6.25%), with JEDEC giving an example implementation of a 9-bit single error correcting code (SEC) plus a 7-bit cyclic redundancy check (CRC). Overall, GDDR7 on-die ECC should be able to correct 100% of 1-bit errors, and detect 100% of 2-bit errors – falling to 99.3% in the rare case of 3-bit errors. Information about memory errors is also made available to the memory controller, via what JEDEC terms their on-die ECC transparency protocol. And while technically separate from ECC itself, GDDR7 also throws in another memory reliability feature with command address parity with command blocking (CAPARBLK), which is intended to improve the integrity of the command address bus.

Otherwise, while the inclusion of on-die ECC isn't likely to have any more of an impact on consumer video cards than its inclusion had for DDR5 memory and consumer platforms there, it remains to be seen what this will mean for workstation and server video cards. The vendors there have used soft ECC on top of unprotected memory for several generations now; presumably this will remain the case for GDDR7 cards as well, but the regular use of soft ECC makes things a lot more flexible than in the CPU space.

This figure is reproduced, with permission, from JEDEC document JESD239, figure 152

Finally, GDDR7 is also introducing a suite of other reliability-related features, primarily related to helping PAM3 operation. This includes core independent LFSR (linear-feedback shift register) training patterns with eye masking and error counters. LFSR training patterns are used to test and adjust the interface (to ensure efficiency), eye masking evaluates signal quality, and error counters track the number of errors during training.

Technical matters aside, this week's announcement includes statements of support from all of the usual players on both sides of the isle, including AMD and NVIDA, and the Micron/Samsung/SKhynix trifecta. It goes without saying that all parties are keen to getting to use or sell GDDR7 respectively, given the memory capacity and bandwidth improvements it will bring – and especially in this era where anything aimed at the AI market is selling like hotcakes.

No specific products are being announced at this time, but with Samsung and Micron having previously announced their intentions to ship GDDR7 memory this year, we should see new memory (and new GPUs to pair it with) later this year.

JEDEC standards and publications are copyrighted by the JEDEC Solid State Technology Association.  All rights reserved.

NVIDIA today is quietly launching a new entry-level graphics card for the retail market, the GeForce RTX 3050 6GB. Based on a cut-down version of their budget Ampere-architecture GA107 GPU, the new card brings what was previously an OEM-only product to the retail market. Besides adding another part to NVIDIA's deep product stack, the launch of the RTX 3050 6GB also comes with another perk: lower power consumption thanks to this part targeting system installs where an external PCIe power connector would not be needed. NVIDIA's partners, in turn, have not wasted any time in taking advantage of this, and today Palit is releasing its first fanless KalmX board in years: the GeForce RTX 3050 KalmX 6GB.

The GeForce RTX 3050 6GB is based on the GA107 graphics processor with 2304 CUDA cores, which is paired with 6GB of GDDR6 attached to a petite 96-bit memory bus (versus 128-bit for the full RTX 3050 8GB). Coupled with a boost clock rating of just 1470 MHz, the RTX 3050 6GB delivers tangibly lower compute performance than the fully-fledged RTX 3050 — 6.77 FP32 TFLOPS vs 9.1 FP32 TFLOPS — but these compromises offer an indisputable advantage: a 70W power target.

Palit is the first company that takes advantage of this reduced power consumption of the GeForce RTX 3050 6 GB, as the company has launched a passively cooled graphics card based on this part, the first in four years. The Palit GeForce RTX 3050 KalmX 6GB (NE63050018JE-1170H) uses a custom printed circuit board (PCB) that not only offers modern DisplayPort 1.4a and HDMI 2.1 outputs, but, as we still see in some entry-level cards, a dual-link DVI-D connector (a first for an Ampere-based graphics card).

The dual-slot passive cooling system with two heat pipes is certainly the main selling point of Palit's GeForce RTX 3050 KalmX 6GB. The product is pretty large though — it measures 166.3×137×38.3 mm — and will not fit into tiny desktops. Still, given the fact that fanless systems are usually not the most compact ones, this may not be a significant limitation of the new KalmX device.

Another advantage of Palit's GeForce RTX 3050 KalmX 6GB in particular and NVIDIA's GeForce RTX 3050 6GB in general is that it can be powered entirely via a PCIe slot, which eliminates the need for an auxiliary PCIe power connectors (which are sometimes not present in cheap systems from big OEMs).

Wccftech reports that NVIDIA's GeForce RTX 3050 6GB graphics cards will carry a recommended price tag of $169 and indeed these cards are available for $170 - $180. This looks to be a quite competitive price point as the product offers higher compute performance than that of AMD's Radeon RX 6400 ($125) and Radeon RX 6500 XT ($140). Meanwhile, it remains to be seen how much will Palit charge for its uniquely positioned GeForce RTX 3050 KalmX 6GB.