A recently discarded NVIDIA graphics card prototype has emerged from an unexpected source—a dumpster. Reported by VideoCardz citing an anonymous tipster from Reddit, the new GPU is identified as an experimental RTX 4090 variant, and has very unique design features and engineering choices. The recovered prototype shares core specifications with the retail RTX 4090 Founders Edition, featuring identical PG147 SKU 330 board architecture and power specifications of 450 W typical draw with a 600 W peak capacity. However, technicalities reveal some differences, particularly in its BIOS version (95.02.01.22.01) and operating frequencies, with a 2115 MHz base clock and 2355 MHz boost clock running approximately 5% lower than the regular Founders Editions model.

Most intriguingly, the prototype implements an interesting triple-fan cooling solution, distinguished by a concealed middle fan integrated within the heatsink structure. This design element aligns with previous industry speculation about NVIDIA's development of an 900 W-capable Founders Edition card. However, the discovered unit maintains standard RTX 4090 branding without any "Ti" or "TITAN" designations, which we previously associated the massive heatsink with. Further analysis suggests the prototype's distinctive three-section PCB layout was likely developed to accommodate additional processing cores and memory components, pointing to NVIDIA's exploration of more powerful variants that ultimately never reached production.

TinyGrad, the company behind the popular TinyBox system, is aiming to commoditize the PetaFLOP. Its latest powerhouse, the TinyBox Pro, is on display on X. This high-performance system boasts an impressive 1.36 PetaFLOPS of FP16 computing power and is based on commercial GPU. The TinyBox Pro configuration features eight NVIDIA RTX 4090 GPUs, surpassing its predecessors with a combined GPU RAM of 192 GB and a memory bandwidth of 8,064 GB/s. This substantial upgrade is complemented by dual AMD Genoa processors and 384 GB of system RAM, delivering a memory bandwidth of 921.6 GB/s. What sets the TinyBox Pro apart is its enterprise-grade architecture. The system utilizes four 2000 W power supplies requiring 200 V+ input, housed in a 4U form factor that spans 31 inches in depth. Despite its compact size, weighing 88 pounds, the unit comes equipped with Supermicro rails for seamless rack integration.

Connectivity options are equally impressive, featuring two open PCIe 5.0 x16 slots that provide extensive expansion capabilities. Storage is managed through a 1 TB boot drive, though this might seem conservative compared to some competitors' offerings. The system runs on Ubuntu 22.04 and is noted for its superior driver quality (compared to commercial AMD GPUs), addressing a common pain point in high-performance computing on commercial hardware. On social media, TinyGrad was very vocal about its fight with the AMD Radeon GPU drivers. However, potential buyers should be prepared for significant noise levels during operation, a trade-off for the remarkable computing power packed into the 4U chassis. With a pre-order price tag of $40,000, the TinyBox Pro positions itself as a serious contender in the professional AI computing market, where regular GPU boxes can cost 100s of thousands of US Dollars. This pricing reflects its enterprise-grade specifications and positions it as an accessible alternative to larger, more expensive computing clusters.

Thanks to the renowned NVIDIA hardware leaker kopite7Kimi on X, we are getting information about the final versions of NVIDIA's first upcoming wave of GeForce RTX 50 series "Blackwell" graphics cards. The two leaked GPUs are the GeForce RTX 5090 and RTX 5080, which now feature a more significant gap between xx80 and xx90 SKUs. For starters, we have the highest-end GeForce RTX 5090. NVIDIA has decided to use the GB202-300-A1 die and enabled 21,760 FP32 CUDA cores on this top-end model. Accompanying the massive 170 SM GPU configuration, the RTX 5090 has 32 GB of GDDR7 memory on a 512-bit bus, with each GDDR7 die running at 28 Gbps. This translates to 1,568 GB/s memory bandwidth. All of this is confined to a 600 W TGP.

When it comes to the GeForce RTX 5080, NVIDIA has decided to further separate its xx80 and xx90 SKUs. The RTX 5080 has 10,752 FP32 CUDA cores paired with 16 GB of GDDR7 memory on a 256-bit bus. With GDDR7 running at 28 Gbps, the memory bandwidth is also halved at 784 GB/s. This SKU uses a GB203-400-A1 die, which is designed to run within a 400 W TGP power envelope. For reference, the RTX 4090 has 68% more CUDA cores than the RTX 4080. The rumored RTX 5090 has around 102% more CUDA cores than the rumored RTX 5080, which means that NVIDIA is separating its top SKUs even more. We are curious to see at what price point NVIDIA places its upcoming GPUs so that we can compare generational updates and the difference between xx80 and xx90 models and their widened gaps.

EK, the leading manufacturer of premium liquid cooling gear, is doing a mid-life upgrade of its Quantum Vector² water blocks by releasing five new products. These new EVO version water blocks are compatible with Strix and TUF 4080 and 4090 series graphics cards. The new Vector² EVO water blocks efficiently tackle the excessive heat generated during high loads on the GPU. These water blocks, an evolution of the Quantum Vector² GPU water blocks from EK's premium Quantum Line, feature a 3rd-generation Vector cooling engine that has been tweaked to maximize cooling power. They cool not only the GPU die, but also the voltage regulation module, voltage controllers, and VRAM, providing comprehensive thermal management.

The main upgrades of the EVO water blocks include:

• A new stand-out end-piece design to shorten the water block
• A custom-made inner O-ring that prevents coolant from seeping between the channels and improves flow across the entire water block, focusing more coolant through the cooling fins
• New, high-performance pre-cut thermal pads
• A better-performing thermal paste