Nvidia shows signs in [2023]

Status
Not open for further replies.
The most important thing to note here is the pricing of the card which is also much higher than 1650. This shows that gamers are willing to pay for a higher-end and costly graphics card if it provides value and given just how many RTX 3060s were dumped into the used market and discounted after the crypto crash, this was to be expected. Over the years, we have seen the NVIDIA 70, 60, and 50-tier cards be the most popular solutions on the market. The GeForce GTX 970 was the last 70-series card to take the top spot and that was all the way back in 2016. Since then, cards such as the GTX 1060, RTX 2060, GTX 1650/1660, and now the RTX 3060 have been the most popular options.

Even higher-end cards such as the GeForce RTX 3070 and RTX 3080 show up in the top-10 list. None of AMD's modern-day GPUs show up in the Steam Hardware Survey but NVIDIA's RTX 4090 do show up though they are definitely not going to show up at the top any time soon as they are more recent releases and will take years (like the RTX 30 series) to make their way to the top spot.

Steam-Hardware-Survey-March-2023-NVIDIA-GeForce-AMD-Radeon-GPU-Share.png
 
Those March 2023 SHS numbers look very odd.

Individual GPU numbers look funky, but I think it's more instructive to look at the Overall Distribution on top. Look at the trend for DX12 GPUs:
90.68% (Nov) -> 90.26% (Dec) -> 90.37% (Jan), 90.79% (Feb) -> 94.47% (Mar)!!

So DX12 GPUs jumped +3.68% in March, largely at the expense of -2.71% DX8 GPUs. What was so special about March 2023 that a bunch of gamers decided to upgrade their 20-year old GeForce4 Ti's to RTXs? I mean, it's a solid move, I'm sure they're going to love the upgrade*.

* -- Upgrade in technology, that is. They may be disappointed in the games. They'll go from the likes of Half Life, NOLF, KOTOR and CnC Generals to... let's see... PUBG_clone_#192 and OpenWorld_Map_Filled_With_Multi_Colored_Objective_Icons_#263. Or who knows, maybe they'll love playing Barbie with their toons with microtransaction-bought cosmetics.
 
Those March 2023 SHS numbers look very odd.

Individual GPU numbers look funky, but I think it's more instructive to look at the Overall Distribution on top. Look at the trend for DX12 GPUs:
90.68% (Nov) -> 90.26% (Dec) -> 90.37% (Jan), 90.79% (Feb) -> 94.47% (Mar)!!

So DX12 GPUs jumped +3.68% in March, largely at the expense of -2.71% DX8 GPUs. What was so special about March 2023 that a bunch of gamers decided to upgrade their 20-year old GeForce4 Ti's to RTXs? I mean, it's a solid move, I'm sure they're going to love the upgrade*.
March was curious in more ways than just that sudden jump. Windows 11 market share also nosedived, and English lost its slot as the most common language.

It's as if the Steam survey was previously blocked by the Great Firewall of China and then was (is?) whitelisted.
 
March was curious in more ways than just that sudden jump. Windows 11 market share also nosedived, and English lost its slot as the most common language.

It's as if the Steam survey was previously blocked by the Great Firewall of China and then was (is?) whitelisted.

AFAIK there's a "Chinese version" of Steam in China. The sudden increase of simplified Chinese (+25%) suggests that many people in China were starting to use "real" Steam. I'm not sure what could be the reason, the trigger could be something very trivial (e.g. some popular game in China got censored and people really want to play that game).
 
Steamdeck by chance? That seems unlikely tho. Maybe it provided the catalyst for other measurement methods to be opened up?
 
So long as they refuse to do transparent statistics, it's safe to assume they've just screwed something up again
 
As usual, NVIDIA ran all the MLPerf benchmarks, including the new networked models that feed the model data to the servers over a network instead of having the parameters already loaded into the system. The NVIDIA H100 Tensor Core GPUs delivered the highest performance in every test of AI inference. Thanks to software optimizations, the GPUs delivered up to 54% performance gains from their debut in September. Remember that NVIDIA has more software than hardware engineers for a reason.
...
The new NVIDIA L4 Tensor Core GPUs made their debut in the MLPerf tests at over 3x the speed of prior-generation T4 GPUs. Again, updated software plays a large role in performance gains over time. Available in a PCIe low-profile form factor, these accelerators ran all MLPerf workloads, consistent with NVIDIA’s belief that customers want a versatile and flexible AI platform.
...
And in the edge market where performance matters, NVIDIA is again the leader. The NVIDIA Jetson AGX Orin system-on-module delivered gained up to 63% in energy efficiency and 81% in performance compared with its results a year ago.
...
Once again, NVIDIA wins in performance, but faces increasing competition for power-constrained environments such as the Edge Data Center and the Embedded Edge,, where Qualcomm and SiMa.ai had winning results.

Notably, Amazon AWS, AMD data center GPUs like the Instinct MI250, Google TPU (which had published great results in the past), Tesla and Intel Gaudi were all no-shows, as were startups Cerebras, Graphcore, Groq and Samba Nova. Its very hard to imagine these companies aren’t interested in showing how well they run the latest models on their latest hardware. Go figure. When I was at IBM Power servers, we had an unwritten rule: don’t publish any benchmark you didn’t win. That worked well then, but in the modern world, not so much.
 

Nvidia doesn't seem interested in 2D MCM packaging. They have been filing patents for face-to-face die stacking over the past few years so it's clear that they'll be moving towards 3D stacking right away for the next generation GPUs.
 
The RTX 4070 Founders Edition sports a dual-slot design, noticeably smaller than its RTX 4080 counterpart. The unique V-shaped PCB allows the secondary axial fan to be fully exposed on either side.

NVIDIA is set to release its RTX 4070 GPU, with the Founders Edition available for purchase. Unlike the non-Ti SKU, the company is encouraging more custom designs for this model. Even with its compact size, NVIDIA has incorporated a 16-pin power connector into the design, while most custom variants will utilize a standard 8-pin connector. Gamers have the option to connect up to four displays through a single HDMI 2.1 or three DisplayPort 1.4 connectors. The RTX 4070 is a versatile addition to the desktop Ada lineup, suitable for systems with or without power adapters.

It's worth noting that the RTX 3080 debuted at $699, while the RTX 3070 Ti launched at $599, the anticipated MSRP of the RTX 4070. Bear in mind that these figures are based on DLSS 2/DLSS 3 performance; thus, it's advisable to await reviews for more accurate information on actual performance, efficiency, and cost-effectiveness. NVIDIA confirms that the GeForce RTX 4070 will be available on April 13th, priced at $599, the RTX 4070 Founders Edition is poised to challenge NVIDIA's AIC partners.
 
A 3080 for $100 cheaper 2.5 years later isn’t exciting at all. Outside of the 4090 this is the worst generation of GPUs from both vendors in many years.
It's the generation with the worst gains over the previous generation - but this is expected development, is been years since we started hearing about diminishing improvements in transistor costs. It will likely get even worse in the following years, btw.

Doesn't mean that its a bad generation per se. The cards are fine, just the perf/price gains are low.
 
  • Like
Reactions: nAo
Status
Not open for further replies.
Back
Top