#TFLOPS
LightNews LightNews
tubazef.com
TFLOPS tá ultrapassado

Vamos medir processamento em VEPS agora (velha empacotando pamonha)
January 23, 2026 at 5:12 PM
necro.universonintendo.com
Levando em conta que isso aqui seja verídico (NÃO SE SABE), tomamos como base os CUDA Cores para "potência bruta", sendo assim:

• Steam Deck // 1.6 TFLOPs
• Switch 2 // 3.09 TFLOPs*
• Xbox Series S // 4.0 TFLOPs
• PlayStation 4 Pro // 4.2 TFLOPs

*Ignorando Tensor Cores
April 4, 2025 at 11:47 AM
nividhia.bsky.social
Como dato anecdótico para que pensemos como avanzan las cosas, en Terminator 3 (2003), Skynet, el superordenador, procesa a 60 Teraflops por segundo. Una gráfica medio buena de hoy ya va a 80 TFLOPS y el supercomputador mas potente conocido chuta a mas de dos millones de TFLOPS.
September 26, 2026 at 3:34 PM
kontronn.bsky.social
Xbox Helix Reportedly Hits 56 TFLOPS, PS6 at 40 TFLOPS www.kad8.com/ai/xbox-heli...
Xbox Helix Reportedly Hits 56 TFLOPS, PS6 at 40 TFLOPS
Xbox Project Helix is reportedly rated at 56 TFLOPS versus 40 TFLOPS for PS6, as both next-gen console chips are said to have completed tape-out.
www.kad8.com
September 25, 2026 at 2:47 AM
gammitin.bsky.social
GigaIO Gryf - portable AI Supercomputer:
- 91.6 TFLOPS (FP32)
- 733 TFLOPS (FP16)
- 1,466 TFLOPS (FP8)
- AMD EPYC 7313
- 256GB RAM
- Up to 4x GPUs (e.g. Nvidia L40S)
- 246TB NVMe storage
- 100GbE + 25GbE networking
- 2.5kW PSU, rugged case
- <55 lbs, fits in carry-on
GigaIO Gryf - portable AI Supercomputer
May 18, 2025 at 6:47 PM
videocardz.com
PS6 and next-gen Xbox chips reportedly taped out, Kepler claims 40 TFLOPS for PS6 and 56 for Helix
PS6 and next-gen Xbox chips reportedly taped out, Kepler claims 40 TFLOPS for PS6 and 56 for Helix - VideoCardz.com
KeplerL2 says Sony and Microsoft have completed tape-out for the chips powering their next-generation consoles. GPU estimates reach 40 and 56 TFLOPS.
videocardz.com
September 23, 2026 at 12:25 PM
keytryer.net
It also has some gems like "ban models using more than 10^12 TFlops" which means banning... GPT-4
September 5, 2025 at 6:57 PM
nineverything.bsky.social
Rumor: Nintendo Switch 2 TFLOPS and clock speed tech specs leaked https://nintendoeverything.com/nintendo-switch-2-tflops-tech-specs-leak/
Nintendo Switch 2 TFLOPS tech specs leak
January 15, 2025 at 12:08 AM
armscontrolwonk.bsky.social
One thing I recently updated: US supercomputing zoomed past the benchmark established for stockpile stewardship (100 tflops) and also China has too. The Chinese have the computers to design new weapons, but what they don't have is the data from nuclear testing -- yet.
October 30, 2025 at 3:32 PM
techpowerup.com
Xbox Helix Reportedly Hits 56 TFLOPS, 40% Ahead of PS6 as Both Chips Said to Be Taped Out
www.techpowerup.com
September 23, 2026 at 5:10 PM
deeb.bsky.social
Steam Machine:
“6X as powerful as Deck” (1.6 TFlops): 9.6 TFlops?
PS5: 10.3 TFlops

Steam Machine: 6 Core Zen 4 CPU @4.8ghz PS5: 8 Core Zen 2 @3.5ghz

Steam Machine: RDNA 3 28CU @ 2.45ghz
PS5: RDNA 2 36CU @ 2.23ghz

Safe to say Steam is targeting PS5 level performance.

Steam Machine: $ ??
PS5: $499
November 12, 2025 at 7:01 PM
sungkim.bsky.social
FlashAttention-4

I hope it is not pain to work with. It changes the algorithm & pipeline so that softmax & SMEM bandwidth no longer dictate speed. Attn reaches ~1600 TFLOPs, pretty much at matmul speed!
March 5, 2026 at 6:47 PM
sungkim.bsky.social
You can try out the newest and greatest Google TPU:

v6e-1 (Trillium) TPUs!! 2x the high bandwidth memory as v5e-1 (32GB) and a whopping peak rating of 918 BF16 TFLOPS (nearly 3x A100)!

on Google Colab.
May 12, 2025 at 12:35 AM
michalblaha.cz
Apple prezentuje MacMini M4 jako ideální stroj na lokální spouštění AI úloh. A do dnešního ohlášení NVIDIA Project Digits to platilo.

Project Digits: 128GB @ 512GB/s, 250 TFLOPS, $3,000
M4 Pro Mac Mini: 64GB @ 273GB/s, 17 TFLOPS, $2,200
M4 Max MacBook Pro: 128GB @ 546GB/s, 34 TFLOPS, $4,700
January 7, 2025 at 9:51 PM
dfeldman.org
GPUs depreciate very rapidly because performance is increasing so fast

In FP16 TFLOPs, which is what matters for LLMs, the best server GPU from 5 years ago is about 5% of the speed of the best today
edzitron.com Ed Zitron @edzitron.com · Nov 24
The tech industry's obsession with GPUs has led every hyperscaler to take on masses of debt, and Oracle, a previously-stable company, has now existentially tied itself to building data centers for OpenAI, a company that burns billions a year. It's all precarious.
www.wheresyoured.at/the-haters-g...
Big Tech Is Having To Take On Debt To Keep Buying GPUs, And It’s Threatening Oracle’s Future Yes, it’s true that the biggest hyperscalers in the world are, incidentally, NVIDIA’s biggest customers. But here’s the thing — that doesn’t actually mean anything, especially when we consider whether this capex spending spree continues into the future, or even whether these companies can service their insane ongoing obligations. There are a few points to note: We’re reaching a point where these companies cannot meet their obligations simply by using their cashflow. In the case of Oracle and Meta, they have to borrow money. Even those companies that can pay their own way are going to turn to the debt markets, in part because financing these big, ambitious projects just makes sense. They can spread the cost over a prolonged timeframe, rather than paying for everything upfront. Oracle’s most recent AI-related debt issue, for example, had bonds that expired in 2065. A recent JP Morgan document said that the company expects the next wave of AI investment to be financed by credit, and that “if the hyperscalers decided to lever their balance sheets to 2.8x net debt to EBITDA (the median for an investment grade company), it could result in an additional $1 trillion of capital to spend.” So far, the five hyperscalers — Meta, Google, Oracle, Microsoft, and Amazon — have issued $121 billion in debt, with around 20% of that (or $27 billion) going to one campus for Meta in Richland Parish, Louisiana. We can expect another $100bn in debt to be raised next year. To be clear, this only applies to traditional long-term debt, specifically bonds. It does not include operating and finance leases, which are another matter entirely, even though they also represent long-term obligations for those companies. Naturally, investors are starting to get worried. Even before it jumped on the AI bandwagon, Oracle was already an incredibly indebted company — and it continues to raise even more d… So far, it’s the most indebted company with an investment-grade credit rating. If it lost that investment-grade rating, it wouldn’t be able to get nearly the same amount of capital, or under the same terms. To be clear, this doesn’t just apply to Oracle, but any company that raises debt on the public markets. That said, in the case of Oracle, it’s already rated BBB- by Standard and Poors, which is at the bottom-end of what we consider “investment grade.” S&P also said it has a negative outlook for Oracle, citing its dependence on OpenAI for as much as a third of its future revenue, which, in turn, is contingent upon OpenAI becoming a sustainable business (ha!) or drawing in more investment capital. These sentiments were echoed by Moody’s, another rating agency, which also puts Oracle at the bottom end of the investment-grade scale. The problem with AI capex spending versus, say, any other capex spending is that it’s a.) expensive, and b.) based on the belief that generative AI will be this big, transformational, money-minting thing. As Reuters noted, some money managers are raising the alarm that the current wave of AI-related debt raising doesn’t properly reflect the risks posed by generative AI. Put another way, it’s like a bank gave a loan to someone with a good credit rating so that they could go to Vegas and play roulette. The intent behind the debt raise matters almost as much as the company raising it. If the terms of these bonds reflected the risks posed by generative AI, the debt would almost certainly be more expensive. Right now, we’re seeing the emergence of instruments that would allow investors to bet against the companies raising money for AI capex. There are now credit default swaps — essentially, insurance products that pay out if a bond fails — for AI bonds. The way that these credit default swaps are priced is in the likelihood of failure over five years, with CDS contracts for riskier bonds costing more — just like house insurance in a… For Oracle, the price of insuring against a bond failure has reached its highest level in nearly two years, representing both demand from investors, as well as a souring of sentiment. The point I’m trying to make is this: NVIDIA’s business is dominated by a handful of companies who are increasingly (though by varying degrees) burdened by debt, and reliant upon debt. For this to continue, people need to be willing to lend these companies money. No money, no GPU sales. Allow me to put this in really stark terms, because it needs to be said. NVIDIA has restructured its business — and its status as the world’s most valuable company — around the fates of a handful of companies, who are themselves servicing two companies, namely OpenAI and (to a lesser extent) Anthropic. Those customers are racking up long-term commitments, both in terms of operating and finance leases, but also in terms of debt. They’re racking up debt at a scale that is unprecedented for these companies, and the evidence suggests that they’ll continue to raise money through debt in the near-term, with much of that going to NVIDIA in exchange for GPUs. For these companies to continue to be able to raise debt, they’re reliant on a handful of factors: The broader macroeconomic picture. Central bank interest rates, and so on. The individual credit ratings of the companies raising debt. The availability of capital in the bond market. The broader sentiment towards these companies. The broader sentiment towards generative AI as a driver of revenue (and profitability). If any of these go in the wrong direction, raising debt becomes harder and more expensive. Additionally, it’s not hard to imagine a scenario where sentiment towards these companies shifts. Oracle, for example, doesn’t really do all that much other than squeeze its legacy customers for as much as it can, for as long as it can.
December 7, 2025 at 10:08 PM
alexstrook.bsky.social
ive got some insider info about nintendo switch 2, and everyone is wrong

-name is actually "2 Nintendo Switch 2" (each 2 represents a joycon)
-Once you dock the console, you cannot remove it. one-time-only docking
-Each joycon has 18.2 TFLOPS (more than ps5 pro)
December 29, 2024 at 9:12 PM
aelkus.bsky.social
“you’re not just getting deep, zesty flavor….but also a generous serving of TFLOPs and clock cycles”
May 29, 2026 at 6:42 AM
mamemomonga.bsky.social
Switch2ってメモリ16GB ストレージ256GBって、それだけみるとMacBookNeoよりもスペックがいいんだよなぁ・・・。ちなみにSwitch2搭載のSoCであるnvidia T239は、ドックモードで3TFLOPSの演算能力がある。これはM1の2.6TFLOPSとM2の3.6TFLOPSの間くらい。GPUはM2~M3くらいの演算能力がある。少なくともMacBook Neoよりも高性能である。Switch2をコンピュータとして見ると激安である。ただコンピュータとして使えるOSは存在しないが。
May 8, 2026 at 11:30 AM
heartade.dev
엑시노스 2100의 GPU 깡성능(1.3TFLOPS)은 닌텐도 스위치(0.4TFLOPS)의 3배가 넘는데 원신 같은 게임들이 버벅거리는 걸 보면 역시 기계가 아니라 개발자를 탓해야 함

며칠전에 나온 엑시 2500은 4TFLOPS를 넘겨서 엑박 시리즈S랑 맞짱뜰 수 있는 수준인데 기껏해야 모바게 주제에 콘솔 현세대기 성능으로도 버벅거리는 거면 반성해야...
heartade.dev 🏳️‍⚧️ heartade | 하세 @heartade.dev · Jul 31
개인적으로는 최신폰을 더 많이 팔기 위해 게임 개발자들이 최적화를 안 하고 놀게 만드는 어둠의 세력이 있다는 음모론을 밀고 있어요(...
July 31, 2025 at 3:36 PM
tosh1125.bsky.social
前のリーク（80 TFLOPS・RTX 5080超え）から後退してる〜。

RTX 5070 Ti：約43.9 TFLOPS
PS6リーク：約40 TFLOPS

レイトレとAI頑張るっぽいけど、DLSSもどんどん進化してるしな〜。

https://t.co/qMyZT6QJQ8
September 27, 2026 at 7:35 AM
gordon.bsky.social
at what point does the global reserve currency transition to TFLOPS?
February 18, 2026 at 2:35 PM
apenwarr.ca
Avery's Adjusted Real Income calculator

iPhone 16: 2.3 Tflops, about $1000
Fastest supercomputer in 2001: 12.3 Tflops, about $110M

For $100k/year (2025 dollars) you could buy 100 iPhones, 230 Tflops, valued at about $2B (2001 dollars)

Tell me again about inflation
July 11, 2025 at 3:17 PM
tweaktown.com
Leaked estimates suggest Xbox Helix's chips have taped out and its GPU could reach 56 TFLOPS versus 40 for the PS6, putting Microsoft about 40 percent ahead on paper. 🎮🧩
Leaked TFLOPS numbers suggest Xbox Helix will outpower the PS6 by 40 percent
Leaked estimates suggest Xbox Helix's chips have taped out and its GPU could reach 56 TFLOPS versus 40 for the PS6, putting Microsoft about 40 percent ahead on paper. 🎮🧩
tweak.town
September 23, 2026 at 10:03 PM
insidexbox.de
XBOX Helix mit 56 TFLOPS, PS6 mit 40 TFLOPS. Ein neuer Leak nennt erste GPU Compute Werte für beide Konsolen. 🎮🔍

➡️ www.insidexbox.de/news/xbox-he...
__________
#Xbox #InsideXboxDE #PS6 #XboxHelix
XBOX Helix: Leak nennt 56 TFLOPS für die GPU
XBOX Helix mit 56 TFLOPS, PS6 mit 40 TFLOPS. Ein neuer Leak nennt erste GPU Compute Werte für beide Konsolen. 🎮🔍
www.insidexbox.de
September 25, 2026 at 6:01 AM