Silicon Graphics Unveils IRIS 2400 to Eliminate Workstation Latency

VintageCG

Citing IBM research on sub-half-second response times, Silicon Graphics argues real-time interactive rendering will dictate market winners because engineers will refuse to tolerate slower workstations.

AI-assisted summary of SGI IRIS 2400 (1985) — original source. Check the original for full context.

Source published: 2010-03-07

Key ideas

Silicon Graphics demonstrated the IRIS 2400 workstation executing real-time 3D graphics captured directly from monitor output to 1-inch videotape without video editing tricks or precomputed frame buffering. (0:00)
Citing formal IBM research, the presentation argues that reducing display latency below 0.5 seconds triggers an exponential rise in engineering productivity and turnaround improvements exceeding 50% to 60%. (0:00)
The IRIS 2400 achieves real-time 3D performance using a hardware pipeline of 12 custom VLSI Geometry Engine chips capable of processing 70,000 coordinates per second with 32-bit floating-point precision. (2:18)
The system features Gouraud shading, hardware Z-buffering, and polygon raster fill rates up to 44 million pixels per second on Unix 4.2 BSD with Motorola 68010 or 68020 processors. (2:18)

Detailed notes

## Background Published on YouTube by VintageCG on March 7, 2010, this promotional demonstration highlights the 1985 Silicon Graphics IRIS 2400 workstation. The video's description humorously reflects on the machine's then-revolutionary ability to animate flat-shaded polygons at several frames per second, noting its 10 MHz 68010 CPU and exorbitant price tag long before SGI's famous Jurassic Park era.

## TL;DR The video showcases the Silicon Graphics IRIS 2400 workstation, presenting unedited real-time graphics captured directly from the display monitor to one-inch videotape. Citing IBM research on response latency, the narrator demonstrates how eliminating interface delay below half a second dramatically elevates engineering productivity. The presentation details the machine's Unix environment and its proprietary pipeline of twelve custom VLSI Geometry Engines that process 32-bit floating-point transformations, Gouraud shading, and hardware Z-buffering in silicon. Ultimately, Silicon Graphics argues that interactive 3D color graphics and real-time motion are fundamental necessities that redefine the entire technical design cycle.

## Atmosphere & style The presentation carries the measured, authoritative tone of a premium mid-1980s corporate technology briefing, starting with a crisp broadcast countdown and an emphatic insistence on optical honesty. Its serious, measured delivery balances hard microprocessor metrics and IBM ergonomics charts with a visionary pitch for the future of interactive computing. Watching the smooth wireframes and Gouraud-shaded solids execute across the display evokes the thrilling dawn of 3D workstation computing, when real-time geometry calculation was treated as an absolute engineering triumph.

## Bulletin - [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) Silicon Graphics demonstrated the IRIS 2400 workstation executing real-time 3D graphics captured directly from monitor output to 1-inch videotape without video editing tricks or precomputed frame buffering. - [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) Citing formal IBM research, the presentation argues that reducing display latency below 0.5 seconds triggers an exponential rise in engineering productivity and turnaround improvements exceeding 50% to 60%. - [02:18](https://www.youtube.com/watch?v=9EEY87HAHzk&t=138s) The IRIS 2400 achieves real-time 3D performance using a hardware pipeline of 12 custom VLSI Geometry Engine chips capable of processing 70,000 coordinates per second with 32-bit floating-point precision. - [02:18](https://www.youtube.com/watch?v=9EEY87HAHzk&t=138s) The system features Gouraud shading, hardware Z-buffering, and polygon raster fill rates up to 44 million pixels per second on Unix 4.2 BSD with Motorola 68010 or 68020 processors.

## Key takeaways - [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) The demonstration footage was captured directly from an IRIS 2400 monitor to videotape in real time, relying entirely on standard high-level graphics software library calls. - [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) SGI avoided standard rendering crutches of the era, such as custom microcoding software or replaying pre-rendered, frame-buffered image loops. - [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) Minimizing display latency is framed as an economic imperative, as engineering productivity directly correlates with instantaneous system responsiveness. - [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) IBM research showed that slashing interactive graphics display latency from 2.0 seconds to 0.5 seconds resulted in project turnaround gains of 50% to 60% or more. - [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) User productivity in terms of transactions per hour climbs virtually vertically once response times drop below the half-second threshold. - [02:18](https://www.youtube.com/watch?v=9EEY87HAHzk&t=138s) The host architecture runs a Motorola 68010 or 68020 processor on Unix with 4.2 BSD extensions, demand-paged virtual memory, and Ethernet networking. - [02:18](https://www.youtube.com/watch?v=9EEY87HAHzk&t=138s) SGI's proprietary Multiple Exposure window manager offers engineers concurrent, overlapping workspaces. - [02:18](https://www.youtube.com/watch?v=9EEY87HAHzk&t=138s) Real-time graphics acceleration is embedded directly into foundational silicon using 12 custom VLSI Geometry Engines containing 80,000 transistors each. - [02:18](https://www.youtube.com/watch?v=9EEY87HAHzk&t=138s) The Geometry Engine pipeline handles 32-bit floating-point coordinate transformations and clipping at rates up to 70,000 coordinates per second. - [02:18](https://www.youtube.com/watch?v=9EEY87HAHzk&t=138s) The workstation draws up to 22,000 depth-cued vectors per second with true perspective correction. - [02:18](https://www.youtube.com/watch?v=9EEY87HAHzk&t=138s) The hardware natively calculates parametric cubic curves in real time, including Bézier, B-spline, and cardinal splines. - [02:18](https://www.youtube.com/watch?v=9EEY87HAHzk&t=138s) Rasterization throughput reaches up to 44 million pixels per second, featuring Gouraud shading and hardware Z-buffer hidden-surface removal. - [04:05](https://www.youtube.com/watch?v=9EEY87HAHzk&t=245s) SGI defines its modern engineering design cycle around three core visual pillars: color, dimension (2D and 3D), and real-time motion. - [04:05](https://www.youtube.com/watch?v=9EEY87HAHzk&t=245s) Color, depth, and kinetic simulation are treated as essential tools for iterative 'what-if' design exploration rather than cosmetic additions. - [04:05](https://www.youtube.com/watch?v=9EEY87HAHzk&t=245s) SGI concludes that once engineers experience true real-time responsiveness, industry expectations permanently shift away from high-latency computing.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) Real-Time Interactive Graphics and the Value of Response Time

Following a crisp broadcast countdown, the demonstration opens with an uncompromised technical declaration: the footage being shown was captured directly from the display monitor of a Silicon Graphics IRIS 2400 workstation straight to one-inch videotape. The narrator stresses that there are zero special video editing effects, trickery, or post-production illusions involved. The fluid motion sequences unfolding on screen are recorded in genuine real time as they execute on the workstation display. Furthermore, the underlying code relies purely on standard calls to Silicon Graphics’ high-level graphics software library—bypassing traditional rendering crutches such as custom microcoding software or playing back precomputed, frame-buffered image loops. Silicon Graphics Computer Systems positions itself as the only workstation manufacturer truly matching the computational horsepower of a 32-bit architecture with equal or superior processing dedicated exclusively to instantaneous display response.

This architectural priority stems from an unforgiving commercial reality: in technical workstation environments, time equals money, and that cost is dictated directly by display latency. To reinforce the stakes, the presentation cites formal IBM research investigating the psychological and economic impact of system responsiveness. When system lag disappears, the dynamic between engineer and terminal undergoes a fundamental shift. In the words of the IBM study highlighted by the narrator, “When a computer and its users interact at a pace that ensures that neither has to wait on the other, productivity soars.” Work costs collapse, employee satisfaction rises, and output quality noticeably climbs.

In IBM’s specific investigations into interactive graphics applications, slashing display response times from two seconds down to half a second yielded measured project turnaround improvements of 50%, 60%, and higher. A related IBM study tracking operator throughput demonstrated that transaction-per-hour productivity spikes almost vertically once latency dips beneath that critical 0.5-second threshold. Silicon Graphics aims straight past sub-second delays toward absolute immediacy. As the narrator underscores: “We believe that the productivity of companies racing to market, or racing to deadline, or maximizing the output of key individuals will require a truly interactive system.” The engineering goal is simple yet uncompromising: provide the fastest interactive display response on the market without saddling the customer with a pricing penalty.

- Direct recording output: Taped directly from the IRIS 2400 monitor to 1-inch videotape without post-processing or optical effects. - Software baseline: Standard high-level graphics software library calls without microcoding tricks or pre-rendered frame buffering. - Empirical backing: IBM interactive graphics studies showing that cutting response time from 2.0 seconds to 0.5 seconds yields 50% to 60%+ project completion gains, with productivity curves going virtually vertical below half a second.

---

[02:18](https://www.youtube.com/watch?v=9EEY87HAHzk&t=138s) The IRIS 2400 Architecture: Geometry Engines and Hardware Acceleration

The hardware underpinning this real-time performance is tailored to satisfy demanding compute and networking standards right out of the box. The core system runs on either a Motorola 68010 or 68020 central processor, paired with the Unix operating system supplemented by 4.2 Berkeley (BSD) extensions, full demand paging, and Ethernet local area network connectivity. On the desktop interface side, SGI’s proprietary Multiple Exposure window manager provides engineers with simultaneous, overlapping, on-demand display workspaces. Crucially, the graphics performance does not rely on third-party outboard accelerators or tacked-on add-in boards; it is integrated directly into the foundational silicon architecture.

At the heart of the machine lies a pipeline of twelve custom VLSI Geometry Engine chips, each packing 80,000 transistors. Arranged in a dedicated hardware pipeline, these custom processors consume high-level Graphics Library commands and execute coordinate transformations and display calculations directly in hardware, slashing the software overhead traditionally required to render dynamic 3D visuals. Operating with full 32-bit floating-point precision, the subsystem transforms, clips, and renders visual data at speeds reaching 70,000 coordinates per second, alongside 22,000 depth-cued vectors per second drawn with perspective correction. The pipeline evaluates and draws complex parametric cubic curves—including Bézier, B-spline, and cardinal splines—in real time, while rendering smooth surfaces with Gouraud shading and hardware Z-buffer hidden-surface removal at raster polygon fill rates up to 44 million pixels per second.

```mermaid flowchart LR accTitle: SGI IRIS 2400 Hardware Pipeline accDescr: Diagram illustrating the flow from CPU and software down through the 12 Geometry Engines to the raster engine and CRT display.

subgraph Host ["Host Workstation"] CPU["Motorola 68010 / 68020 CPU"] OS["Unix 4.2 BSD + Demand Paging"] GL["Graphics Software Library"] CPU --- OS OS --> GL end

subgraph Geometry ["Geometry Pipeline"] GE["12x Custom VLSI Geometry Engines (80,000 transistors each)"] Transforms["32-bit Floating-Point Transforms & Clipping"] GE --- Transforms end

subgraph Raster ["Rasterization & Display"] ZB["Hardware Z-Buffer & Gouraud Shading"] Display["Color CRT Display (Real-Time Interactive Output)"] ZB --> Display end

GL -->|"High-level graphics calls"| GE Transforms -->|"70,000 coords/sec 44M pixels/sec fill"| ZB ```

- Processor options: Motorola 68010 or Motorola 68020 microprocessor. - Operating environment: Unix operating system with 4.2 BSD extensions, demand-paged virtual memory, Ethernet networking, and the Multiple Exposure window manager. - Custom silicon: 12 pipelined VLSI Geometry Engines containing 80,000 transistors per chip. - Floating-point & vector metrics: Full 32-bit floating-point arithmetic; up to 70,000 coordinates per second transformed, clipped, and plotted; 22,000 depth-cued lines per second with true perspective. - Spline and surface support: Real-time generation of Bézier, B-spline, and cardinal spline cubic curves. - Shading & raster throughput: Gouraud shading, hardware Z-buffering for hidden surface removal, and fill rates scaling up to 44 million pixels per second.

---

[04:05](https://www.youtube.com/watch?v=9EEY87HAHzk&t=245s) Color, Dimension, and Motion: Defining the Modern Design Cycle

Silicon Graphics separates its platform from conventional workstations by structuring its visual architecture around three core pillars: the system thinks in color, it thinks in two and three dimensions, and it thinks in motion. The narrator stresses that in the iterative, exploratory "what-if" workflow central to modern engineering and industrial design, these three pillars are not cosmetic luxuries or optional enhancements—they represent the absolute essence of the design task itself.

Color allows engineers to instantly distinguish intricate layers in complex circuit layouts, inspect parts under varied simulated lighting setups, and evaluate surface finishes prior to manufacturing. True three-dimensional depth allows designers to properly judge spatial relationships and accelerate model building and iterative adjustments. Motion provides the interactive capability required to simulate real-world mechanical behaviors and kinetic performance before committing to physical prototypes.

The presentation closes with an ultimatum for organizations evaluating technical computing hardware. Design cycles inevitably follow speed: whichever process is fastest and most intuitive establishes itself as the standard for the entire discipline. The narrator warns that expectations cannot be rolled back once this threshold is crossed: “Once your users or your competitors' users have experienced real-time response, they will not settle easily for lesser productivity.” Silicon Graphics pitches the IRIS 2400 not merely as another piece of office hardware, but as the benchmark by which all future high-performance engineering workstations must be judged.

## Notable quotes > [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) When a computer and its users interact at a pace that ensures that neither has to wait on the other, productivity soars.

> [04:05](https://www.youtube.com/watch?v=9EEY87HAHzk&t=245s) Once your users or your competitors' users have experienced real-time response, they will not settle easily for lesser productivity.

> [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) We believe that the productivity of companies racing to market, or racing to deadline, or maximizing the output of key individuals will require a truly interactive system.

## Mentioned Silicon Graphics Computer Systems, IRIS 2400, IBM, Motorola 68010, Motorola 68020, Unix, 4.2 BSD, Multiple Exposure, Geometry Engine

## Video description BEHOLD the AWESOME POWER of the Silicon Graphics IRIS 2400 workstation. Able to animate DOZENS of flat-shaded polygons at SEVERAL frames per second! Long before SGI's Jurassic Park heyday, their early IRIS 2400 system cost more than a nice condo and came equipped with a 68010 CPU running at a blazing TEN megahertz!

[Source](https://www.youtube.com/watch?v=9EEY87HAHzk)

中文简报

Silicon Graphics推出IRIS 2400以消除工作站延迟 Silicon Graphics援引IBM关于半秒以下响应时间的研究指出，由于工程师将不再容忍速度较慢的工作站，实时交互式渲染将决定市场赢家。该视频展示了Silicon Graphics IRIS 2400工作站，呈现了直接从显示器采集至一英寸录像带、未作剪辑的实时图形。解说员引用IBM关于响应延迟的研究，展示了将界面延迟降至半秒以下如何显著提升工程生产力。演示详细介绍了该机器的Unix环境及其由12颗定制VLSI几何引擎（Geometry Engine）构成的专有流水线，这些引擎通过芯片硬件处理32位浮点变换、Gouraud着色以及硬件Z缓冲。Silicon Graphics最终认为，交互式3D彩色图形与实时运动是重塑整个技术设计周期的根本必需品。 Silicon Graphics展示了IRIS 2400工作站运行实时3D图形，画面直接从显示器输出采集至1英寸录像带，未使用视频剪辑特效或预先计算的帧缓冲。 (0:00) 演示援引IBM的正式研究指出，将显示延迟降至0.5秒以下会引发工程生产力呈指数级增长，周期改善幅度超过50%至60%。 (0:00) IRIS 2400采用由12颗定制VLSI几何引擎芯片组成的硬件流水线实现实时3D性能，能够以32位浮点精度每秒处理70,000个坐标。 (2:18) 该系统支持Gouraud着色、硬件Z缓冲，并在运行Unix 4.2 BSD、配备Motorola 68010或68020处理器的环境下，实现高达每秒4400万像素的多边形光栅填充率。 (2:18)

Reading summary

Silicon Graphics Computer Systems positions the IRIS 2400 workstation as an integrated solution to technical design bottlenecks, arguing that real-time visual responsiveness is essential for engineering productivity. The system couples general-purpose Unix processing with dedicated custom silicon to execute dynamic 3D graphics calculations directly in hardware. The Economic Value of Display Latency Reduction (0:00) Silicon Graphics bases its architectural strategy on the principle that system latency directly dictates technical project costs. Citing research from IBM on interactive graphics environments, the presentation notes that reducing display response times from 2.0 seconds to 0.5 seconds produced measured project turnaround improvements between 50% and over 60%. The cited IBM data further indicates that operator throughput and completed transactions per hour increase sharply once response lag drops below the 0.5-second threshold. To prove that the IRIS 2400 achieves true real-time display speeds without synthetic acceleration, the demonstration emphasizes that its footage was recorded directly from the workstation monitor onto one-inch videotape without post-production effects or editing. The visuals were driven entirely by standard calls to SGI's high-level graphics software library, operating without custom microcoded routines or pre-rendered frame-buffer playback loops. Custom Geometry Engines and Workstation Architecture (2:18) The IRIS 2400 is powered by either a Motorola 68010 or 68020 central processor running the Unix operating system with 4.2 Berkeley (BSD) extensions, demand-paged virtual memory, and Ethernet networking. System interaction is handled via SGI's proprietary Multiple Exposure window manager, which enables overlapping, simultaneous workspaces on the desktop without relying on third-party add-in accelerator boards. Graphics acceleration is performed by a dedicated pipeline of twelve custom VLSI Geometry Engine chips, each containing 80,000 transistors. Operating with 32-bit floating-point precision, this custom array calculates coordinate transformations and clipping in hardware, reaching processing rates of 70,000 coordinates per second and drawing 22,000 depth-cued vectors per second with true perspective. The pipeline also generates parametric cubic curves—such as Bézier, B-spline, and cardinal splines—in real time, and supports Gouraud surface shading alongside hardware Z-buffering at raster polygon fill rates reaching 44 million pixels per second. Color, Dimension, and Motion in Engineering Workflows (4:05) Silicon Graphics defines its visual architecture around three essential capabilities: color, multidimensional depth, and motion. In iterative design cycles, the company argues that these features are functional necessities rather than aesthetic enhancements. Color enables engineers to inspect multilayer circuit board designs and evaluate simulated surface finishes, while true three-dimensional depth provides the spatial feedback needed for rapid geometric modeling. Real-time motion allows designers to evaluate dynamic mechanical interactions and kinetic behaviors directly on screen before building physical prototypes. The presentation contends that incorporating these capabilities directly into the interactive design loop accelerates decision-making and minimizes costly revisions. The presentation concludes that real-time interactive response establishes an irreversible standard across technical computing disciplines. Silicon Graphics asserts that once design teams experience lag-free 3D manipulation, organizations cannot revert to higher-latency systems without incurring severe competitive and productivity penalties.

精炼阅读摘要

Silicon Graphics（SGI）在1985年的宣传展示中指出，IRIS 2400工作站通过专有硬件加速实现了完全无后期修饰的实时三维交互，将瞬时系统响应时间确立为技术工作站提升工程生产率的核心决定因素。系统即时响应时间对工程生产率的商业价值 (0:00) 展示视频开篇强调，演示画面是从IRIS 2400监视器直接录制到一英寸录像带上，没有借助后期剪辑或预渲染循环帧缓冲等技巧，完全依赖标准高级图形库调用。SGI认为，在技术工作站领域，时间成本直接由显示延迟决定，因此将32位计算架构与专门负责即时显示的图形处理硬件紧密结合。 IBM人机交互实证研究与亚秒级响应阈值 (0:00) SGI援引IBM关于人机交互心理学与经济效应的研究作为依据。IBM的研究发现，当计算机与用户的交互达到互不等待的节奏时，工作成本显著下降且质量提升。具体到交互式图形应用中，当显示响应时间从2.0秒缩短至0.5秒时，项目周期缩短了50%到60%以上；一旦系统延迟突破0.5秒的关键阈值，操作员每小时的处理效率曲线呈现近乎垂直的飙升。SGI据此主张，企业缩短上市周期必须依赖完全交互式的亚秒级响应系统。几何引擎流水线与专用硅芯片加速架构 (2:18) IRIS 2400的核心计算单元采用Motorola 68010或68020处理器，运行带有4.2 BSD扩展及请求分页虚拟内存的Unix系统，配备以太网接口与Multiple Exposure窗口管理器。其图形性能直接内建于底层硬件架构中，核心由12颗专有超大规模集成电路（VLSI）几何引擎芯片（Geometry Engine）组成的硬件流水线构成，每颗芯片包含8万个晶体管，专职硬件化处理坐标变换与显示计算。 32位浮点渲染与曲面光栅化性能指标 (2:18) 该硬件流水线采用全32位浮点精度运算，每秒可完成多达70,000个坐标的变换、裁剪与绘制，以及22,000条具备真实透视矫正的深度提示向量。系统支持实时生成Bézier曲线、B样条和Cardinal样条等三次参数曲线，并结合Gouraud着色与用于消除隐藏面的硬件Z缓冲器，光栅多边形填充速率最高可达每秒4,400万像素。色彩、三维与动态交互重塑工业设计流程 (4:05) SGI将IRIS 2400的架构支柱概括为具备色彩、二维及三维空间以及动态模拟的运算能力，并指出这些特性并非可有可无的装饰，而是现代工程探索性设计流程的基础。色彩有助于区分复杂电路图层和评估材质表面；真实三维深度加速了空间关系判断与建模；动态交互则允许工程师在制造物理样机前直接模拟机械运动特性。 SGI总结认为，设计流程的标准往往由效率最高、最直观的工具确立；一旦用户或其竞争对手体验过实时响应带来的生产率提升，便很难再退回低效的旧工作模式。该演示将IRIS 2400定位为衡量未来高性能工程工作站的核心基准。

Detailed notes — zh

## Background 该视频由 YouTube 频道 VintageCG 于 2010 年 3 月 7 日发布。视频展示了 1985 年发布的 Silicon Graphics IRIS 2400 图形工作站，页面描述感叹了这台搭载 10MHz 68010 CPU、价格堪比高档公寓的早期系统在当时能够以每秒数帧的速度实时渲染数十个平滑着色多边形的惊人算力。 ## TL;DR 本片是 Silicon Graphics 为其 IRIS 2400 工作站制作的宣传演示，画面全程直接从监视器录制，不含任何后期特效，力证其三维实时图形能力的真实性。视频引用 IBM 的权威研究指出，将系统显示响应时间缩短至半秒以内能够呈指数级提升工程师的生产率。为此，IRIS 2400 搭载了由 12 颗专用 VLSI 几何引擎组成的硬件流水线与硬件 Z 缓冲区，无需手写微码即可实现每秒 7 万个顶点的变换与高达 4400 万像素的填充率。最后，SGI 提出了“以色彩、三维维度和动态来思考”的设计哲学，强调唯有这种无延迟的实时交互才能确立工程行业全新的生产力标准。 ## Atmosphere & style 全片洋溢着 1980 年代工业宣传片特有的严谨与自信，倒计时的开场和严格的技术免责声明奠定了硬核、真实的基调。画外音旁白沉稳有力、条理清晰，既有商用论证的冷静说服力，又带着前沿计算技术突破时的自豪感。观看时能直观感受到复古 CRT 屏幕上实时矢量线条与平滑多边形翻转所带来的科技震撼，仿佛亲历了现代 3D 计算图形技术破晓初期的激荡时刻。 ## Bulletin - [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) Silicon Graphics（SGI）发布的 IRIS 2400 演示视频完全从显示器直接录制至 1 英寸录像带，全程未加任何后期特效，以此力证其高阶图形软件库配合专用硬件即可实现真实的三维实时渲染。 - [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) SGI 引用 IBM 生产率研究指出，将图形响应时间从 2 秒降至 0.5 秒以内可缩短 50% 至 60% 以上的项目交付工期，使用户每小时事务处理量呈近乎垂直的爆发式增长。 - [02:30](https://www.youtube.com/watch?v=9EEY87HAHzk&t=150s) IRIS 2400 放弃了传统外挂加速卡方案，改用 12 颗专用 VLSI“几何引擎”（每颗 8 万晶体管）构成底层硬件流水线，具备全 32 位浮点精度与高达每秒 4400 万像素的多边形填充率。 - [04:20](https://www.youtube.com/watch?v=9EEY87HAHzk&t=260s) SGI 确立了以“色彩、二维/三维空间与动态”为核心的交互显示哲学，强调最快、最直观的设计周期必将取代落后流程，成为工业界衡量图形工作站的新标杆。 ## Key takeaways - [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) 宣传视频直接从 IRIS 2400 监视器转录至广播级录像带，杜绝任何视频后期特效与预计算帧缓冲拟态，保证演示效果完全为实时生成。 - [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) SGI 强调自身率先在 32 位通用计算能力与超高速显示响应性能之间实现了真正平衡。 - [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) 系统响应时间是工作站效率的核心指标；IBM 研究证明让计算机与用户无需相互等待可大幅降低研发成本并提升产出质量。 - [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) 测试数据表明，响应时间从 2 秒缩短至 0.5 秒能削减 50%～60% 的工程周期，而延迟跌破半秒后生产率吞吐呈垂直上升趋势。 - [02:30](https://www.youtube.com/watch?v=9EEY87HAHzk&t=150s) IRIS 2400 基础计算架构采用 Motorola 68010 或 68020 处理器，运行支持按需分页（Demand Paging）的 Unix 4.2 BSD 操作系统。 - [02:30](https://www.youtube.com/watch?v=9EEY87HAHzk&t=150s) 配备标准以太网通信与多重曝光窗口管理器（Multiple Exposure Window Manager），支持高效多任务并发显示。 - [02:30](https://www.youtube.com/watch?v=9EEY87HAHzk&t=150s) 系统摒弃了外挂式加速卡做法，选择将 12 颗专用 VLSI“几何引擎”流水线直接整合进系统底层硬件架构。 - [02:30](https://www.youtube.com/watch?v=9EEY87HAHzk&t=150s) 12 颗几何引擎各集成 8 万个晶体管，无需手动编写微代码即可由硬件原生执行高阶图形库命令。 - [02:30](https://www.youtube.com/watch?v=9EEY87HAHzk&t=150s) 硬件流水线具备 32 位全浮点运算能力，可实现每秒 70,000 个三维坐标的空间变换、裁剪与透视投影。 - [02:30](https://www.youtube.com/watch?v=9EEY87HAHzk&t=150s) 在真实透视条件下，系统每秒能绘制 22,000 条深度提示线，并支持实时计算贝塞尔、B样条与基数样条等三次曲线。 - [02:30](https://www.youtube.com/watch?v=9EEY87HAHzk&t=150s) 集成硬件 Z 缓冲区，支持实时隐藏面消除、高洛德着色（Gouraud shading），多边形填充率高达 4400 万像素/秒。 - [04:20](https://www.youtube.com/watch?v=9EEY87HAHzk&t=260s) SGI 提出三大核心显示理念：以色彩思考、以二维和三维空间思考、以动态思考，三者是工程“假设分析”必不可少的基础。 - [04:20](https://www.youtube.com/watch?v=9EEY87HAHzk&t=260s) 色彩用于区分电路布线与评估打光细节，空间维度加速空间关系评估，动态效果则用于模拟物理运动与机构行为。 - [04:20](https://www.youtube.com/watch?v=9EEY87HAHzk&t=260s) SGI 指出最快且最直观的设计周期将成为行业通用标准，体验过即时响应的用户将无法倒退容忍延迟低效的传统工作流。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=9EEY87HAHzk&t=0s) 实时图形与响应时间的重要性：时间就是金钱演示在一声倒计时中拉开帷幕，并特别对视频画面的真实性做出了严格声明：整部宣传片直接从 Silicon Graphics IRIS 2400 工作站的显示器转录至 1 英寸录像带（1-inch videotape），全程绝未采用任何特殊的视频后期特效。观众所看到的每一段三维动态显示序列，全部都是工作站在屏幕上实时运行并实时录制的。软件的编写仅调用了高阶图形软件库（high-level graphics software library），并没有借助微代码软件（microcoding software）或预先计算好放入帧缓冲区的静态图像序列等传统拟态动画手段。Silicon Graphics（SGI）强调，自己是当时唯一一家将工作站的 32 位通用计算能力，与同等甚至更强的高速显示响应能力完美平衡的工作站厂商。在解释为何要如此聚焦显示响应速度时，旁白指出在工作站应用环境中，“时间就是金钱”是一条颠扑不破的铁律，而这其中的时间核心就是显示器的响应时间。视频引用了 IBM 针对系统响应时间改进效果所发表的权威研究报告： > “When a computer and its users interact at a pace that ensures that neither has to wait on the other, productivity soars. The cost of the work done on the computer tumbles. Employees get more satisfaction from their work and its quality tends to improve.” 随后，旁白通过数据进一步展示响应时间对项目交付周期的深远影响。在 IBM 的交互式图形应用测试中，仅仅将显示响应时间从 2 秒降低到 0.5 秒，项目完成时间就实现了 50% 到 60% 甚至更多的惊人削减。在另一项以“每小时事务处理量（transactions per hour）”衡量生产率的研究中，数据显示随着响应时间跌破半秒，用户生产率曲线近乎呈现垂直攀升。SGI 提出，对于那些争分夺秒抢占市场先机、赶工期截止日或旨在激发核心技术人员最大产出的企业而言，他们需要的系统绝不能是数秒的等待，甚至不能是几分之一秒的延迟，而是真正的“实时（real-time）”与“动态（motion）”。 - 录制格式：直接从 IRIS 2400 监视器录制至 1 英寸广播级录像带，无后期视频剪辑特效。 - IBM 研究数据：显示响应时间由 2.0 秒缩短至 0.5 秒，项目工期缩短 50%～60% 以上。 - 生产率表现：当响应延迟进入 0.5 秒以内时，用户单位小时的事务吞吐曲线近乎垂直上升。 --- ### [02:30](https://www.youtube.com/watch?v=9EEY87HAHzk&t=150s) 核心硬件架构：VLSI 几何引擎流水线与图形性能为了在不产生价格惩罚（no price penalty）的前提下提供极致的显示响应，Silicon Graphics 拿出了软硬结合的整套解决方案。在基础计算与通用环境方面，系统紧贴市场在处理与网络层面的主流标准，配备 Motorola 68010 或 68020 CPU，运行具备 Berkeley 4.2 扩展（含按需分页 demand paging）的 Unix 操作系统，内置以太网（Ethernet）局域通信，并配备多重曝光窗口管理器（multiple exposure window manager），支持按需多显示任务。更重要的是，SGI 坚决拒绝使用外挂式加速卡（add-on accelerators）这种修修补补的方式，而是将超大规模集成电路（VLSI）方案直接深植于系统底层架构之中。 ```mermaid flowchart LR accTitle: IRIS 2400 渲染管线架构 accDescr: 展示从高阶图形库命令到 12 颗几何引擎流水线，再到光栅化输出的整个处理过程。 subgraph CPU_SYS["主机系统与控制"] A["CPU (MC68010 / MC68020) + Unix 4.2 BSD"] B["高阶图形软件库 (Graphics Library)"] A --> B end subgraph GE_PIPE["VLSI 几何处理流水线"] C["12 颗 VLSI 几何引擎 (每颗 8 万晶体管)"] D["32 位浮点流水线 变换、裁剪、透视投影"] C --- D end subgraph RASTER_OUT["光栅化与显示输出"] E["硬件 Z 缓冲区 (消隐与 Gouraud 着色)"] F["CRT 实时显示输出 (多边形填充 4400 万像素/秒)"] E --> F end B -->|"高阶图形调用"| C D -->|"顶点与线框数据"| E ``` 该架构的核心是流水线配置的 12 颗专用 VLSI“几何引擎（Geometry Engines）”，每颗集成 80,000 个晶体管。它们通过纯硬件方式直接执行高阶图形库指令，彻底省去了手工编写微码的痛苦，极大缩减了创建动态三维图形的时间。系统具备全 32 位浮点运算精度，能够以每秒高达 70,000 个坐标的速度进行空间变换、裁剪和显示；在具备真实透视（true perspective）的前提下，每秒可绘制 22,000 条深度提示线（depth-cued lines）。工作站不仅能实时计算并绘制贝塞尔曲线（Bézier）、B样条曲线（B-spline）、基数样条曲线（Cardinal spline）以及其他三次曲线，还通过硬件 Z 缓冲（hardware Z-buffering）实现了平滑表面渲染、高洛德着色（Gouraud shading）和隐藏面消除，多边形填充速率最高可达每秒 4400 万像素。 - 处理器平台：Motorola 68010 或 68020 CPU，搭载支持 Demand Paging 的 4.2 BSD Unix。 - 网络与窗口：标准以太网通信接口，结合 Multiple Exposure Window Manager 实现多窗口实时显示。 - 几何加速核心：12 颗专用 VLSI 几何引擎（单颗集成 80,000 晶体管），纯硬件流水线驱动。 - 顶点与线段性能：32 位全浮点精度，每秒 70,000 个坐标变换/裁剪/显示；每秒 22,000 条真实透视深度提示线。 - 曲线与渲染能力：实时生成 Bézier、B-spline、Cardinal spline 等三次曲线；硬件 Z 缓冲区支持 Gouraud 着色与消隐；多边形填充率达 44,000,000 像素/秒。 --- ### [04:20](https://www.youtube.com/watch?v=9EEY87HAHzk&t=260s) 三大显示理念与工程设计的新标准针对显示哲学，SGI 指出有三大根本性的显示思维让该系统遥遥领先于业界同行：第一是“以色彩来思考（thinks in color）”，第二是“以二维和三维空间来思考（thinks in two and three dimensions）”，第三则是“以动态来思考（thinks in motion）”。这三者之所以至关重要，是因为在工程与设计的“假设分析（what-if）”探索过程中，色彩、维度与动态绝非锦上添花的奢侈品，它们构成了设计的本质核心。旁白详细拆解了三者的实际应用场景：工程师需要色彩来区分电路版图（layout）上的复杂走线、创建直观的视觉图形、在不同光源条件下检查设计效果，或是预览完成的曲面细节；需要三维空间维度来精准评估空间相对关系，成倍加速模型的构建与修改；更需要动态效果来真实模拟整个设计的行为机制与物理运动。旁白对设计工具演进的市场规律做出了极具预见性的论断： > “The fastest, most intuitive design cycle becomes the industry standard design cycle.” 随着系统演示走向尾声，视频向潜在客户发出了明确的选型警示：无论是用户自身，还是竞争对手的设计师，一旦体验过这种指哪打哪的“实时响应”，就再也无法回退容忍低效率的工作流程。这正是客户在拍板选定工作站供应商之前，必须认真考量 Silicon Graphics 解决方案的原因所在——SGI 为工程界带来的这一套生产力与创造力体系，必将成为衡量所有高性能图形工作站的行业新标杆。 ## Notable quotes > [04:20](https://www.youtube.com/watch?v=9EEY87HAHzk&t=260s) The fastest, most intuitive design cycle becomes the industry standard design cycle. ## Mentioned Silicon Graphics (SGI), IRIS 2400, IBM, Motorola 68010, Motorola 68020, Unix 4.2 BSD, Ethernet, Multiple Exposure Window Manager, Geometry Engine (几何引擎) ## Video description BEHOLD the AWESOME POWER of the Silicon Graphics IRIS 2400 workstation. Able to animate DOZENS of flat-shaded polygons at SEVERAL frames per second! Long before SGI's Jurassic Park heyday, their early IRIS 2400 system cost more than a nice condo and came equipped with a 68010 CPU running at a blazing TEN megahertz! [Source](https://www.youtube.com/watch?v=9EEY87HAHzk)

Source quotes

[4:20] The fastest, most intuitive design cycle becomes the industry standard design cycle.

Source quotes

[0:00] When a computer and its users interact at a pace that ensures that neither has to wait on the other, productivity soars. [4:05] Once your users or your competitors' users have experienced real-time response, they will not settle easily for lesser productivity. [0:00] We believe that the productivity of companies racing to market, or racing to deadline, or maximizing the output of key individuals will require a truly interactive system.

Original transcript

[0:00] 10, 9, 8, 7... Six. 5. Four. three, two, This demonstration video is taped directly from the display monitor of the Silicon Graphics Iris 2400 workstation to 1 inch videotape. No special video editing effects are used. Motion display sequences such as the one you are now watching are taped in real time as they run on the workstation display. The software was written using calls to the high-level graphics software library. Traditional motion techniques, such as microcoating software or frame-buffered, pre-computed images, are not employed. Silicon Graphics Computer Systems is in fact the only workstation vendor balancing the 32-bit computing power of the workstation with equivalent or greater power for fast display response. Why? The equation time equals money is a fact of life in the workstation environment, particularly where time is the response time of the workstation display. IBM has published studies on the effect of improved system response time. Highlight. When a computer and its users interact at a pace that ensures that neither has to wait on the other, productivity soars. The cost of the work done on the computer tumbles. Employees get more satisfaction from their work and its quality tends to improve. Consider the effect of response time on project time to completion. In IBM studies of interactive graphics applications, reducing display response time from two seconds to just one half second resulted in documented time to completion improvements of 50%, 60%, and more. In a separate response time study, IBM scientists document dramatic jumps in user productivity measured in transactions per hour as display response time is reduced. Notice that the curve measuring user productivity is nearly vertical as response time passes below one-half second. The silicon graphics system provides display response that is instant. We believe that the productivity of companies racing to market, or racing to deadline, or maximizing the output of key individuals will require a truly interactive system. not seconds, not fractions of seconds, but real time. Motion. Our goal is to be able to do this. Deliver a workstation with the fastest display response possible with no price penalty. The Silicon Graphics Solution. a workstation that meets the processing and networking requirements of the marketplace. Motorola 68010 or 68020 CPU. The Unix operating system with 4.2 Berkeley extensions including demand paging. Ethernet local area communications, and our multiple exposure window manager for multiple displays on demand. Display performance is not addressed by add-on accelerators. but rather through a VLSI solution designed into the fundamental system architecture. Twelve of these 80,000 transistor VLSI geometry engines in a pipelined configuration provide the power to put high-level software in motion through direct hardware support and execution of displays. Displays created using the high-level software commands of the graphics library. radically reducing the time required to create dimensional visuals that move. It all comes together in a system that transforms, clips, and displays data at rates to 70,000 coordinates per second. gives you 22,000 depth-cued lines per second with true perspective. The system performs operations with full 32-bit floating point precision. Your workstation calculates and draws Begier, B-spline, cardinal spline, and other cubic curves in real time. It gives you smooth surfaces, garrot shading, and hidden surface removal using hardware Z-buffering. and fills polygons at up to 44 million pixels per second. Three fundamental approaches to the display set the silicon graphics system ahead. First, The system thinks in color. Second, the system thinks in two and three dimensions. Third, it thinks in motion. color Dimension motion. Why are they important? because in the what-if process that is engineering and design, color, Dimension and motion are not luxuries. Color, dimension, and motion are the essence of the design. Whether you are using color to differentiate the circuits on a layout, Create graphic visuals. look at designs under different light sources, or preview completed surface designs. whether you are using dementia [5:00] to evaluate spatial relationships. speed the construction and alteration of models or motion to simulate the behavior of your design. Color. dimension and motion are essentials. The benchmark is always the same. The fastest, most intuitive design cycle becomes the industry standard design cycle. Once your users or your competitors users have experienced real-time response, they will not settle easily for lesser productivity. That is why the Silicon Graphics solution deserves your serious consideration before you commit to a workstation vendor. because the productivity and creativity that silicon graphics brings to engineering are certain to become the standard by which high-performance workstations are measured.