Beyond3D
Sunday, Jun 21, 2015 · 4300 words · approx 21 mins to readAs it approaches 10 years of me being steward of Beyond3D, I thought I’d write down some memories of my time before and during. Kind of a retrospective in technical product analysis on the web in general, but with Beyond3D as the background. Beyond3D’s a funny old place, and while I’m not able to write down a complete, unabridged history, I’ve got a pretty good view of at least the last 12 years.
I joined the Beyond3D forums in 2003, eager to learn everything I could about how GPUs worked. I was Technical Editor of Hexus at the time, and I wanted my coverage of the latest graphics AIBs to be as good as it could possibly be. While we covered performance of things like the GeForce FX 5700 — remember NV36?! I certainly do! — reasonably well, I was never really able to explain why the products I reviewed went as fast as they did.
I didn’t really have any understanding of GPU microarchitecture, the underlying semiconductor manufacture, or an appreciation of how they were really driven by client drivers implementing the APIs of the time. Our ambitions at Hexus were grand. We did really well against our British colleagues like Bit-tech, but what we really wanted was to be as great, as good and as big as Anandtech. It’s funny, because that’s still what everyone wants, 18 years after Anand registered the domain name. A lot has obviously changed in the nearly 2 decades since, but honestly a lot has also stayed the same.
In 2003, as we cemented Hexus’ reputation as an in-depth analysis house, my job was not just to write reviews, but figure out how we could look at products in a deeper, more knowledgeable way. That meant really understanding what made PC components tick, from the transistor up. I figured that if we were to compete with Anandtech, we had to know more about what we were writing about than Anand and his team did. Hardware reviewing on the web is still played out in the same old way it was back then, pretty much. Those on the inside with the hardware vendors get given samples to review ahead of an embargoed launch, shared with everyone else on the sampling list. When the embargo is lifted, everyone publishes at the same time, with the end game hopefully that you gain readers because your view of the product is better and more insightful than any other you can read for free, all funded by ads.
The mechanics of how that actually plays out on the web these days is a bit more nuanced, compared to the late 1990s and early 2000s, and things are changing maybe a bit more rapidly now than they ever have done. The focus these days is on mobile technology, rather than PCs and PC components, at least in terms of numbers of eyeballs willing to read your content, and thus get you paid. Ad blocking is a prevalent thing. The big sites are now really big, which means the smaller ones are struggling more than they ever have done. I wrote about that a little while ago, so I won’t rehash it here.
So as Hexus played that game, the strategy to win seemed obvious to me: learn as much about how everything works as I could, then impart that knowledge into my copy, adding more insight into the analysis than anyone else to build strong reasons to repeatedly come back to Hexus for more. Pretty straightforward, and a plan that still makes sense today in many ways. I figured gaining that knowledge needed two plans of attack. The first, just hoovering up all existing expert knowledge that’s out there already, so I could stand on the shoulders of giants. The second, finding and inventing novel ways to ask new questions of the products using software. So when it came to looking at GPUs, I knew I needed to put the plan into action. I’d go and figure out what everyone already knew, and then teach myself how to get new information on my own.
So I joined Beyond3D on 9th October 2003. My public post history says a lot about what I knew about GPUs back then, which was pretty much just the bare minimum needed to get by. I knew how to measure their performance, and that’s a good place to start, but I honestly had no idea how the GPU created that performance. I could tell you that the GPU had x vertex shader units, y pixel shader units, and ran at z MHz. So I was great at playing GPU Top Trumps, but not much more.
The great thing about Beyond3D, and something that’s been the case since before I took over, is that the people who design and implement GPUs post there. I didn’t know that, though! So unwittingly, it turned out to be the perfect place to go hang out and figure things out; you weren’t just hanging out with enthusiasts, but the people who built them. Get something wrong and often someone at the company who made the exact thing you’re talking about will post a correction. I’ve never really been afraid of being wrong about things in public, so posting dumb stuff at Beyond3D was a great learning experience.
And I was at least 5 years late to the party. Beyond3D has around 20,000 members now, not all of them active by any means, but when I joined in 2003 most of them were. I’m user ID number 1816, and being surrounded by 1800 other GPU enthusiasts was great, lots of them posting daily. I owe my career to that experience, honestly. Not only that, but the first two years of my time at Beyond3D were some of the most illuminating and fun that I’ve had in technology.
Beyond3D was based in the UK at the time, run by “Wavey” Dave Baumann, now a senior product manager at AMD. David Ross, head honcho at Hexus, finally got the misguided courage in 2003 and 2004 to let me out of the box from time to time and attend UK press events, and later some international ones around graphics, so I got to meet and talk to Wavey. I admit to being a bit star struck when we first met. There was an event in London, I think for an ATI product, and we needed to cross the city to either get there or head to dinner afterwards, I can’t quite remember now. As we headed where we were headed, I formed a very fond memory that I’ve loved recalling over the years, of me talking up at him as we rode an escalator together on the London Underground, asking questions about GPUs. I vividly remember hoping he wouldn’t find them stupid. It was one of my first events, and one of my first times in London, and so I was literally making a first name for myself with my peers in the UK tech publishing industry, including Wavey. Other journalists were around and I remember being conscious and really nervous that they could hear what I was asking, and I was young, just 23 or 24, so the whole experience was completely overwhelming, but I loved it.
I’d found a real love for real-time graphics hardware in the first few years of my time at Hexus, since starting in 2000, and to be talking to the best public GPU analyst working at the time — and about GPUs! — is something that’ll stick with me until I can’t recall memories any more. Looking back on the time now, it was pretty much the golden age of real-time graphics in many ways. The industry was established, thanks to the founding hardware forays of 3Dfx, Rendition, S3, ATI and NVIDIA, and helped in no small part by GLQuake as the killer app to kick things into high gear.
The thing that got me hooked on 3D acceleration was the 3Dfx Voodoo. I remember heading into the city to buy some PC games one weekend in 1996, and right there next to the games on the shelf was a box for the Guillemot Maxi Gamer 3D add-in board. An uncustomised 3Dfx Voodoo 1 with 4MB of memory and not much in the box other than a driver CD and the VGA passthrough cable. It was £99, and I didn’t really know what to make of it at first, but it had a Quake logo on the box. That’s all the impetus I needed to grab it and go and pay, and I’ve had dedicated 3D acceleration in all of my computers ever since, be they standalone boards like the Voodoo and Voodoo 2, or integrated single-chips like Voodoo Banshee and almost every GPU on an add-in board that I’ve owned since.
I could write a terribly boring and long-winded book about my journey from double clicking on the GLQuake shortcut to the bit where I joined the site. There’s so much to talk about in that part of my graphics life, from trying to write a basic MiniGL driver for the S3 Virge (total failure!), after being hooked on what was possible via open source drivers on the Utah GLX project, through to writing a popular overclocking utility for 3Dfx and early Riva boards as I got hooked on extracting the maximum from my hardware for minimal outlay, and so much in between. My first spinning shaded triangle on a Voodoo 2 in Glide. Adding support for fog volumes in Quake after the source code release in 1999. Most of it revolved around Quake as you can see!
In 2003 and 2004 the GPU hardware industry, as quickly as it had started in the late 90s with many more major players, had already consolidated into basically just NVIDIA and ATI. Discrete graphics has been GeForce versus Radeon for over a decade now, with no other major players, yet in the years when the industry was in its infancy, especially before programmable 3D graphics was even a thing and graphics chips did just 2D and video, you needed more than the standard number of human digits, feet and hands, to count the competitors. So my industry baptism into how GPUs really worked happened long after the original industry big bang had cooled. I picked it all up when it was just the Red Team versus the Green Team, which is really just as it is now in the desktop PC space for discrete GPUs. Wavey and the folks posting at Beyond3D were a fine team to pick up the workings of dichotomous GPU architectures, and in many ways I’m thankful for only having to focus on GPUs from two companies, rather than 20.
I’d long convinced Tarinder Sandhu to join Hexus at this point. He’s still there now, I believe as Editor-in-Chief. David, I think knowing how much I loved graphics more than anything, had given Tarinder a lot of the processor and core PC platform analysis to do, leaving me to primarily focus on graphics AIBs (although looking back at my old Hexus content archive, I wrote an unhealthy amount of mainboard reviews too). That meant spending time on Beyond3D for a living! I’ve long learned that being paid to do the things you really love is the best way to enjoy a career, and that really kicked into gear around then.
A frequent face on the forums throughout 2005, I setup the Beyond3D IRC channel on the 5th December, on Freenode, where it still lives (#beyond3d if you want to come and say hello). Wavey joined, along with a core of others, shortly after I set it up, and on top of the discussion about graphics we all had on the forums, we had an outlet to discuss graphics in real time on IRC as well. That proved, and still does to this day, to be invaluable to me. Hanging out with a small set of hardcore GPU and PC enthusiasts, many of whom have also found subsequent homes in the GPU and wider industry as I did, was fantastic. It still is. That constant dialogue about graphics in front of and with Wavey proved to be a big part of the reason why, in early 2006, he asked me if I’d like to look after it. I didn’t even have to think about it. I just typed yes and hit Enter without a single thought as to any other outcome.
That “look after it” is important when it comes to Beyond3D, since it’s more than just running it, writing for it, or steering it. Those things happen, too, but looking after it is really what the steward of Beyond3D is responsible for doing. I was still working at Hexus at the time and with Beyond3D not a profitable concern by any means, the natural thing for me to do was continue to work at Hexus full-time, while the Hexus team figured out how to bring Beyond3D into the fold and make it profitable. The basic idea was that we’d do in-depth architecture analysis and new reference design reviews at Beyond3D, and have Hexus do the customised AIB reviews, with their coolers, clock changes, price wars and games bundles. That’d let me do what I really loved doing most of all, Hexus’ commercial team figuring out how to make it generate revenue.
Lots of great stuff happened in that period in 2005 and 2006, graphics and site wise. I don’t know how often fellow graphics hardware enthusiasts look back on those years, but they were golden for our industry. ATI introducing the first GPU with a unified shader architecture in Xenos, along with Wavey’s pioneering analysis of the microarchitecture. NVIDIA G80. The story of how Quake got its fast inverse square root function was uncovered. ATI R520. DirectX 10 was released. It’s hard to understate how much new pioneering hardware and software was developed and released across 2005 and 2006. I plan to do a podcast series, talking about the evolution of real-time graphics acceleration. I can’t wait to talk about what happened in those years.
Sadly, what I remember as a great time in graphics with some real momentum didn’t quite carry over to what was happening at Beyond3D. Behind the scenes, things weren’t going well, although looking back at it now, my feelings about what happened are completely different now. After doing the Beyond3D and Hexus in parallel thing for a while, things got reasonably tense between Hexus and I. I wasn’t working on Hexus as much, spending almost all my time on the thing I really wanted to be working on. In hindsight, I missed a trick in not working more closely with David and Paul to figure out how to balance my time and my output at Hexus and Beyond3D. They paid the bills, but we didn’t work together to figure out how to make that work properly, and a lot of that failing was mine. At the time, I didn’t feel like Hexus wanted to really look after Beyond3D. That phrase again.
Because it wasn’t making money and because I was spending all of my time on it, so not really helping Hexus make any money, no wonder they were frustrated with me and Beyond3D. I got the impression that David and Paul planned to significantly devalue the founding ethos of the site and do less in-depth technical analysis and more straightforward but less Beyond3D-y content. Honestly, given the landscape of making money on the web at the time, I don’t blame them in hindsight, but at the time I was angry. I promised Wavey I’d do the right thing and look after it, but I didn’t see how I could do that with Hexus involvement. Eventually, things came to a head one day and I just quit, separating Beyond3D from Hexus in late 2007.
David and I didn’t talk for a long time after that, until well into my tenure at Imagination! No wonder, given the way I just up and quit with no notice. It worked out that way because — for better or worse for both parties I’m not sure to this day — we didn’t have an informal employment contract with each other. I started at Hexus back in the day because David saw me post on the Overclockers UK forums that I had an Abit VP6 with a pair of socketed Pentium 3s, and wanted a review of the VP6 for the site. He rang me up one evening, asked if I wanted to review it and if the review was OK do some more. I said yes, back in late-ish 2000, and the rest was history. We worked together pretty much on trust. He had kit sent to me, I wrote reviews. We never formally codified that in any legal language, not even after 6 years of working together. I had no fixed salary, just a small amount amount per review and the use of the review hardware if the rest of the site didn’t need it. No contract, no salary. Just David running a business and me filing copy periodically. So when the time came to ensure, in my eyes anyway, that Beyond3D would be looked after properly, something I couldn’t see happening with Hexus any more, I just stopped answering the phone and email. There was no contract of any kind, so there was nothing they could do.
I mentioned earlier that I was really busy on Beyond3D in the run up to the working divorce with Hexus. The majority of that time wasn’t writing content, but writing a content management system for the site that would modernise its publishing ability, let the team and I collaborate on content, let us visualise data much better and generally bring the design up to scratch. Until the new CMS went live in early-ish 2007 (I think March 12th!), the website was running on a home-grown system that could barely be called a CMS. I was used to something much better at Hexus, and because Hexus couldn’t spare development time to improve or design it, or set Beyond3D up with a cut of the Hexus system, I did it myself.
In late 2006 and early 2007, web content management systems weren’t really a thing. Or at least I’ve convinced myself of that because that’s the rationale I used to write a new one. Written from scratch and underpinning a new website that I designed myself, it all still powers the site today. Of all of the software I’ve ever written in my life, Absinthe, as it was lovingly codenamed by Farid, one of the team at the time, is what I’m most proud of on balance. That’s not because it’s great code, or as robust or secure system as I’d maybe build today, but because the back-end has stood up and the front-end has worked pretty much constantly since I switched it on 8 years ago, and the site and content still renders as it did back then in old browsers. I’m a terrible web designer — just look at this site for proof of that, where there isn’t any design to speak of at all — but you can still read all of the content Beyond3D has ever published, right there in a modern browser in 2015 without having to do anything special.
I used that development experience to get a job at BT’s Plusnet ISP in October 2007. Having that full-time job at BT meant that, despite being divorced from Hexus shortly afterwards, I was financially secure, but I didn’t really have the time for Beyond3D in the way I used to. The last thing I wrote for Beyond3D was in June 2008, an architecture analysis of NVIDIA’s GT200 GPU. The state of the graphics industry at the time was one of intense competition and interesting microarchitecture, so it was really sad to not be able to look at it for a viable living. Making money on the web with adverts has always been hard, but I’ve always been particularly inept at it. The good thing is that the site doesn’t cost me a lot to keep going. For a long time we were hosted by The Tech Report’s hosts, effectively for free, via a deal we got by being associated with TR. When that host was bought out by a corporate megalith and they reviewed the accounts, we were asked to start paying or find a way off of their platform. Since then I’ve paid for it myself, moving the servers back to the UK where they’ve been for years now. It’s cheap enough to run that I don’t mind, and in many ways it’s me giving back to Beyond3D something while I’ve not been able to write for it. I look after it that way now, keeping the lights on.
For whatever reason (maybe I should ask him one day!), Alex Voicu, who was doing GPU analysis for Rage3D around the same time, came to me in mid-2008 and asked if I’d be interested in him writing more in-depth analysis for Beyond3D. He did great work for Rage3D, so in absence of me being able to do it myself, I bit his hand off and said yes. His RV740, Cypress and Fermi analyses are some of the best things we’ve ever published on the site. Alex had a PhD to do, though, and then found subsequent tenure at Microsoft where he’s been for the last year or so, finding himself without the time to write anything for Beyond3D. Sounds familiar!
We haven’t published anything of note, especially not an in-depth architecture analysis, since late 2010 when Alex took a close look at NVIDIA Fermi. So for the last 4.5 years the site’s ticked along as just the forums, with no active analysis of either modern discrete GPUs, or the rise of the embedded GPUs in today’s smartphones, tablets and other highly integrated devices powered by cutting edge SoCs. In 2011 I decided to pass the reins to Alex formally in an announced way, although on the understanding that I’d still keep the site running and look after its safekeeping. I built all of the technology the main site stands on, and I’ve done all of the sysadmin stuff since the first day Wavey handed me the keys, so it made sense for me to keep doing all of that while Alex was in charge of content. It let me look after the site and feel like I was still a part of it, while working on graphics again for a living at PowerVR. So I’ve kept the lights on ever since.
The classic career path of a Beyond3D Editor-in-Chief is to go work for a GPU hardware vendor eventually. Dave Barron was at 3Dfx. Kristof Beets is now my boss at Imagination! Dave Baumann went to AMD, née AMD. So, not keen to leave the UK but completely sick of working for an ISP and still desperate to do graphics for a living, I interviewed with Imagination in late 2009, and in very early 2010 I started work on PowerVR Graphics in the Developer Technology group doing part demos, part competitive analysis. I helped spin competitive analysis out into its own team shortly after, going to work for Kristof and founding what’s now a flourishing engineering group doing a bunch of stuff, mostly competitive and performance analysis, deep within PowerVR Graphics.
In recent times we’ve jointly discussed getting Beyond3D back on the publishing-about-discrete-GPU-microarchitecture horse, but it’s felt dishonest to do so without making certain changes to the site and forums first. The forums are what can potentially pay enough of Beyond3D’s bills, so they needed love first and foremost. After much wailing and gnashing of teeth, sitting on beta migrations for almost a year, I pulled the trigger on a move to new forum software in November last year. I’ll revamp the forum structure in the next couple of months, to reinvigorate the populace and make prominent and happy the growing and more fertile areas of discussion in order to foster them properly and make sure they can grow. We’ll change how threads are handled and how moderation works, not massively but enough to set a slightly new tone. After that, I’ll work on a new CMS, but certainly not one of my own making or design, so we can write new stuff again.
That’s one of the great things about the site, really. Despite me not writing anything for over 6 years and Alex for over 4, we have ideas for great content we could write, that would be real Beyond3D-specific content and not anything you could find elsewhere. Stuff on boring old discrete GPUs that would have impact. If we can find a great way to look at mobile architectures without breaking NDAs and things, we could write about things that nobody else has a hope of uncovering. The stuff that I wanted to write back when I thought Beyond3D was going to be my life for a long time, now fuelled by actually having some idea about how GPUs actually work, not just how the public is deigned to understand them.
So while it’s languished for more than a few years now, the forums are healthy and we will have another go at adding to what’s published about GPU architectures on the Web. I’m a strong believer that the current way of funding independent publishing on the Web is fundamentally changing, and I’d like Beyond3D to have a go at proving me right. It’s exciting to think back about it so far, and wonder where it might go in the future with a bit of luck.