I've told this story before on HN, but my biz partner at ArenaNet, Mike O'Brien (creator of battle.net) wrote a system in Guild Wars circa 2004 that detected bitflips as part of our bug triage process, because we'd regularly get bug reports from game clients that made no sense.
Every frame (i.e. ~60FPS) Guild Wars would allocate random memory, run math-heavy computations, and compare the results with a table of known values. Around 1 out of 1000 computers would fail this test!
We'd save the test result to the registry and include the result in automated bug reports.
The common causes we discovered for the problem were:
- overclocked CPU
- bad memory wait-state configuration
- underpowered power supply
- overheating due to under-specced cooling fans or dusty intakes
These problems occurred because Guild Wars was rendering outdoor terrain, and so pushed a lot of polygons compared to many other 3d games of that era (which can clip extensively using binary-space partitioning, portals, etc. that don't work so well for outdoor stuff). So the game caused computers to run hot.
Several years later I learned that Dell computers had larger-than-reasonable analog component problems because Dell sourced the absolute cheapest stuff for their computers; I expect that was also a cause.
And then a few more years on I learned about RowHammer attacks on memory, which was likely another cause -- the math computations we used were designed to hit a memory row quite frequently.
Sometimes I'm amazed that computers even work at all!
Incidentally, my contribution to all this was to write code to launch the browser upon test-failure, and load up a web page telling players to clean out their dusty computer fan-intakes.
I remember one of the first impressions I had in GW1 during test events was the sense of scale in the world that still managed to avoid excessive harsh geometry angles for the most part. Not surprised to hear it was pushing more polygons than average.
P.S. GW1 remains one of my favorite games and the source of many good memories from both PvP and PvE. From fun stories of holding the Hall of Heroes to some unforgettable GvG matches, y'all made a great game.
As a mobile dev at YouTube I'd periodically scroll through crash reports associated with code I owned and the long tail/non-clustered stuff usually just made absolutely no sense and I always assumed at least some of it was random bit flips, dodgy hardware, etc.
For the Mastodon Android app, I also sometimes see crashes that make no sense. For example, how about native crashes, on a thread that is created and run by the system, that only contains system libraries in its stack trace, and that never ran any of my code because the app doesn't contain any native libraries to begin with?
Unfortunately I've never looked at crashes this way when I worked at VKontakte because there were just too many crashes overall. That app had tens of millions of users so it crashed a lot in absolute numbers no matter what I did.
GW1 was my childhood. The MMO with no monthly fees appealed to my Mom and I met friends for years. The 8 skill build system was genius, as was the cut scenes featuring your player character. If there's ever a 3rd game I would love to see something allowing for more expression through build creation though I could see how that's hard to balance.
The PvP was so deep too. You would go 4v4 or 8v8 and coordinate a “3, 2, 1 spike” on a target so that all your damage would arrive at the same time regardless of spell windup times and be too much for the other team’s healer to respond to.
Could also fake spike to force the other team’s healer to waste their good heal on the wrong player while you downed the real target. Good times.
I still remember summoning flesh golems as a necromancer! Too much of my life sunk into GW1. Beat all 4(?) expansions. Logged in years later after I finally put it down to find someone had guessed my weak password, stole everything, then deleted all my characters. C'est la vie.
I don't understand why ECC memory is not the norm these days. It is only slightly more expensive, but solves all these problems. Some consumer mainboards even support it already.
I’ve had plenty of servers with faulty ecc dimms that didn’t trigger , and would only show faults when actual memory testing. I had a hard time convincing some of our admins the first time ( ‘no ecc faults you can’t be right ‘ ) but I won the bet.
Edit: very old paper by google on these topics. My issues were 6-7 years ago probably.
That shouldn’t make sense. It’s not like the ECC info is stored in additional bits separate from the data, it’s built in with the data so you can’t “ignore” it. Hmm, off to read the paper.
Well for DDR5 that's 25% more chips which isn't great even if you don't get ripped off by market segmentation.
It's possible DDR6 will help. If it gets the ability to do ECC over an entire memory access like LPDDR, that could be implemented with as little as 3% extra chip space.
Also, in a game, there is a tremendously large chance that any particular bit flip will have exactly 0 effect on anything. Sure you can detect them, but one pixel being wrong for 1/60th of a second isn't exactly ... concerning.
The chance for a bit flip to affect a critical path that is noticeable by the player is very low, and quite a bit lower if you design your game to react gracefully. There's a whole practice of writing code for radiation hardened environments that largely consists of strategies for recovering from an impossible to reach state.
For safety critical systems, one strategy is to store at least two copies of important data and compare them regularly. If they don't match, you either try to recover somehow or go into a safe state, depending on the context.
You can have voting systems in place, where at least 2 out of 3 different code paths have to produce the same output for it to be accepted. This can be done with multiple systems (by multiple teams/vendors) or more simply with multiple tries of the same path, provided you fully reload the input in between.
Interesting, I was not aware! Do you have a statistics for the bit flips in RAM %? My feeling would be its the majority of bit flips that happen, but I can be wrong.
It would be quite hard to gather that data and would be highly dependent on hardware and source of bit flip.
But there's volatile and nonvolatile memory all over in a computer and anywhere data is in flight be it inside the CPU or in any wires, traces, or other chips along the data path can be subject to interference, cosmic rays, heat or voltage related errors, etc.
It should be fairly easy to see statistically if ECC helps, people do run Firefox on it.
The number of bits in registers, busses, cache layers is very small compared to the number in RAM. Obviously they might be hotter or more likely to flip.
> And then a few more years on I learned about RowHammer attacks on memory, which was likely another cause -- the math computations we used were designed to hit a memory row quite frequently.
For that one I'd guess no, because under normal circumstances hot locations like that will stay in cache.
Thanks to asrock motherboards for AMD’s threadripper 1950x working with ECC memory, that’s what I learned to overclock on.
I eventually discovered with some timings I could pass all the usual tests for days, but would still end up seeing a few corrected errors a month, meaning I had to back off if I wanted true stability. Without ECC, I might never have known, attributing rare crashes to software.
From then on I considered people who think you shouldn’t overlock ECC memory to be a bit confused. It’s the only memory you should be overlocking, because it’s the only memory you can prove you don’t have errors.
I found that DDR3 and DDR4 memory (on AMD systems at least) had quite a bit of extra “performance” available over the standard JEDEC timings. (Performance being a relative thing, in practice the performance gained is more a curiosity than a significant real life benefit for most things. It should also be noted that higher stated timings can result in worse performance when things are on the edge of stability.)
What I’ve noticed with DDR5, is that it’s much harder to achieve true stability. Often even cpu mounting pressure being too high or low can result in intermittent issues and errors. I would never overclock non-ECC DDR5, I could never trust it, and the headroom available is way less than previous generations. It’s also much more sensitive to heat, it can start having trouble between 50-60 degrees C and basically needs dedicated airflow when overclocking. Note, I am not talking about the on chip ECC, that’s important but different in practice from full fat classic ECC with an extra chip.
I hate to think of how much effort will be spent debugging software in vain because of memory errors.
Similar experience. I played with overclocking the DDR5 ECC memory I have on my system, it would appear to be stable and for quite a while it would be. But after a few days I'd notice a handful of correctable errors.
I now just run at the standard 5600MHz timing, I really don't find the potential stability trade off worth it. We already have enough bugs.
If you look around you'll see people already putting the new, chinese made DDR4 through its paces, it's holding up far better than anyone expected.
Every single time I've had someone pay me to figure out why their build isn't stable, it's always some combination of cheap power supply with no noise filtering, cheap motherboard, and poor cooling. Can't cut corners like that if you want to go fast. That is to say, I've never encountered "almost ok" memory. They're quite good at validation.
The danger is we’ll start to see more QA rejects coming into the market. The temptation to mix in factory rejects into your inventory is going to get very high for a lot of resellers.
DDR4 and 5 both have similar heat sensitivity curves which call for increased refresh timings past 45C.
Some of the (legitimately) extreme overclockers have been testing what amounts to massive hunks of metal in place of the original mounting plates because of the boards bending from mounting pressure, with good enough results.
On top of all of this, it really does not help that we are also at the mercy of IMC and motherboard quality too. To hit the world records they do and also build 'bulletproof', highest performance, cost is no object rigs, they are ordering 20, 50 motherboards, processors, GPUs, etc and sitting there trying them all, then returning the shit ones. We shouldn't have to do this.
I had a lot of fun doing all of this myself and hold a couple very specific #1/top 10/100 results, but it's IMHO no longer worth the time or effort and I have resigned to simply buying as much ram as the platform will hold and leaving it at JEDEC.
> From then on I considered people who think you shouldn’t overlock ECC memory to be a bit confused. It’s the only memory you should be overlocking, because it’s the only memory you can prove you don’t have errors.
This attitude is entirely corporate-serving cope from Intel to serve market segmentation. They wanted to trifurcate the market between consumers, business, and enthusiast segments. Critically, lots of business tasks demand ECC for reliability, and business has huge pockets, so that became a business feature. And while Intel was willing to sell product to overclockers[0], they absolutely needed to keep that feature quarantined from consumer and business product lines lest it destroy all their other segmentation.
I suspect they figured a "pro overclocker" SKU with ECC and unlocked multipliers would be about as marketable as Windows Vista Ultimate, i.e. not at all, so like all good marketing drones they played the "Nobody Wants What We Aren't Selling" card and decided to make people think that ECC and overclocking were diametrically supposed.
[0] In practice, if they didn't, they'd all just flock to AMD.
>[0] In practice, if they didn't, they'd all just flock to AMD.
only when AMD had better price/performance, not because of ECC. At best you have a handful of homelabbers that went with AMD for their NAS, but approximately nobody who cares about performance switched to AMD for ECC ram, because ECC ram also tend to be clocked lower. Back in Zen 2/3 days the choice was basically DDR4-3600 without ECC, or DDR4-2400 with ECC.
At the beginning of your comment I was wondering if the "attitude" that was corporate serving was the anti-ECC stance or the pro-ECC stance (based on the full chunk that you quoted). I'm glad that by the end of the comment you were clearly pro ECC.
Any workstation where you are getting serious work done should use ECC
As a community alpha tester of GW1, this was a fun read! Such an educational journey and what a well organized and fruitful one too. We could see the game taking shape before our eyes! As a European, I 100% relied on being young and single with those American time zones. :D Tests could end in my group at like 3 am, lol.
Oh yeah, those were some good times. It was great getting early feedback from you & the other alpha testers, which really changed the course of our efforts.
I remember in the earlier builds we only had a “heal area” spell, which would also heal monsters, and no “resurrect” spell, so it was always a challenge to take down a boss and not accidentally heal it when trying to prevent a player from dying.
Every interesting bug report I've read about Guild Wars is Dwarf Fortress tier. A very hardcore, longtime player who was recounting some of the better ones to me shared a most excellent one wrt spirits or ghosts, some sort of player summoned thing that were sticking around endlessly and causing OOM errors?
That's a really cool anecdote. The overclock makes sense. When we released Need For Speed (2015) I spent some time in our "war room", monitoring incoming crash reports and doing emergency patches for the worst issues.
The vast majority of crashes came from two buckets:
There's a famous Raymond Chen post about how a non-trivial percentage of the blue screen of death reports they were getting appeared to be caused by overclocking, sometimes from users who didn't realize they had been ripped off by the person who sold them the computer: https://devblogs.microsoft.com/oldnewthing/20050412-47/?p=35.... Must've been really frustrating.
This was a design choice by AMD at the time for their Athlon Slot A cpus. Use the same slot A board which you could set the cpu speed by bridging a connections. Since the Slot A came in a package, you couldn't see the actual cpu etching. So shady cpu sellers would pull the cover off high speed cpus, and put them on slow speed cpus after overclocking them to unstable levels.
Wow, that’s really interesting! I always suspected bit flips happened undetected way more than we thought, so it’s great to get some real life war stories about it. Also thanks for Guild Wars, many happy hours spent in GW2. :)
Some multiplayer real-time strategy (RTS) games used deterministic fixed-point maths and incremental updates to keep the players in sync. Despite this, there would be the occasional random de-sync kicking someone out of a game, more than likely because of bit flips.
For RTS games I wish we could blame bit flips, but more typically it is uninitialized memory, incorrectly-not-reinitialized static variables, memory overwrites, use-after-free, non-deterministic functions (eg time), and pointer comparisons.
God I love C/C++. It’s like job security for engineers who fix bugs.
I kind of wanted to confirm that. At that time I was still using a Compaq business laptop on which I played Guild Wars.
The Turion64 chipset was the worst CPU I've ever bought. Even 10 years old games had rendering artefacts all over the place, triangle strips being "disconnected" and leading to big triangles appearing everywhere. It was such a weird behavior, because it happened always around 10 minutes after I started playing. It didn't matter _what_ I was playing. Every game had rendering artefacts, one way or the other.
The most obvious ones were 3d games like CS1.6, Guild Wars, NFSU(2), and CC Generals (though CCG running better/longer for whatever reason).
The funny part behind the VRAM(?) bitflips was that the triangles then connected to the next triangle strip, so you had e.g. large surfaces in between houses or other things, and the connections were always in the same z distance from the camera because game engines presorted it before uploading/executing the functional GL calls.
After that laptop I never bought these types of low budget business laptops again because the experience with the Turion64 was just so ridiculously bad.
Did you/he ever consider redundant allocation for high value content and hash checks for low value assets that are still important?
I imagine the largest volume of game memory consumption is media assets which if corrupted would really matter, and the storage requirement for important content would be reasonably negligible?
I think the most reasonable take would be to just tell the users hardware is borked, they're going to have a bad outside the game too, and point them to one of the many guides around this topic.
I don't think engineering effort should ever be put into handling literal bad hardware. But, the user would probably love you for letting them know how to fix all the crashing they have while they use their broken computer!
To counter that, we're LONG overdue for ECC in all consumer systems.
I put engineering effort into handling bad hardware all the time because safety critical, :)
It significantly overlaps the engineering to gracefully handle non-hardware things like null pointers and forgetting to update one side of a communication interface.
80/20 rule, really. If you're thoughtful about how you build, you can get most of the benefits without doing the expensive stuff.
I think I sit in another camp. A lot of my engineering efforts are in working around bad hardware.
Better the user sees some lag due to state rebuild versus a crash.
Most consumers have what they have, and use what they have. Upgrading everything is now rare. If they got screwed, they'll remain screwed for a few years.
That's an interesting idea. How might you implement that? Like RAID but on the level of variables? Maybe the one valid use case for getters/setters? :)
As another user fairly pointed out, ECC. But a compiler level flag would probably achieve the redundancy, sourcing stuff from disk etc would probably still need to happen twice to ensure that bit flips do not occur, etc.
Every frame (i.e. ~60FPS) Guild Wars would allocate random memory, run math-heavy computations, and compare the results with a table of known values. Around 1 out of 1000 computers would fail this test!
We'd save the test result to the registry and include the result in automated bug reports.
The common causes we discovered for the problem were:
- overclocked CPU
- bad memory wait-state configuration
- underpowered power supply
- overheating due to under-specced cooling fans or dusty intakes
These problems occurred because Guild Wars was rendering outdoor terrain, and so pushed a lot of polygons compared to many other 3d games of that era (which can clip extensively using binary-space partitioning, portals, etc. that don't work so well for outdoor stuff). So the game caused computers to run hot.
Several years later I learned that Dell computers had larger-than-reasonable analog component problems because Dell sourced the absolute cheapest stuff for their computers; I expect that was also a cause.
And then a few more years on I learned about RowHammer attacks on memory, which was likely another cause -- the math computations we used were designed to hit a memory row quite frequently.
Sometimes I'm amazed that computers even work at all!
Incidentally, my contribution to all this was to write code to launch the browser upon test-failure, and load up a web page telling players to clean out their dusty computer fan-intakes.