Epic’s Tim Sweeney has been advocating AMD’s 64-bit platform since its official unveiling at Comdex last November. With this in mind, Firingsquad decided to ask Tim about Epic’s 64-bit UT2K3 port, Epic’s 64-bit plans for their next generation Unreal engine, and his thoughts on the Opteron architecture.
Tim Sweeney 64-bit Interview
Submitted by Sagres 2003-04-27 Games 16 Comments
Just goes to show that the 2Gig limit on address space is real and impacting applications already and that 64 bit on the desktop is needed now.
Many OSes struggle with accessing more than 1gig of ram due to address space cram and yet those same OSes demand at least 1 Gig of ram to do their job well to deal with the habits of the typical user. I have observed the work habits of my staff over the years, and the dominant habit is to have as many applications open on the desktop as possible. Recent applications are becoming memory hungry and so the demands on large memory machines are increasing.
While it’s theoretically possible to squeeze more RAM in, the work to do so is less than trivial and potentially results in a performance hit. It is simpler and more logical to switch to a larger address space CPU.
Given these arguments, I think that 64 bit CPUs have more going for them than just the esoteric large applications that struggle with a 32 bit address space, and I expect that we’ll see a good uptake on the Athlon 64 when it arrives for these very reasons.
I for one just want to see x86 die
640K should be enough for anybody.
the improvements amd made in amd64 are the next best thing to RISC
I’m fairly sure the 32bit memory addressing limit is 4GB, or 2^32 bits.
The bottom line of Sweeney’s argument is that development of content for a higher fidelity of output requires a 64 bit address space. Despite vast improvements of graphics fidelity over the past decade, games are still fairly crude affairs and have yet to make significant advances in areas that are arguably more important. Not only that, I’m not convinced that the software solutions to the content creation problems haven’t been explored well enough, giving rise to the suspicion that 64 bits is a lazy mans optimisation. In conclusion, I’d say the move to 64 bits hasn’t been justified, and wonder how much Sweeney is being paid to say what he’s saying.
what current or *almost ready* solution do you suggest to take the place of x86?
you see, seaslug, the seaslug was an incredibly efficient sea creature (it’s also SeaSLUG, a somewhat redundant acronym for Seattle Local Unix Group)…in many cases discarding it’s shell, evolving to the point where it was more efficient for mating and absorbtion of nutrients sans shell.(efficient much like the powerpc, sparc, alpha, mips or any other cpu that an elitist little x86 hater like you might prefer).
but a distant cousin to the seaslug, the cephalopods seem to be giving the efficient seaslug a run for their money. Despite having redundant appendages, some might say useless extraneous body parts, mating routines that often require the being to die shortly afterwards, and finicky appetites…cephalopods seem quite versatile and capable, in the end showing all non x86 architectures, i mean non-cephalopods it’s nimbleness, and proficiency in the sea.
is it possible, just possible that x86 exists for a reason?
affordable information processing & computing available to more socio-economic classes then any other cpu?
cost effective powerful systems so good that universities are building large clusters around them?
i think seaslug is not a slug at all.
but a troll.
maybe seaslug should quit trying to grow legs and go back to the depths of the ocean.
x86 is dead. long live x86.
Charles, I dunno if he is getting paid, but I like a bit of healthy scepticism
I want to see something jaw-droppingly good. And I want to be told it’s only possible with 64bit computing. Until then I’ll wait.
I thought that he made a good point for the kind of task that content-creation took, and the options available: use diskspace, addressing extensions, or AMD64; and he said that AMD64 was the easiest route. Seems like a good reason.
Personally I don’t play FPS, but loads of people do. You can have FPS and have some story, e.g. halflife.
Every time I see one of those “64-bit” articles here on OSNews, and especially while reading this interview, I find myself asking, “Why?”.
Not “Why 64 bits?”, there are probably some people out there needing that. But rather, “So why did you wait for such a long time?”. When was the R4400 released? Ten years ago. The Alpha? Dito. And any processor made after the seventies will offer more than enough registers to make any Epic programmer happy.
Now, given another ten years, the PC world may have made yet another giant stride, finally realising the benefits of such things as OpenFirmware or function keys.
I suspect cost, compatibility and market want/need were big obstacles for implementation in mainstream hardware.
(And even then Athlon 64 won’t be out before September…)
yes, 4gb is the limit, but not for user space applications. most OSes use half the vm space for kernel land and the other half for user space
so a user space app can only see 2 GB (this is most of the time, yes there are exceptions)
My feeling is that another 8 bits of address space would be enough to solve most address limit problems for the next 10 or 20 years (at least, maybe longer). However, 5-byte 40-bit data moves are somewhat odd, so no vendor seems to be going in that direction. 64-bit addressing is probably overkill to some extent, but there is certainly a problem with the current 32-bit limitations.
IIRC, the current AMD 64-bit CPUs only use 48-bit addressing externally. Even with that, I believe the memory controllers are limited to 64GB of RAM. But that leaves plenty of room to expand in the forseeable future.
> I’m fairly sure the 32bit memory addressing limit is 4GB, or 2^32 bits.
Yeah, I know. I thought that might be interpreted wrong. I’m talking more about the 2 Gig limit that most OSes force on user level address space. As a result of this, the kernel is also constrained to 2 gig of address space to itself as well. When you start splitting it up into arenas, you run out of address space pretty quick.