X-Box hacker interview

Fafalada, could you answer me some questions about ps2?

Does vu1 have write access to SPRAM (presumably not)?

What is the machines physical address length (24bit?)/ memory wordsize(prob. 64bit as the caches)?

having a 3.2GB/sec memory sub-system on a 2GB/sec bus seems to be a strange design desicion, as all kinds of other data(e.g. IPU --> GIF, SPRAM --> VU1) also have to use that pipe, esp. when taking into consideration the near absence of caches in EE.

The edram seems to have two access ports for the framebuffer (1024bit read & write..). Does this mean you can interleave fetches and writes to the framebuffer (how long is edrams access latency then??)?

 

My mistake, so it is legal as a large company to subsidize your products with income from other areas to drive out competitors as long as you do it to the same extent in all markets? :/

No, bottom line for the producer (loss) and bottom line for the consumer (gain).

If I tried to sell that idea to investors they would not agree.

Xbox might end up being good for m$, but that doesnt mean their strategy is particularely good for us ... benefiting the consumer is a mere side effect of the open market, and not one you can always depend on.

Marco

PS. saying consoles with embedded framebuffers, in addition to seperate main memory, use UMA seems disingenuous.

 

I wasn't aware that you could sell the same product in different markets in the same region. Would seem to cancel itself out, maybe I'm crazy.

 

Not geographical areas, areas of business.

 

i can speak for what happened here

first they gave two games for each people who bought the xbox at its original price.

from the 15-6 to the 15-9, they gave two games to the people who bought a xbox and a gamepad. operation "coup double".

and of course packs with two games are coming for christmas..

MS have other activities, very lucrative.. they can do this type of bizness as long as they want. in the worst case, they just have to increase prices of software they sell to their captive customers, as they did already many time.. (see the gigantic prise increases for the price of the OS or the office suite..)

other companies like nintendo can not rely on a cash cow to live, they have to live with the software they sell.

yes microsoft dumping the xbox console and software could be seen as a good thing for the consumer, but it would only be short-term thinking.

in fact it can cause a great damage for the sector, to the other console manufacturer and games publishers that have to bear this unfair concurrence. (who have no racket to finance them)

if they succeed in driving their competition out, there will be no more "gifts".

 

How so?

 

I think all of you MS detractors have just proven the point.

Bunnie slams the technology, but he's really just slamming MS, just like many people here.

The Xbox is nice enough hardware. It's good value for consumers and developers. As a result, we Xbox gamers buy a lot of software so it all works out.

 

Granted, yes, the chips are identical. However the implementation is very different.

Thing is, the smallest RIMMs you can get are 64MB, and they have four cores each. Add to that the distance from the socket to the Northbridge and you have some built-up latency simply due to the signal's travel time.

One thing DRDRAM controllers have to do at startup is 'ping' the channel to find the latency to the last chip in line, to make sure all accesses have that same delay... so everything stays in order - if you have 16 chips, say your addresses are on chip 1, then chip 16, then chip 2 - they're going to come back in the wrong order, chip 1, then chip 2, then chip 16, because the signal takes less time to reach chips 1 and 2.

On the PS2, I don't know if you've ever seen it, but the system uses two 400MHz channels (just like a dual-channel PC800 PC implementation), but each of those channels has exactly one chip... and the chips are mere centimetres away from the EE CPU, making actual access latency near nothing.

There's still some address latency, as per DRDRAM's actual specs, but the signal latency is all but gone, plus there is no need for any wait states on PS2.

 

Thanks for dropping the "goods", Tagrineth! You are teh cool! 8)

 

So what is the actual latency?

 

Lower than normal DRDRAM, but I don't know of any straight measurements offhand. Fafalada could probably cite a few numbers...

And actually dual-channel DRDRAM on dual Xeons in a recent shootout at 2CPU had LOWER latency than single-channel DDR with dual AthXP's. Funfunfun.

 

MS didn't, although some retailers did. I actually never saw one of the forced bundles in a B&M although I know some of them did partake in it.

 

Interesting. Though dual processor anything changes the latency.

The xbox's memory architecture is dual channel DDR @ 200(400) mhz.

In PC equivalents it would roughly be dual-channel PC3200.

It also most likely has similar optimizations as a GeForce video card or any other non-expandable memory bus would have -- like PS2, since it is non-expandable, I bet you can run the timings much tighter than on a real PC.

 

I know what you mean Mfa, and wholeheartedly agree. Any company who can sell substantially below cost and cover its' losses with earnings from a seperate sector (or an entirely seperate division of the company), strikes me as a dangerously anti-competitive practice (though in the short term makes consumers happy by driving down prices).

 

neither bunnie or somebody told here that the xbox was not a good console to buy and was not enjoyable or was a bad console to develop for..

of course, as soon as we dare criticizing MS and/or its product we have to be some religious fanatic MS-haters, isn't it ?
how could it be possible for somebody with an healthy mind to do such things ?
i wonder..

you have just proved that you are immunized against the reality.

you are biased enough to be blind, like these race horses that wear blinkers..

you prefer to believe in PR garbage than in the sad reality.

 

I'm not Faf, but...

Not directly... But data from VIF1 can be DMAd to SPRAM

Physical address length is 32bits. Word length is also 32bits. As for the caches, the line size is 64 bytes.

Having a little extra memory bandwidth just helps to assure data can be queued up as quickly as possible when bus arbitration causes additional latency.

Near absence? There's like 7 caches on-chip (80KB). Since many of the execution resources are rather discrete it makes sense to have cache pools distributed so each device can operate on data without fighting over the bus.

Max row access is 45ns, a transfer of 2 bytes per/beat (1.25ns) or a qword (16 bytes) every 10ns, and it takes 4 bus cycles (8 CPU) to fill a cacheline (4 qwords)...

Not necessarily, if you're talking latency to the CPU. When processing data on several processors with a high locality of reference, good cache snooping can kill a lot of latency. Of course on point to point topologies, it doesn't really work.

It's a wee bit more complex than that. It doesn't have much to do with expandability (well yeah in the case of Rambus it does, but not so much with DDR-RAM).

The thing is there is no N-bridge controller to reach out of the chip to fetch memory addresses.

Also, controller optimizations that work for a GeForce don't necessarily translate well to a CPU. It works for a GeForce because that's the only device it has to work with, so you can tweak the memory controller for wide alignments for blitting large tracts of predicated data around. However the XGPU has to provide support for not only the GPU portion of itself, but also MCPX and the CPU, each with their own demands. This can make itself visible with regards to address granularity for one (think cache pollution).

 

antitrust laws are here to prevent this kind of situation and ensure the market remains a free market.

 

That indicates any company that turns a decent profit. Perhaps we aren't seeing an anti MS slant here, simply anti capitalism?

 

I have not read such a enlightening interview in eons.

Give the guy a break, there's no way he can back up all his arguements with counter arguments and couters for those; an arguement only goes so deep. Right now, those of you who are flaming him are flaming a person who can't respond, makes you feel good doesn't it?