Enter the 360 beta kits

[PC-Engine]

I get that they are pissed off that its not x86 and can't be lazy .

Spoken like a true non-programmer who hasn't seen the XCPU specs.

The XCPU is slower for FLOPs than the G5s from the previous alpha kit, and IIRC doesn't support out of order instructions - forcing the developer to worry about optimising that rather than letting the CPU.

You may call this "lazy", but until you've seen how god-awful it can be to optimise code, especially multi-threaded code, you may want to hush up.

Slower for FLOPS? Care to elaborate?

OOO can be taken care of with compilers...

 

[jvd]

Mech . Read the article. One dev clearly states lazy code wont work well on it and they will have to clean up the code

 

[mech]

I did read the article... "lazy code" is an already established term. Saying the devs are just pissed because they can't be lazy and cos it's not x86 is a totally different kettle of fish.

The XCPU is dramatically stripped down compared to a G5, and although is theoretically more powerful (in peak performance), it takes a lot to get this performance out of it. Even Microsoft knows and admits this.

 

[PC-Engine]

The CPU is capable of 115 GFLOPS. The dual G5 is capable of less than 50GFLOPS total IIRC. In order CPUs will require more attention to get maximum performance, but I don't see how it's lower in GFLOPS...

 

[mech]

I was talking about how quickly it executes floating point code, not how many FLOPs it can do in a second.

From Anandtech's yanked article:

To put it in perspective, floating point multiplies are apparently 1/3 as fast on Xenon as on a Pentium 4.

The reason for the poor performance? The very narrow 2-issue in-order core also happens to be very deeply pipelined, apparently with a branch predictor that’s not the best in the business.

Also, take that 115GFLOP figure with a HUGE grain of salt - Microsoft's been steadily revising it upwards as they figure out more ways to inflate it (without the hardware actually changing). A more accurate figure would probably be around the 70-90GFLOP mark. But it's like MIPS - totally useless to predict real world performance, because there are so many other variables that influence the final speed. Having shit branch prediction for instance.

 

[PC-Engine]

I was talking about how quickly it executes floating point code, not how many FLOPs it can do in a second.

From Anandtech's yanked article:

To put it in perspective, floating point multiplies are apparently 1/3 as fast on Xenon as on a Pentium 4.

The reason for the poor performance? The very narrow 2-issue in-order core also happens to be very deeply pipelined, apparently with a branch predictor that’s not the best in the business.

Also, take that 115GFLOP figure with a HUGE grain of salt - Microsoft's been steadily revising it upwards as they figure out more ways to inflate it (without the hardware actually changing). A more accurate figure would probably be around the 70-90GFLOP mark. But it's like MIPS - totally useless to predict real world performance, because there are so many other variables that influence the final speed. Having shit branch prediction for instance.

And from that you come to the conclusion that the 3 core CPU will perform worse than the dual G5s in the majority of game apps? So MS went backwards with the CPU? You do realize that higher clocks mean less ops per clock right? You do realize that 3 cores are more than 2 cores right? You do realize that the 3 cores are clocked higher than the 2 cores right?

 

[mckmas8808]

I was talking about how quickly it executes floating point code, not how many FLOPs it can do in a second.

From Anandtech's yanked article:

To put it in perspective, floating point multiplies are apparently 1/3 as fast on Xenon as on a Pentium 4.

The reason for the poor performance? The very narrow 2-issue in-order core also happens to be very deeply pipelined, apparently with a branch predictor that’s not the best in the business.

Also, take that 115GFLOP figure with a HUGE grain of salt - Microsoft's been steadily revising it upwards as they figure out more ways to inflate it (without the hardware actually changing). A more accurate figure would probably be around the 70-90GFLOP mark. But it's like MIPS - totally useless to predict real world performance, because there are so many other variables that influence the final speed. Having shit branch prediction for instance.

And from that you come to the conclusion that the 3 core CPU will perform worse than the dual G5s in the majority of game apps? So MS went backwards with the CPU? You do realize that higher clocks mean less ops per clock right? You do realize that 3 cores are more than 2 cores right? You do realize that the 3 cores are clocked higher than the 2 cores right?

Look PC-Engine the best thing I can say is not listen to this guy. It seems obvious to me that he finds a need to put down the X360's power. He will soon wake up when the system is released.

 

[Mefisutoferesu]

Nah, I think the guys's background is very very grounded in PCs is all. So he's not particularly thrilled with the new architecture changes and what that entails. Anand wasn't neither is he.

Though I do wonder how he's come to the conculsion that MS inflated the FLOPs rating of the XeCPU? THere really isn't any fixed function FLOPs on CPUs, so... or is there?

 

[PC-Engine]

Nah, I think the guys's background is very very grounded in PCs is all. So he's not particularly thrilled with the new architecture changes and what that entails. Anand wasn't neither is he.

Though I do wonder how he's come to the conculsion that MS inflated the FLOPs rating of the XeCPU? THere really isn't any fixed function FLOPs on CPUs, so... or is there?

ERP already explained how MS came up with the 115GFLOPS figure.

 

[mech]

And from that you come to the conclusion that the 3 core CPU will perform worse than the dual G5s in the majority of game apps? So MS went backwards with the CPU? You do realize that higher clocks mean less ops per clock right? You do realize that 3 cores are more than 2 cores right? You do realize that the 3 cores are clocked higher than the 2 cores right?

Hey PC-Engine, if you can point out where in my posts you've proved me wrong, go ahead. You've totally misinterpreted what I wrote. The fact is that clock for clock, core for core, the XCPU has been made SLOWER than G5 dev kits for a bunch of common operations. HOWEVER, due to clock speed increases, a faster vector engine, and a third CPU, it's theoretically faster. But for devs writing general purpose code it's not.

My point was that the developer has to go to a lot of effort to squeeze out the performance Microsoft has stated for the XCPU. Being a programmer, I realise how much that can suck, and that it's totally harsh to call a developer "lazy" for being less than thrilled about the situation. Because rather than just worrying about getting things working (which can be difficult in itself), they have to worry about the nitty gritty shit that was meant to have been left behind in the 80s, but with the extra complexity of handling 3 cores. FUN.

Anyway, there's a great article on Arstechnica about this kind of debate here.

PC-Engine, you guys can argue and flame all you like against me, but I've posted nothing but cold hard facts. If you don't want to accept that, there's nothing I can do for you. I'm not bagging out the XCPU at all, are you guys that insecure about your console of choice that you have to intepret everything as an attack?

 

[pc999]

But consoles had almost always been in order, most of consoles dev should be used to do in order code, PC(OoO) dev should worry more because they are not, anyway even in the article a dev say that if you follow the rules you should be able to get the performace.

Now the more work thing, it really will be needed...

 

[PC-Engine]

Sounds like a strawman argument to me. You need to work harder period, whether it's Xbox 360 or PS3. It's not exactly a mystery. Bottom line is the final CPU is more powerful than the dual G5s when programmed correctly. That's life, nobody said it would be easy.

 

[mech]

Apology accepted PC-E

 

[PC-Engine]

Apology accepted PC-E

If you look at your first post in this thread, it offers little of anything so don't be surprised if people can't predict what's in your little head.

 

[mech]

Apology accepted PC-E

If you look at your first post in this thread, it offers little of anything so don't be surprised if people can't predict what's in your little head.

Offers little of anything?

What part of "it's slower for floating point operations and doesn't support out of order execution so stop ragging on the programmers for being lazy" didn't you get?

You just cannot admit being wrong.

 

[Sonic]

This thread was locked because PC-Engine stated "If you look at your first post in this thread, it offers little of anything so don't be surprised if people can't predict what's in your little head." in regard to the poster mech.

Threaqd locked until action is taken.