Hyperthreading in a console environment

[zurich]

I made a post about this earlier, but it got burried in a thread somewhere..

Basically wondering how much of a benefit HyperThreading (SMT) would be in a console environment, where there wouldn't be any heavy resource sharing/OS overhead like we currently see in Windows. HT has shown some very impressive gains for FPU intensive stuff like 3ds.. but would this be a 'must have feature' on their XCPU2 for MS?

How far 'to the metal' can a developer code, to take advantage of HT?

 

[Sonic]

zurich, you have presented a very good question. As a quick and dirty answer to the question would be to wait and see how PC games react to HyperThread (SMT). I haven't seen any examples myself as I haven't really looked lately, if anyone can post a link to info or an article about it then it may prove useful for this question.

But to answer it in the long way would have to see what devs do on next gen consoles. I'm sure with the paralellity of the PS3 it might be able to take advantage of it. You know, if you have Cells not being used to their full potential you put in some HT and bang you get extra performance. Then again, the PS3 being so paralell may not need SMT at all, but I am sure IBM will have a solution that works.

For Xbox2 I would see that getting good use of HyperThreading. I will assume it will run an Intel chip, and this that being the case and HT helping to improve in floating point performance I would say there is a good chance it could improve performance quite nicely. (It could also be a nice marketing kick for MS in the next generation, but that's besides the point.) Again, I would have to look at data coming from PC games to make a judgement call on HT if it would improve a game. And it would have to be from a game that doesn't use the CPU for geometry but other things that require FPU performance. I say that because I am sure most or all geometry would be handled by the GPU in Xbox 2.

Still, this is very interesting to read about and I look forward to the other responses to this topic.

Thank you zurich for making me think.

 

[zurich]

Ya, it's a tough question because after browsing the 3.06ghz P3 reviews, HT showed virtually no increase in performance whatsoever.

http://www.tech-report.com/reviews/2002q4/pentium4-3.06/index.x?pg=9

1-2% tops.

However, these games are SMT unaware, and as I said Windows XP has a known resource sharing issue with HT.

Can each logical processor execute SSE2 instructions, or are those registers shared?

*lights a fire under our resident developers butts!*

 

[V3]

How far 'to the metal' can a developer code, to take advantage of HT?

If you could code to the metal, HT wouldn't give advantage, better off putting a second core in there.

 

[Grall]

If AI and physics ran in separate threads, the CPU could alternate between either if it ran into a cache stall on the other. Programmers wouldn't have to code to the metal at all, just break stuff up into a couple different threads instead of running everything sequentially in the same thread.

*G*

 

[[maven]]

I remember from a presentation done by the guys that did "Aquanox" at Dawn2Dusk that they made a seperate thread for terrain occlusion culling in "Aquanox 2" to support HT (more so than SMP) systems. Said gave them about 10-20%, but "thread-synchronisation was hell" (paraphrased).
Otherwise, the presentation was boring as hell.

 

[Brimstone]

How far 'to the metal' can a developer code, to take advantage of HT?

If you could code to the metal, HT wouldn't give advantage, better off putting a second core in there.

A dual core cpu should get a signifigant boost from HT.

A single core cpu with HT probably would probably only get minor improvements.


If Microsoft doesn't have Intel provide a dual core CPU for the X-Box 2, the impact of HT will be marginal in the grand scheme of things. On the other hand if the X-Box 2 is blessed with a dual core cpu with Intels HT technology, game programmers will have been given a big leap in power.

 

[London Geezer]

A dual core cpu should get a signifigant boost from HT.

A single core cpu with HT probably would probably only get minor improvements.


If Microsoft doesn't have Intel provide a dual core CPU for the X-Box 2, the impact of HT will be marginal in the grand scheme of things. On the other hand if the X-Box 2 is blessed with a dual core cpu with Intels HT technology, game programmers will have been given a big leap in power.

i'm pretty sure they will have dual (or more...?) cores in the CPU or maybe in the GPU even though i think they should not do the whole crappy-CPU-good-GPU thing. i mean, animation and physics will be too important in the next generation to have an underpowered CPU... especially compared to what PS3 CPU will be, or better, is supposed to be...

some people still think Microsoft will be fine with a good GPU and a crappy CPU. i don't think it will cut it in the next generation... polygons and textures are nice and all, but it's the animation that sets apart Pixar movies from games graphics, and come next generation, when we will have the IQ of these movies, animation will step in as the main factor, or one of the main factors at least. and unless they put all the animation and physics code on GPU, then i dont see NextBox as very impressive.
but maybe its just me

 

[jvd]

Well under win98 apps show either a decrease in performance or an increase depending on the app. I think it wil lbe the same way in games. So software will run faster than others .


I still say that the boost from x86-64 is more signifigant (mostly because of the extra registers it can use) than hyperthreading . I guess we will see how much it effects games when ut2003k 64 bit comes out .

 

[Citan]

I read an interview with Gabe Newell a while back and he basically said that for them coding for HT was more trouble then it was worth, especially compare to a relativly trivial matter(in comparison) of recompiling for x86-64.

 

[jvd]

I read an interview with Gabe Newell a while back and he basically said that for them coding for HT was more trouble then it was worth, especially compare to a relativly trivial matter(in comparison) of recompiling for x86-64.

Yes I'm all hearing that utk2003 64bit sees a drastic improvement of alomst 20% in some cases . How true that is I don't know but if its true i can't wait to see the future pc games .