Linus Torvalds on Nehalem

pxc

Extremely [H]
Joined
Oct 22, 2000
Messages
33,063
http://www.realworldtech.com/forums/index.cfm?action=detail&id=93813&threadid=93745&roomid=2

>
>What really bugs me about Nehalem is the 4-cycles L1 cache.

From what I've seen (I've had one for a couple of months
now), the L2 more than makes up for it. Yes, the L1 is one
cycle slower, and no, you won't always make up for it, but
the L2 is several cycles faster.

I can't do public benchmarks, but I'm sure Intel won't mind
me saying that Nehalem certainly doesn't look like a step
backwards. My only personal gripe with the machine I have
is actually just that I'd rather have a dual-core and a
lower TDP, than the quad I have. But that's obviously not
a core issue, it's just a "which version comes out first"
thing.

And in the kinds of loads this thread talks about, the
quad-core (with SMT) will obviously be the thing to go for.
I suspect it will do quite well.

Linus
 
Who cares, Linus is a crazy goon now, he nitpicks at everything for no reason that really have almost no effect.
 
He loves to argue and even admits that. :p I usually don't pay much attention to him with his comments on hardware. He usually gets owned hard by others when he makes crazy claims. Within his domain (x86 software development), he is actually very knowledgeable.
 
I wouldn't really regard him as any more knowledgeable about this than anyone else who knows anything about how cache levels affect performance, which I imagine is a lot of people in today's communities.
While we're at it, why not post what Michael Schumacher thinks about ABS?
 
Linus no longer works at Transmeta... he works at OSDL now I believe.
He is extremely knowledgeable about microprocessors and OS design (Obviously.) His opinion *is* valuable, although many times OSS developers' opinions are unnecessarily criticized for being practical when many would expect altruism.
 
He invented Linux, thus what he says about computers is of value.

Unless he is wrong.

Just because someone does something cool and impressive like say land on the moon, doesn't mean you should believe him when he says aliens are real and are all around us.
 
He is extremely knowledgeable about microprocessors and OS design (Obviously.)
I'd strike the part about microprocessors. :p He knows x86 very well on the software side, but he's definitely not an expert on x86 (or most other processors') architecture(s). He sticks his foot in his mouth almost every time he strays there.
 
Still, it doesn't seem he's overstepped himself this time.

Yes, Nehalem is going to pwn. I doubt there are more than a few on these forums who would argue that.
 
Yes, Nehalem is going to pwn. I doubt there are more than a few on these forums who would argue that.

I haven't read anything that makes it seem like i7 is going to be anything like conroe was in 2006. It sounds like it will be a bit better, but at this point in late 2008 we are still looking at programs that do not use more than 1 or 2 cores, and usually don't use the second one very efficiently anyway. We can't even use the power of quads for many things, and now we have eight threads. What the hell is the point of that? If this thing were 40% more powerful than penryn clock for clock, I would be pumped. But it is looking more like 10 or 15%

Considering all the things I would have to buy (new motherboard, ram) to get one of these up and running, I am going to decline until the situation changes.
 
Well for the guy (me) with the X2 4400 and DDR1...I am LOOONG overdue to upgrade and I am excited about the i7s.
 
I read a while ago that even a spokesperson for Intel said the i7s wont do much for game performance (I cant be bothered to look for the source, it was a few months ago). Obviously thats not what everyone here is looking for, but I know for me personally that's what I would be looking for if I were in the market for a new CPU/Mobo/RAM setup.

Right now, I'm happy with my Penryn. :)
 
Well among the things for me is it really is future proof. Core is a great archetecture (I have a Penryn), but its not wide enough to last for another couple years by itself.

Clock a bloomfield to 3.6 or so and you will have PLENTY of single threaded performance, along with 4 hyperthreaded core for parallel performance. Win/win. Socket will last a while, meaning if you really want to upgrade its going to be available.

Plop in a new gen GPU (or a second GTX280) next fall and ride it out. A bloomfield chip should satisfy on CPU power for 2 years, until sandy bridge at least. Either Im getting less greedy or performance is reaching a level of diminishing returns. Either way, I dont feel the need to upgrade every 6 months anymore.
 
Either Im getting less greedy or performance is reaching a level of diminishing returns.

Feeling that too. Nothing seems to be needing/using the amount of power of today's tech. I'm definitely not seeing much of a difference vs. say, when I went from single core to 2core which was a gawdsend for even just switching windows/simple programs.
 
I read a while ago that even a spokesperson for Intel
Fudzilla is not Intel. :p

Intel is claiming (Core i7-965 vs QX9770): "52% more 3D gaming performance, 38% more rendering performance and 41% more movie editing and conversion performance." Looks like Intel is going the AMD route with performance claims.
 
Feeling that too. Nothing seems to be needing/using the amount of power of today's tech. I'm definitely not seeing much of a difference vs. say, when I went from single core to 2core which was a gawdsend for even just switching windows/simple programs.

Agree.
 
Fudzilla is not Intel. :p

Intel is claiming (Core i7-965 vs QX9770): "52% more 3D gaming performance, 38% more rendering performance and 41% more movie editing and conversion performance." Looks like Intel is going the AMD route with performance claims.

It's going to cater more toward the ''rendering/ movie editing'' people and the benchmarkers, but in terms of gaming performance Core i7 is a pretty big disappointment over the Core 2's.
 
GPUs are holding us back for the most part anyhow, there's no where for the chips to go really. The only applications that seem to be able to take more power appear to be rendering, media editing, etc stuff like that.

I don't think people should be worried at all about gaming performance, a faster chip won't make up for your video card holding you back.
 
This is crazy,it seems to be at a point of where the performance isn't gonna be increasing that much over current hardware.
The cpu side is great with the intel core 2 duo.A e8400 stock @3ghz can hit 4ghz easily and it is more then enough for applications and games today.
The video card side seems just to be a fight between nvidia and ati.They will come out with something a little bit faster than the previous gen and the price is $500 again.

I am still running the 64x2 @2.6 and I wont upgrade to I7.I have no need to since I will be spending so much cash and I feel like for nothing.Even for me to go to a E8400 is gonna cost me over $500.I will sit back,wait,and watch people go crazy over the new intel design since I feel we are barely making steps forward now.It isn't like when it went from single core to dual core that was like a friggin leap in technology but now we are just spending useless money
 
This is crazy,it seems to be at a point of where the performance isn't gonna be increasing that much over current hardware.
The cpu side is great with the intel core 2 duo.A e8400 stock @3ghz can hit 4ghz easily and it is more then enough for applications and games today.
The video card side seems just to be a fight between nvidia and ati.They will come out with something a little bit faster than the previous gen and the price is $500 again.

I am still running the 64x2 @2.6 and I wont upgrade to I7.I have no need to since I will be spending so much cash and I feel like for nothing.Even for me to go to a E8400 is gonna cost me over $500.I will sit back,wait,and watch people go crazy over the new intel design since I feel we are barely making steps forward now.It isn't like when it went from single core to dual core that was like a friggin leap in technology but now we are just spending useless money

In my book the Nehalem is a greater advancement than the C2D was and I can use the extra CPU power and it quote a powerboost, when you look at none_GPU limited cases.
 
the future of high-powered processing is in CUDA and equivalents. new PC games dont even max out CPUs from 2 years ago and programs supporting CUDA have already proven to be 3-4 times faster than the best Extreme Quad for 1/3 the price or less.

Extra cores will almost always increase encoding and rendering in apps that support multi-threading (which is excellent), but once Video codecs or rendering apps start utilizing CUDA, it's game over for people who do these things professionally. a GTX280 is already encoding movies at least 2 times faster (some benchmarks show as high as 4x under some conditions) than the QX9770 and it's only been in beta testing for about 6 months.
 
No i7 for me bro.I will not be able to afford nor do I need it right now
 
What he says makes sense, no?
Basically he says this:
L1 went from 3 to 4 cycles, but L2 went from 14 to 10 cycles. On average, the faster L2 can compensate for the slower L1.

It reminds me of the Pentium 4. The first iterations (Willamette, Northwood) had only 8k L1 cache. This was much smaller than the 32k that Pentium 3 had. However, it was also faster, and the L2 was much bigger.
 
It's going to cater more toward the ''rendering/ movie editing'' people and the benchmarkers, but in terms of gaming performance Core i7 is a pretty big disappointment over the Core 2's.

when core 2 first came out. HOW MUCH benefit did it have over the previous fastest cpu for GAMES? did u see a huge jump in fps?

i have heard from people here INCLUDING Kyle that no, the core 2 diddnt make ANY difference to games back when it first launched.

why? because games are GPU, yes GPU dependant. Kyle even made a article about it. this is deja vu as far as i can see.

infact here it is by the guy himself. http://www.hardforum.com/showpost.php?p=1033156927&postcount=75

all those who complain about i7 not making a difference seriously need to read the above link as core 2 was in the SAME position when it first came out!

stop thinking about todays games and thing about tomorrows games. i bet u alot of cash that your beloved core 2 thats less then 4 cores will suffer with alan wake WHICH HAS been confirmed to be optimtized for quad core

12211450832r35WCHvRr_5_2.gif
 
when core 2 first came out. HOW MUCH benefit did it have over the previous fastest cpu for GAMES? did u see a huge jump in fps?

i have heard from people here INCLUDING Kyle that no, the core 2 diddnt make ANY difference to games back when it first launched.

why? because games are GPU, yes GPU dependant. Kyle even made a article about it. this is deja vu as far as i can see.

infact here it is by the guy himself. http://www.hardforum.com/showpost.php?p=1033156927&postcount=75

all those who complain about i7 not making a difference seriously need to read the above link as core 2 was in the SAME position when it first came out!

stop thinking about todays games and thing about tomorrows games. i bet u alot of cash that your beloved core 2 thats less then 4 cores will suffer with alan wake WHICH HAS been confirmed to be optimtized for quad core

12211450832r35WCHvRr_5_2.gif

/thread.

(I have used same link and same argument on other sites, thumbs up)
 
Funny how all Nehalem threads end up in a debate on game performance and GPU-limitness.

The opening post doesn't make any mention of games or GPUs at all. It discusses the increased latency of the L1 cache, which Torvalds claims to be masked by the lower L2 latency in practice.
Discuss.
 
Funny how all Nehalem threads end up in a debate on game performance and GPU-limitness.

The opening post doesn't make any mention of games or GPUs at all. It discusses the increased latency of the L1 cache, which Torvalds claims to be masked by the lower L2 latency in practice.
Discuss.

i was only replying to a person claiming to say that the i7 doesnt have a big impact on games like how core 2 have while in reality the core 2 in its infantry diddnt provide a game boost at all just like a i7. but look how many people have a core 2 now eh?

the simple reason why i7 dont make much difference in games is two reasons. Games are GPU bound and two. hardly any games are optimised for more then one core let alone 4 with i7's are.
 
i was only replying to a person claiming to say that the i7 doesnt have a big impact on games like how core 2 have while in reality the core 2 in its infantry diddnt provide a game boost at all just like a i7. but look how many people have a core 2 now eh?

the simple reason why i7 dont make much difference in games is two reasons. Games are GPU bound and two. hardly any games are optimised for more then one core let alone 4 with i7's are.
Fail. This has nothing to do with cache latencies.
Funny how all Nehalem threads end up in a debate on game performance and GPU-limitness.

The opening post doesn't make any mention of games or GPUs at all. It discusses the increased latency of the L1 cache, which Torvalds claims to be masked by the lower L2 latency in practice.
Discuss.
Even a broken clock is right twice a day. Linus may or may not understand what he's talking about, but I think it's clear that Nehalem will represent a significant step forward over Conroe/Penryn. Thus, he ends up being right.
 
Fail. This has nothing to do with cache latencies..

But it has everything to do with this quote :
Originally Posted by 3far4shot View Post
It's going to cater more toward the ''rendering/ movie editing'' people and the benchmarkers, but in terms of gaming performance Core i7 is a pretty big disappointment over the Core 2's.

You fail :D
 
But it has everything to do with this quote :

You fail :D
That post was irrelevant because it was off-topic.

There are a dozen other "Nehalem vs. Conroe for games" and "Dual core vs. Quad core for games" threads you could waste your breath in. Why don't you go there, instead of continuing to post things that have nothing to do with the topic in this one?
 
Thus, he ends up being right.

Not really. He specifically talks about the caching, you talk about performance compared to Conroe/Penryn, which is something different altogether.
Namely, the L2 might actually compensate for the L1 (which I believe will be the case), but the overall performance will still be slower than Conroe/Penryn (because of factors not related to the caching itself).
Or, conversely, L2 does not compensate for L1 but Nehalem is faster anyway (because of factors not related to the caching itself).

Personally I think the most likely option is that L2 does compensate for L1 most of the time, AND Nehalem is faster than Conroe/Penryn on most tasks (the new architecture has many improvements to increase instruction throughput, both per thread, and in multithreading).
 
That post was irrelevant because it was off-topic.

There are a dozen other "Nehalem vs. Conroe for games" and "Dual core vs. Quad core for games" threads you could waste your breath in. Why don't you go there, instead of continuing to post things that have nothing to do with the topic in this one?

why dont u post there :D relax dude
 
Back
Top