Conroe performance claim being busted
Make of it as you will. The guy is clearly an AMD zealot. but he does make some good points.
Make of it as you will. The guy is clearly an AMD zealot. but he does make some good points.
Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
God be with us.Dillusion said:God, more fuel for the immature AMD fan club, this is never going to end.
basically saying that conroe's main advantage is with programs that fit within it's unified 4mb cache. he points out how the benches in sciencemark, the ones that use a very small amount of memory, conroe wins easily, but the ones that use more, the X2 catches up and surprasses.. it also explains why conroe is able to get an exceptionally fast superPI 1m time, since 1m seems to allocate just over 8mb.. 1/2 of that can be thrown in the cache at any given time, and with decent prefetching, a cache miss is gonna be pretty rare.Mayhs said:can someone summarise it please the artilce wont show up
If you compare the BLAS score in Sciencemark which uses x87, Conroe is superior.(cf)Eclipse said:also keep in mind one other thing: it appears that conroe is gonna suck at FPU intensive tasks. luckily for intel, there's SSE.
(cf)Eclipse said:basically saying that conroe's main advantage is with programs that fit within it's unified 4mb cache. he points out how the benches in sciencemark, the ones that use a very small amount of memory, conroe wins easily, but the ones that use more, the X2 catches up and surprasses.. it also explains why conroe is able to get an exceptionally fast superPI 1m time, since 1m seems to allocate just over 8mb.. 1/2 of that can be thrown in the cache at any given time, and with decent prefetching, a cache miss is gonna be pretty rare.
ontop of this, it seems that with anything that is multithreaded and likes lots of cache, there's potential negative sides to the unified cache that might pop up, similar to what happened with hyperthreading..
how this will translate into real world performance, we'll have to wait and see. 4mb is a considerable amount of cache.... and games did seem to do really well in the anand test, most likely due to three things:
cache size
integer power
exceptionally optimized SSE core
however, two other interesting things:
1. how do they compare in 64bit mode? i'm using xp-64 almost exclusively now, and use 64bit apps whenever possible. it really saddens me how little progress we've made into the x86-64 world thus far, especially since all new CPUs have x86-64 capability. along with going multithreaded, this is something that programmers REALLY need to focus on, especially since it should be easier to optimize things for 64bit than making it multithreaded
2. conroe is apparently exceptionally similar to K8, most likely aside from a few things such as how the ALU/FPU units are set up (it seems that intel went for a ton of ALU power while scaling back the FPU a bit and focused on SSE instead.. ironic, considering the K8L is supposed to double the number of FPU units ), and the SSE execution engine, which is an area where conroe definitly holds a tremendous advantage in atm.
Conroe can just run off the cache with very high speed -- another cheap gimmick at the expense of very large die size.
yeah, i'm keeping my fingers crossed, but it's gonna be mighty tough for just a memory controller change to catch up to what conroe is doing.Bona Fide said:Frankly, I was impressed with the sheer efficiency of the memory controller on the K8 processors. If AMD could do the same, or even close to the same, with AM2, we'd have a real winner.
i went to look, but XS is down again.SLee said:If you compare the BLAS score in Sciencemark which uses x87, Conroe is superior.
(cf)Eclipse said:i went to look, but XS is down again.
yup, i see. thanksSLee said:Here are a couple links to posts made by one of the main Sciencemark programmer (with scores):
http://www.aceshardware.com/forums/read_post.jsp?id=115160317&forumid=1
http://aceshardware.com/forums/read_post.jsp?id=115160639&forumid=1
i am amd biased by definitionTetrahedron said:I love how its ALWAYS AMD biased.. but never ever ever could there be Intel biasness happening right?
pxc said:In the other thread (XS or the intel forum?) where VictorWang's conroe results showed the only benchmark that conroe lost was sciencemark, someone mentioned that sciencemark will become the most important benchmark now. The link in the first post proves it.
SLee said:Here are a couple links to posts made by one of the main Sciencemark programmer (with scores):
http://www.aceshardware.com/forums/read_post.jsp?id=115160317&forumid=1
http://aceshardware.com/forums/read_post.jsp?id=115160639&forumid=1
HighwayAssassins said:I think i will just wait for the benchies when teh chips actually come out.
coldpower27 said:Conroe is only about 14x mm^2 by current die size estimates and that is with the "enormous" 4MB cache, thanks to 65nm fabrication technology, and Intel's exceptionally good cache density, Intel can afford to throw in this muc cache on the 65nm and still price it reasonably aka 316US for the E6600 model.
It isn't Intel's problem that AMD lags behind Intel on process technology, and has worse cache densities then Intel does. Hence AMD designed their architecture to not rely on cache too much, they simply can't afford to put that much cache on in the first place.
Interesting that you say you work in a Fab and say why AMD doesn't have 2X2Mb because of cost is silly, well care to shed some light on why AMD didn't throw on 2x2Mb cache? I gave more then just cost reason why AMD didn't do it if you read my posts.powerade said:all ill say is i work in a fab, saying amd doesnt have 2x2 mb caches yet simply cuz of price is laughable
i cant remember if conroe is due out in the fall or the summer( no time to keep up on other company's products and shit, and not due for a comp upgrade til later) so anyway, it takes up to 3 months to manufacture chips from start to finish, so the samples anad got to benchmark is basicaly final silicon (if its coming out when i think)
and keep in mind intel was saying ~20% gain over fx 60 which had been out for a lil while, purly comical. I guess theres plenty of dell buyers out there that will buy into the hype though so i guess its kinda ingenious of intel.
coldpower27 said:Interesting that you say you work in a Fab and say why AMD doesn't have 2X2Mb because of cost is silly, well care to shed some light on why AMD didn't throw on 2x2Mb cache? I gave more then just cost reason why AMD didn't do it if you read my posts.
I think cost and yield do play an important role in determining how much cache can be put on a processor Windsor the Socket AM2 Athlon 64x2 has a Die Size of 220 mm^2 with 2x1MB and that is much larger then Conroe is. Hence why you don't see AMD throwing more cache onto it as they can't afford to at the moment not on 90nm DSL SOI, or won't because they want to enjoy the margins they have had the luxury of charging when Intel made this mistake.
So the margin of victory
baldrik said:wait a minute! aer they handing out medals for this victory? if they aer, i bet amd just doesnt know about it yet and thats why they are lagging behind!!
look out hector!! they a3r here to steal you medals!! dont let them have teh 1337n3ss!!
oh for sure! if I didnt let you know, how else would you know?osalcido said:well thanks for letting us know
Comparing Conroe die size to Windsor is very legitimiate, as they will be out in the same time frame, as AMD will be still on the 90nm DSL SOI process when Conroe ships, AMD has no one to blame but itself for this. It's not Intel's problem that it is 9-12 months ahead of AMD on the 65nm transistion.mwarps said:Comparing a die size of a 65 nm process to a 90nm process is, at best, moronic. Not for just for coldpower, but also powerade. If we assume that conroe is 14mm on a side at 65, we'll just go linearly to 21mm on a side at 90nm, which magically takes us to 440?sqmm.. more than twice the size of the X2. Kiddies need to learn how to play nice, and to do a little simple math before they get in big fights on the intarweb.
(btw, even if the scales aren't linear between generations, I don't care, this is utterly stupid)
such is the problem when your slower then your competitor on process transitions.
such is the problem when you dont have infinite capitol to squander on getting the process right
Rofl, is that the best you can come up with, it also not Intel's problem that AMD is poorer and is lacking in financial reosurces due to lack of a competitive product for a good portion of it's lifetime. This is AMD's own doing.baldrik said:let me fix that for you
yes it is.not Intel's problem that AMD is poorer and is lacking in financial reosurces due to lack of a competitive product for a good portion of it's lifetime.
Why on earth would AMD having no financial reosurces be Intel's problem?baldrik said:yes it is.