Sign in with
Sign up | Sign in
Your question

Cache size the next mhz race?

Last response: in CPUs
Share
February 6, 2007 6:39:26 PM

I hear a lot of people whining about amd not having big enough L3 cache's...but honestly do you think intel would be able to surpass AMD just by putting 25 megs of cache per core?! I say no! I think intel is being retarded like always...remember with multi cores and huge cache's it causes even more FSB cloggin (every1 here knows y...one core trying to get stored data off of the other cores cache. I know the L3 is supposed to help avoid this...but anyway)

5200mhz Hypertransport speeds (isn't that what the chart some1 posted said for the 1000 series of opterons?) Cray-Zee

More about : cache size mhz race

February 6, 2007 6:43:22 PM

Quote:
I hear a lot of people whining about amd not having big enough L3 cache's...but honestly do you think intel would be able to surpass AMD just by putting 25 megs of cache per core?! I say no! I think intel is being retarded like always...remember with multi cores and huge cache's it causes even more FSB cloggin (every1 here knows y...one core trying to get stored data off of the other cores cache. I know the L3 is supposed to help avoid this...but anyway)

5200mhz Hypertransport speeds (isn't that what the chart some1 posted said for the 1000 series of opterons?) Cray-Zee


First intel has successfully competed by increasing cache as Tulsa has 16MB L2. Second, HT3 is at 5200MT/s not MHz. The clock speed is 2000MHz 4000MHz effective.

Cache can be considerably larger than logic so there is a limit based on die size.
February 6, 2007 6:44:31 PM

Quote:
I hear a lot of people whining about amd not having big enough L3 cache's...but honestly do you think intel would be able to surpass AMD just by putting 25 megs of cache per core?! I say no! I think intel is being retarded like always...remember with multi cores and huge cache's it causes even more FSB cloggin (every1 here knows y...one core trying to get stored data off of the other cores cache. I know the L3 is supposed to help avoid this...but anyway)

5200mhz Hypertransport speeds (isn't that what the chart some1 posted said for the 1000 series of opterons?) Cray-Zee


No, it is not. TDP, IPC, PPW and overall performance are the new Ghz races. Then again, with K10 and with a much tighter gap in IPC and TDP, the Ghz race might start again. Which Intel will almost definitely win.

Big cache is coming because, although it eats up wafer space like hell, it's a simple way to increase performance. In terms of K10, it's apparently an important architectural feature.
Related resources
February 6, 2007 6:49:18 PM

No it wont. Unlike with the GHz race, larger cache sizes generally yield diminishing returns, whereas upping clock frequency generally gives you an across the board linear increase in performance.

The next GHz race is for a short while going to be number of cores I think. The reason for this is that its the simplest thing to sell to your average consumer - things like IPC and PPW are too complicated for them to understand or care about. GHz is out of the window, so the next thing that they will look to is more cores. Retailers will market more cores = more power, even if it's not always true. Even AMD is getting stuck into this arms race with the 4x4 platform - who on earth wants 8 cores, unless you do large amounts of video processing?
February 6, 2007 6:52:31 PM

Quote:
No it wont. Unlike with the GHz race, larger cache sizes generally yield diminishing returns, whereas upping clock frequency generally gives you an across the board linear increase in performance.

The next GHz race is for a short while going to be number of cores I think. The reason for this is that its the simplest thing to sell to your average consumer - things like IPC and PPW are too complicated for them to understand or care about. GHz is out of the window, so the next thing that they will look to is more cores. Retailers will market more cores = more power, even if it's not always true. Even AMD is getting stuck into this arms race with the 4x4 platform - who on earth wants 8 cores, unless you do large amounts of video processing?


Enthusiasts, gamers with too much money, people who actually DO do a lot of rendering and video editing. For them, a QuadFX with two K10 quad-cores is a productivity boon.
February 6, 2007 6:58:00 PM

Yeah but in reality, the people who could utilise all those cores represents a small fraction of all computer users. It serves the purpose of giving enthusiasts bragging rights sure and gamers could throw away their money yeah, and see no benefit over 4 cores. That's precisely the point - except for the few people who really can utilise all those cores, more than 4 cores is a gimmick, and by the time people really do need 8 cores, I suspect that we'll have seen a major shift towards assymetric and potentially dynamically reconfigurable processors.
February 6, 2007 7:06:19 PM

vista 2.0 thats why you'll need a quadfx platform just to boot your compy in under 10 minutes :p 
February 6, 2007 7:31:43 PM

Quote:
I hear a lot of people whining about amd not having big enough L3 cache's...but honestly do you think intel would be able to surpass AMD just by putting 25 megs of cache per core?! I say no! I think intel is being retarded like always...remember with multi cores and huge cache's it causes even more FSB cloggin (every1 here knows y...one core trying to get stored data off of the other cores cache. I know the L3 is supposed to help avoid this...but anyway)

5200mhz Hypertransport speeds (isn't that what the chart some1 posted said for the 1000 series of opterons?) Cray-Zee


Oh here we go again with the intel cache thrashing argument again. I'm sorry to say, sir that you don't know what you are talking about when it comes to the Core micro-architecture. I strongly recommend you educate yourself before making such biased and untrue comments in public.

I would suggest you start by reading this article
Anandtech Core 2 article

Pay particular attention to the section where they go into detail about how the memory subsystem works. In the future I would recommend you do some research before you go regurgitating misinformation.
February 6, 2007 7:41:31 PM

Quote:
I hear a lot of people whining about amd not having big enough L3 cache's...but honestly do you think intel would be able to surpass AMD just by putting 25 megs of cache per core?! I say no! I think intel is being retarded like always...remember with multi cores and huge cache's it causes even more FSB cloggin (every1 here knows y...one core trying to get stored data off of the other cores cache. I know the L3 is supposed to help avoid this...but anyway)

5200mhz Hypertransport speeds (isn't that what the chart some1 posted said for the 1000 series of opterons?) Cray-Zee


No, it is not. TDP, IPC, PPW and overall performance are the new Ghz races. Then again, with K10 and with a much tighter gap in IPC and TDP, the Ghz race might start again. Which Intel will almost definitely win.

Big cache is coming because, although it eats up wafer space like hell, it's a simple way to increase performance. In terms of K10, it's apparently an important architectural feature.I think there will always be a gigahertz race. New architectures will still benefit from faster clock-speed. If GHz doesn't matter anymore, then why did AMD bring out a 3.0GHz K8...that for all intents and purposes.. is at the ragged edge? It's similar to cars. You can lighten the car, mess with gearing, aerodynamics, etc. for better acceleration and top-speed, even use more efficient engines(double-overhead cams, variable valve timing, 4 valve heads,etc), but even with these super-efficient engines, bumping the displacement will bring more horspower/torque(as long as other systems are designed to keep up...freer-flowing exhaust/better intake,more appropriate cam(s), timing changes, bigger injectors, etc). The old saying still holds true...."you can't beat cubes". :wink:
February 6, 2007 8:39:11 PM

Or whatabout: "there's no replacement for displacement"?
February 6, 2007 8:42:30 PM

Quote:
Or whatabout: "there's no replacement for displacement"?
That one too. :wink:
February 6, 2007 9:05:57 PM

AMD will b doomed. when AMD release its K8L architecture, Intel will release faster Core processor, probably above 3Ghz. 3GHz Core vs 2GHz K8L, which one u will buy if it is the same price?
February 6, 2007 9:20:56 PM

That would depend on the benches and prices.
February 6, 2007 10:16:27 PM

Who cares about intel and their 3.33 ghz!?!?! haven't we been going over this a billion trillion times? Crapburst proved at even at over 5ghz a core you couldn't touch a fx60 tweaked by the same kind of overclocking nut to 3.4ghz.

I think the quadcores will dominate the server segment, and probably giving people like Kingpin something new to crackout on and get those 3dmark06 score to 30k.

maybe they'll optimize the K10 to run with the R600...HEAVEN FORBID!!

Then intel will have to buy nvidia...i'm gonna run via cpu's and matrox cards!! blech!

I think Intels early adoption of ddr2 is the only reason they are ontop right now...because just like intel amd saw almost no gain in performance when they transitioned to ddr2...but now both companies have their sights (sites?) set on ddr3 which will be the true showdown. If AMD's kuma's are just their standard 64 bit cpu's revamped only slightly (which most of the reports are saying to the otherwise) then that'll be superboneheaded on their part and i will probably loose all faith in AMD.

1250mhz ddr2 ram any1? imagine the dominator series of ddr3!!

i think amd should buy ocz while their at it...
February 6, 2007 10:51:15 PM

Was that reply meant for me? Cause I'm not sure how that applies to my comment.
February 6, 2007 11:46:35 PM

If Intel were to put outrageous amounts of well connected cache on their Xeons, then yes I think that'll steal much of AMD's thunder since they're partially banking on better caching for Barcelona to perform so well in servers.

But cache shows much stronger signs of diminishing returns on the desktop because users tend to run small chunks of random code, where just a little cache matters.

Depending on the market segment, you'll be pushing different metrics for your chip designs.

More cache never causes a higher strain on the FSB, by the way. But enormous disconnected cache banks negotiating over the same FSB are hardly more effective than dumping that cache and accessing main memory directly. My point is that it doesn't hurt performance directly, though it'll certainly hurt fab capacity.
!