![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
What will you buy after the price drops?
- Thread starter osalcido
- Start date
InorganicMatter
[H]F Junkie
- Joined
- Oct 19, 2004
- Messages
- 15,461
Remember, AMD is discontinuing non-FX 1MB chips, and I think not cutting prices on them is the first step. If you want a 4400+, 4800+, or Opteron, you'd better get it quick because they will be no more shortly. Myself, I'd buy the 5000+ and forget the cache since it doesn't make a difference.Yashu said:
the extra cache does make a difference... at the very least it is ~200mhz plus a few other advantages.
the integrated memory controller is the reason why cache is not *as* important, but when you have alot of RAM, the extra cache means you have fewer cache misses.
in all it depends on your application. I could use the extra cache, or would at least like to have it if I could.
the 5000+ looks nice if I was going AM2, however.
AMD discontinuing them points to their innability to increase their silicon wafer size. Intel uses one of the largest wafer sizes in the industry *and* uses a 65nm process... they can afford 4mb caches...
AMD can't get their wafer size up or their process down so they have to resort to cutting away their larger cache part. it doesn't look good for amd regardless of the performance difference.
I hope they can get to 65nm soon, maybe we will see large caches again then. (k8L has L2 and L3, so that tells you no k8L at least until they get past their first gen 65nm)
the integrated memory controller is the reason why cache is not *as* important, but when you have alot of RAM, the extra cache means you have fewer cache misses.
in all it depends on your application. I could use the extra cache, or would at least like to have it if I could.
the 5000+ looks nice if I was going AM2, however.
AMD discontinuing them points to their innability to increase their silicon wafer size. Intel uses one of the largest wafer sizes in the industry *and* uses a 65nm process... they can afford 4mb caches...
AMD can't get their wafer size up or their process down so they have to resort to cutting away their larger cache part. it doesn't look good for amd regardless of the performance difference.
I hope they can get to 65nm soon, maybe we will see large caches again then. (k8L has L2 and L3, so that tells you no k8L at least until they get past their first gen 65nm)
I just bought an x2 4600+ (939) for $249 + s/h from outpost.com. When I ordered it last night, it showed as in stock - alas it's back ordered. They'll be getting another shipment in on Wednesday and they'll ship it to me the same day (plus i chose overnight). Kind of annoying since I'll have all the other new stuff i bought this week (x1800xt, asus mobo, x-fi, sata hd, psu, etc). Ah well..I guess I just have to concentrate on my summer class for a few days.
I really did want to build a Core 2 duo system, but I have 2gigs of ddr400 ... don't really feel like blowing 200 bucks on ddr2 plus the 200 bucks for mobo. But I figure for BF2, Oblivion, etc the AMD setup will be a HUGE improvement over my athlon xp-m 2800+ / 6600gt agp rig.
btw, what's up w/ newegg? takin forever to drop their prices.
I really did want to build a Core 2 duo system, but I have 2gigs of ddr400 ... don't really feel like blowing 200 bucks on ddr2 plus the 200 bucks for mobo. But I figure for BF2, Oblivion, etc the AMD setup will be a HUGE improvement over my athlon xp-m 2800+ / 6600gt agp rig.
btw, what's up w/ newegg? takin forever to drop their prices.
InorganicMatter
[H]F Junkie
- Joined
- Oct 19, 2004
- Messages
- 15,461
The 5000+ IS the the gap filler...bandit390 said:
dude, I own two AMD systems (one in my sig).
how about just because AMD is having to keep up with demand by reducing the production of larger die sizes doesn't mean I become an intel !!!!!!. look it up, intel uses a larger ingot.
It's just buisness, man, not a religion... calm down.
how about just because AMD is having to keep up with demand by reducing the production of larger die sizes doesn't mean I become an intel !!!!!!. look it up, intel uses a larger ingot.
It's just buisness, man, not a religion... calm down.
Yashu said:
Learn about what you are posting before you post it. Any idiot knows intel uses larger wafers. It's been that way for years.
All cores come from the same wafer. They disable failing parts at test, including the caches.
AMD may have moved to a new design which only has 512KB, but it's not going to save them waferspace, or at least not enough to make a large difference to their bottom line. And if they HAVE gone to this design to save their bottom line, they are DONE.
Wafersize has *nothing* to do with demand. It has everything to do with cost.
if you know this then why are you even bothering to argue with me...
AMD cutting half the cache saves alot of wafer space lets see here:
click here
ok now look at that pic there, the cache takes up at least half the space... so AMD can gain ~50% more space by cutting the cache in half... that is alot of extra CPUs right there. This is not an insignificant amount, my friend. **
AMD cannot afford extra waferspace, plain and simple. AMD has not been able to move to a larger ignot size either... they have to keep pace with demand somehow while keeping costs down (they cut their prices by up to ~50%... no coincedence).
Understanding the reality of things doesn't make me less of an AMD supporter. I still prefer their chips... I do not want to buy an intel chip until they are proper 64bit CPUs, this includes conroe (but that is not our topic).
**Keep in mind that opteron uses the same die as athlon64, just with more HT links enabled.
coldpower27
[H]ard|Gawd
- Joined
- Jan 27, 2005
- Messages
- 1,154
No it doens't save that much, did you even research on what the die size of Windsor based processor actually are?Yashu said:
I will tell you right now, 2x1MB parts are 230mm2, while 2x512Kb based parts are 183mm2. So you save some but not 50%
Removing half the cache on a Socket AM2 based Dual Core saves 20.5% die area. Nowhere near 1 half, it helps some, but it's only part of the solution, they still need to continue to ramp capacity up at Fab 36, and move onto Brisbane which will be a much more effective solution then this is right now in the meantime.
InorganicMatter
[H]F Junkie
- Joined
- Oct 19, 2004
- Messages
- 15,461
I'm pretty sure he meant 1/2 relative to the cache (half of 1MB is 512KBcoldpower27 said:
).coldpower27
[H]ard|Gawd
- Joined
- Jan 27, 2005
- Messages
- 1,154
But you don't gain 50% more space by halving the cache do you?InorganicMatter said:I'm pretty sure he meant 1/2 relative to the cache (half of 1MB is 512KB
Yashu said:
Because you have no idea what you're talking about, and you are trolling in here.
You do realize that it would take a new toolset, new maskset and waste millions of dollars to run a line of just 512KB chips alongside the 1MB versions which they are still producing(FX line, opterons) right? Wait, no, you don't realize it. I am informing you.
Edit for the uninformed: Chips with lesser cache are made on the same wafers as chips with more cache, they just have it disabled. (So I don't get flamed)
Here's a clue about semi manufacturing. Tools are tweaked like mad for specific processes. Retooling to run a different process loses them money. They would Never Do It while in production.
What they can't afford to do is redesign the entire chip to save space this late into the game. Yeah, they can redesign to cut space by about 25-33%. Still have to retool for it, get a new mask set, wait for the process to become mature. Time that takes? 3-6 months. Not to mention they would have to re-qual each tool each time they switched between the processes.
They cut their prices because they are forced to by Intel to stay competitive, not because they are making more chips. It's counterintuitive to good business practice.
*snark*
Yes, let's spend millions to get another process, new masks, and wait a few months for the new process to mature! oh look, we have 20 more chips on a wafer! (out of 200+) Now let's cut our prices in half!
*/snark*
That said, I am done with this thread.
The last paragraph.. that's just *funny*, I actually laughed.
yes you are right and about my math too... ~25% is what I should have said... 25% is a huge difference in terms of wafer size... (I am not sure where I got 50%... I posted with haste)
Moving to a 65nm process will gain even more area then 25% (I think this is where I was freudingly typing 50%) but we are not there yet... that is why I originally posited that the larger caches could return on that process, or at least no k8L until there are good yealds on 65nm (perhaps at least one generation in). k8L on 90nm would be too expensive.
anyway... the extra cache would be nice but the 5000+ is in a position that the 4800+ has been on s939 in terms of average performance yet way cheaper. it's a good deal.
coldpower27
[H]ard|Gawd
- Joined
- Jan 27, 2005
- Messages
- 1,154
This is known information, it's on the web by multiple reviews of when AMD launched the AM2 based Dual Cores.mwarps said:
AMD makes native 512k cache per core chips... I have one in my PC, manchester core. All those shiny new x2s they just dropped the prices on... they aren't all half cache disabled. most of them will be native 512k per core.
I am not sure what your point is. You almost agree with everything I say but you do it as if you are trying to argue.
coldpower27
[H]ard|Gawd
- Joined
- Jan 27, 2005
- Messages
- 1,154
Well since I actually have the figure it's actually closer to just 20% then 25%. Please don't do any sugarcoating with me unless you can explain why you think it's 25% when you only save 20% space.Yashu said:
Well I did the calculations and Brisbane 2x512KB would be about the same size as Intel Allendale based cores. AND AMD is not abandoning 2x1MB chips, on the 65nm node, there may or may not be 2 chips with 1 being 2x512Kb and 1 being 2x1MB we will see.
Obviously there will be no K8L on 90nm, considering it timeframe of introduction of Mid 2007, 65nm should be mature enough to use by that timefame as well it's a damn Quad Core so no way In hell would it be on 90nm process, though I already know this.
Well I don't really consider buying AMD products, but a 5000+ for 301US is not bad, but the Core 2 Duo E6600 supercedes it at 316US giving FX62 or better performance.
Yashu said:
*bashes head against a wall*
It is not a native 512K piece.
Take the heatspreader off.
Take the core off the mount, you can put it back on later, I promise.
Flip it over. Take out a microscope.
The second part of the cache is cut off (disabled) with a laser (or it's just scribed off)
*bashes head against a wall again*
EDIT:
If someone can prove me wrong with an intelligent post, I will go out and buy a Conroe right now. Right fricken now! (not AM2 stuff, they had time to get everything right for that. Doesn't make SENSE to run 512K parts along side 1MB parts, but if they do, good (or really bad) for them. )
coldpower27
[H]ard|Gawd
- Joined
- Jan 27, 2005
- Messages
- 1,154
mwarps, in this instance yashu is right.mwarps said:
AMD does fabricate both 2x1MB and 2x512KB chips on Socket 939 and Socket AM2. This allows AMD multiple avenues of creating SKU's.
An Athlon 64x2 3800+ Socket 939 can come from either a Toledo core with half it's cache disabled which has a die of 199mm2, or the native 2x512kb chip Manchester which has a die of 147mm2, though the Manchester core was only introduced with the advent of the 3800+ from what I remember. Intially all Socket 939 Dual Cores were based on the Toledo die at launch this changed though.
This isn't really a new thing with AMD, there are Venice (512kb) and San Diego (1MB) cores with differing die sizes, sometimes the dsibaled San Diego are used for the 512Kb chips, but alot of chips come from Venice based products.
There are even Manchester based Athlon 64's with a disabled core and cache from what I remember.
Intel doesn't do this typically, as they have massive maufacturing capacity relatively so that maybe why It may come as a s bit of a shock that AMD does it. I am surprised you didn't know this.
coldpower27
[H]ard|Gawd
- Joined
- Jan 27, 2005
- Messages
- 1,154
It does if your capacity constrained like AMD is. For socket AM2, they have both SKU's, with the advent of the removal of the 4000+, 4400+ and 4800+ SKU's more production can now be dedicated to the 2x512KB chips, as we need less 2x1MB chips now.mwarps said:
coldpower27 said:
Guess I'm an intel fan boy then. At least their manufacturing makes sense.
If you're going full gun to win, then just f-ing doing it (thus the nature of my hard-headed posts). Make every single damned chip like it was the best you can make.
Why the f*** does AMD do that?!?!!?
Appologies to Yashu.
Here it is, get ready for it...
No wonder AMD can't compete.
coldpower27
[H]ard|Gawd
- Joined
- Jan 27, 2005
- Messages
- 1,154
You gotta give AMD credit though, at least they are trying and giving us competitive price/performance ratio on thier mid to low end products.mwarps said:
I prefer Intel, but I do want AMD to try nonetheless, or Intel will fall into the same trap and rest on their laurels, having someone nip at thier heels is a good thing.