PDA

View Full Version : 31 stage pipeline!!!!!!!!!!


@md0Cer
02-03-04, 04:30 PM
Ok, I had better leave the Intel section fast before I get rotten tomatoes thrown at me, anyways, I have a question about the presscot's 31 stage pipeline.

Based off of some reviews and benchmarks, it seems as though a prescott at the same clockspeed as a northwood performs the same or better in most cases. How is this so considering it has 11 more stages? I know they made some modifications to the branchpredictor, but it couldn't have been that good. Does anyone know why?

Thanks
-0cer
:cool:

seamadan000
02-03-04, 04:41 PM
the prescott at lower frequencies generally is a bit (a very small bit) slower than northwood, but as the frequencies increase this is less of an issue. Intel lengthened the pipeline to allow for better clockrate scaling, but they had to trade some efficiency for that. The 1meg cache helps to mitigate that as well, as it allows the pipe to stay full.

brywalker
02-03-04, 04:42 PM
Think of it this way...

There is double the cache. Why does it perform the same or *slightly* better at the same clock speed?

The 31 stage pipeline is bad. It does less work per clock now. This will help get to higher clock speeds, but the performance would be on par to a Northwood core if it could hit those speeds. All along Intel said that this wouldn't be a super awesome new knock your socks off core. Its just moving forward with current performance.

(loose) ex: 3.7G Northwood 800 FSB w/8K and 512K = 3.7G Precott 800 FSB w/16K and 1MB

Thing is, Northwood can't hit those clock speeds reliably from the factory, so it doesn't exist. Thats why the Prescott is here. If they were able to use the current 20 stage with the 2x cache on .09, that thing would whoop ass. But the manufacturing process isn't there yet.

I am still waiting for the dual (and quad) core CPUs that we were supposed to get 3 years ago from AMD and Intel.

Here's to dreaming.

@md0Cer
02-03-04, 05:18 PM
ah I see.

But from seeing some reviews on presscott, 1.4Vcore, aircooling and 4.2Ghz, damn! AMD had better do something very good to keep me a customer. Although I probably won't be upgrading for a long time.


I am still waiting for the dual (and quad) core CPUs that we were supposed to get 3 years ago from AMD and Intel.

What exactly would be the benefit of a quad cored CPU? Just that it would be like a duallie but all in one? BTW, I saw a thread somewhere a couple months ago about some CPU that had like 8 or 6 cores or something like that. The thing was the size of a hand!

seamadan000
02-03-04, 07:47 PM
look to anandtech.com's review. It shows that the prescott closes the performance gap with the northwood at higher speeds when they are equal. They also hypothesized that the prescott at around 3.7ghz would be as fast if not faster than a 3.7ghz northwood. I guess we will have to see. Batboy has a NW2.8 and is getting a prescott 2.8, so after overclocking both we will have an answer.

Zuzzz
02-04-04, 12:17 AM
I just read that is a 32 stage pipeline not 31. . Either way HOLY S*** !!

Z

RoadWarrior
02-04-04, 07:14 AM
As far as I understand it, from reading Xbit labs article, the performance hit doesn't show so bad because of all the other heavy optimisation and improvement they have done to the core. It's not a northwood with more cache and a longer pipeline.

I'm thinking that the celeron prescott might be worth waiting for for overclocking. Apparently it's due in the early summer (Q2,04) and will be intended for a 533Mhz FSB, that's a lot of FSB headroom! :D Hopefully they won't cripple the cache too bad. Though if it has the same cache as a northwood then that should tell us a bit about how much effect the cache has in hiding the latency issue with the longer pipeline. Anyhoo, I would think that the celly prescotts are gonna be an easy 4G chip with good un's doing 4.5, and if they only perform as good as a woody at 3.5-3.6 when at 4G then that's all good because it's harder to get a woody going that fast. HT is gonna be disabled on them but I don't really think that's a biggie.

With regard to the dual and quad cores, there may have been some miscomprehension of the rumours about HT, which is a virtual dual core, and the prescott HT which was rumoured it was going to be a virtual quad CPU but has now proved to be only dual as per the northwood.

Road Warrior

Krowa 02
02-04-04, 08:31 AM
Those celeron prescotts will have 256k cache, which is the same ammount from the willy p4's. Supposidly they are the same chips that intel couldnt get running up to spec at normal prescott settings, so they down graded them, so basically they wont be good overclockers until later steppings become available.

Cereal Killa
02-04-04, 09:09 AM
Originally posted by @md0Cer
What exactly would be the benefit of a quad cored CPU? Just that it would be like a duallie but all in one? BTW, I saw a thread somewhere a couple months ago about some CPU that had like 8 or 6 cores or something like that. The thing was the size of a hand! [/B]
I think those were processors from IBM, they were big no doubt about it :eek:
--Cereal

@md0Cer
02-04-04, 03:59 PM
Originally posted by RoadWarrior
As far as I understand it, from reading Xbit labs article, the performance hit doesn't show so bad because of all the other heavy optimisation and improvement they have done to the core. It's not a northwood with more cache and a longer pipeline.

I'm thinking that the celeron prescott might be worth waiting for for overclocking. Apparently it's due in the early summer (Q2,04) and will be intended for a 533Mhz FSB, that's a lot of FSB headroom! :D Hopefully they won't cripple the cache too bad. Though if it has the same cache as a northwood then that should tell us a bit about how much effect the cache has in hiding the latency issue with the longer pipeline. Anyhoo, I would think that the celly prescotts are gonna be an easy 4G chip with good un's doing 4.5, and if they only perform as good as a woody at 3.5-3.6 when at 4G then that's all good because it's harder to get a woody going that fast. HT is gonna be disabled on them but I don't really think that's a biggie.

With regard to the dual and quad cores, there may have been some miscomprehension of the rumours about HT, which is a virtual dual core, and the prescott HT which was rumoured it was going to be a virtual quad CPU but has now proved to be only dual as per the northwood.

Road Warrior

Well, I am not sure about the celerons. Although I do not know much about Intels, it seems as though they had better have some pretty damn good optimizations becuase 256K cache and a 31 or 32 stage pipeline, yea it might do 6.5Ghz(being sarcastic), but would it compare to a 1Ghz p3?

RoadWarrior
02-05-04, 08:21 AM
If the cellys are rejected P4s, then it might be that they had problems with the HT or that one or two or three banks of cache are bad, (Presuming 256KB banks) so if they're only disabling the bits that don't work so good then it might make the same or better speeds as P4 prescotts. Less cache can be good heatwise. They've got it spread all round the core now too, if the cache was making half or more of the heat, then the cellys could run a lot cooler. Look at the Tualatin celerons as an example, they had half the cache disabled from the PIII tualatin, people were getting only around 1.5-1.6G o/c with the PIIIs but 1.7-1.8 with the celerons. I guess you could argue that was due to the lower multis on the PIIIs though. But anyway there's quite a few examples from the past where the identical core with less cache clocked a heck of a lot better than the chip with more (PII vs celeron A, K6-2 vs K6-3 etc etc) Granted the efficiency of the chip is down, but the extra Mhz usually make up for it. Another example is that of AMD "Thorton" chips, when the cache is 256k people are hitting 2.6G with them, when they unlock the extra cache they find they can only hit 2.4G.

Intels are gonna be hitting 5Ghz on some peoples systems by the end of the year, inefficient or not, it's gonna take 3Ghz or better on an Athlon to match it. Though on AMDs next core shrink for Athlon XP 32bit, we should I guesstimate see it topping at 3.5Ghzor 3.7 or so with chilled water etc. However we'll need to get to 300Mhz FSB with those to stay efficient and maintain the same ipc equivalence with the P4s.

Heh, I'm not getting pro-Intel or anything, just a brand new core and process size and technology is exciting. AMD is looking like abondoning the 90% of the performance for 50% of the price bandwagon I've been riding though, if they go for price parity with intel, then you've gotta be short of a little common sense not to re-evaluate your CPU choices. I'm keeping an eye to the future is all.

Road Warrior

uvaman
02-05-04, 01:31 PM
What I dont understand is why Intel didnt just added those ICP improvements to the current Northwood processor..

I mean, Intel claims 4-5ghz for prescott, but that seems what they should get from the process shrink alone, Look @ the .13 change, from 2ghz to 3.4ghz, so thats a 1.4ghz grow, so 3.4 + 1.4 = 4.8ghz well within Intels 4-5ghz...
Im under the impression that they pinned their hopes with this 31stage pipelines in that they would have a 6ghz chip by the end of this year..


4.2ghz for a prescott, @ those temps I dont think they will last long, in fact I think that OCers will begin to notice that smaller and smaller CPU die faster and faster, I know im noticing.. I just killed my 2500+ athlon @ 3200 speed with a mere 1.75volts, it just isnt stable @ any speed anymore... thats just in a few weeks.. while my duron 800 was working @ 1030 for all the time I cared to use it.. I still have it laying around somewere...

I dont think I would overclock.... from the terrible experince I just had, I mean that 2500 booted @ stock voltage for the most part it worked @ stock volts, just crashed after some 10 min or so of gaming..., and was ROCK stable at 1.75v, it looked like I was gonna have this CPU for very long, but puff, down goes the system..... I might get hammer soon, depends on the $$.
those 754 look good, considering I always upgrade CPU/mobo at the same time, I wouldnt mind I AMD kills support next month, which they wont do anyway... I bet 754 will become durons, and will last for a loong time.. beside Its been some time since AMD has a real Duron/Normal CPU division like Intel has right now, execpt Intels Durons are so abismal,..., those 2.4c look good...