Questions about Core 2 and L2 Cache

[Sabrewulf165]

Correct me if I'm wrong:

Core 2 Duo E6300/E6400 = 2MB total L2 cache shared between both cores

Core 2 Duo E6600/6700 = 4MB total L2 cache shared between both cores

If you refer to the recent digit-life review, you'll see they claim both the 6400 and 6700 have 4096KB L2. Did they simply pull a PCPer and underclock the 6700 or what?

Also, how important do people expect the larger cache on the 6600/6700 to be? Having been a long-time AMD user, I'm pretty accustomed to the CPU not benefitting much from larger cache except for respectable bumps in cache-intensive apps like archiving or code compiling, but I've heard that Intel's new "smart fetch" or whatever it's called works much better with the 4MB CPUs than the 2MB ones. I'm just wondering whether it's worth it to upgrade from a 2.4ghz (2x512k) X2 to something like an E6300. I assume I'll be able to OC the 6300 at least a resonable amount, but I'm wondering if the smaller cache is going to prevent it from really giving me the full Core 2 "experience."

 

[Centauri]

Most of the reason you're used to not seeing L2 cache make much of a difference with AMD is that you're used to having your processor ride on a bus architecture that doesn't suck. Intel's bus architecture does suck, making cache a key for good performance.

 

[Sabrewulf165]

Most of the reason you're used to not seeing L2 cache make much of a difference with AMD is that you're used to having your processor ride on a bus architecture that doesn't suck. Intel's bus architecture does suck, thus cache really is a pretty strong player in the world of Intel performance.

ROFL OK... that didn't really answer any of my questions, and it's sure to draw a lot of flames at you, but thanks I guess.

 

[coldpower27]

ROFL OK... that didn't really answer any of my questions, and it's sure to draw a lot of flames at you, but thanks I guess.

Agreed with the flaming part with the poster above your post.

If your talking a Athlon 64x2 4600+, I would say the E6400 should be the minimum you would need to make it worth while upgrade preferably the E6600 if you want some gains.

The Cache makes a difference, but not really dramatic. I had a site that showed cache differences in one of my posts, on a High IPC architecture such as Pentium M/Athlon 64/Core 2 Duo large cache usually produce limited gains. It's the latency and how advanced the cache is that play more of a role on performance. On COre Architecture clockspeed should have the largest impact on performance not cache.

Hold on while I pull up the bench showing the difference at equivalent clockspeeds.

http://www.hardforum.com/showpost.php?p=1029600365&postcount=53

 

[Donnie27]

ROFL OK... that didn't really answer any of my questions, and it's sure to draw a lot of flames at you, but thanks I guess.

In Cache dependent apps like some games, less cache will matter. Many streaming apps it will not matter as much unless you multitask a lot while running them.

If you refer to the recent digit-life review, you'll see they claim both the 6400 and 6700 have 4096KB L2. Did they simply pull a PCPer and underclock the 6700 or what?

The 6400 is supposed to have 2MB unless Intel changed the rules like they did by holding off on the 8000 series and calling the E6800 they Core 2 Duo E. There was clearly a rumor that the 4MB 2.13 was supposed to be the E6500 model.

 

[burningrave101]

Core Duo 2 can share the L2 cache between the two cores so with the E6600 for example a single core could effectively make use of 8MB L2 cache. Cache seems to make the most difference in areas like games. Its not a huge difference but its usually a few fps.

 

[Aganerral]

Core Duo 2 can share the L2 cache between the two cores so with the E6600 for example a single core could effectively make use of 8MB L2 cache. Cache seems to make the most difference in areas like games. Its not a huge difference but its usually a few fps.

Only 4MB total, not 4/core

 

[Rishi M.]

Core Duo 2 can share the L2 cache between the two cores so with the E6600 for example a single core could effectively make use of 8MB L2 cache. Cache seems to make the most difference in areas like games. Its not a huge difference but its usually a few fps.

Huh? The E6600 has 4MB shared cache, not 2x4MB.

 

[burningrave101]

Huh? The E6600 has 4MB shared cache, not 2x4MB.

Actually i've heard it might have 2x4MB shared L2 cache.

This speed is achieved by running at a 9x multiplier at 266MHz FSB. We initally thought that 4MB L2 cache would be shared between the two cores on this CPU but the motherboard BIOS reported 8MB as total L2 cache which means 4MB per core.

http://www.tbreak.com/reviews/article.php?cat=cpu&id=456&pagenumber=1

Is it for sure only 4MB total shared L2 cache? I guess the BIOS just probably isn't reading the correct amount of cache then since its shared.

 

[Megalomaniac]

Engineering samples is what they had, not production.
if you look at some other reviews, they mention this.
that is why some of 6400 chips you see have 4MB cache.
but a production model, only 6600 and up will have 4MB
all others will have 2 MB

 

[coldpower27]

Actually i've heard it might have 2x4MB shared L2 cache.



http://www.tbreak.com/reviews/article.php?cat=cpu&id=456&pagenumber=1

Is it for sure only 4MB total shared L2 cache? I guess the BIOS just probably isn't reading the correct amount of cache then since its shared.

I cna gurantee 100% that Conroe has 4MB of cache in total, this is a fact, that site screwed up as it relied on CPU-Z as that views how much cache a core has access to.

Which is 4MB in either cores case.

 

[Donnie27]

Agreed with the flaming part with the poster above your post.

If your talking a Athlon 64x2 4600+, I would say the E6400 should be the minimum you would need to make it worth while upgrade preferably the E6600 if you want some gains.

The Cache makes a difference, but not really dramatic. I had a site that showed cache differences in one of my posts, on a High IPC architecture such as Pentium M/Athlon 64/Core 2 Duo large cache usually produce limited gains. It's the latency and how advanced the cache is that play more of a role on performance. On COre Architecture clockspeed should have the largest impact on performance not cache.

Hold on while I pull up the bench showing the difference at equivalent clockspeeds.

http://www.hardforum.com/showpost.php?p=1029600365&postcount=53

Ouch!

 

[Donnie27]

Engineering samples is what they had, not production.
if you look at some other reviews, they mention this.
that is why some of 6400 chips you see have 4MB cache.
but a production model, only 6600 and up will have 4MB
all others will have 2 MB

Link for hardheaded folks.

http://www.dailytech.com/article.aspx?newsid=1619

The 2.67GHz Conroe CPU featured in the IDF benchmarks last month will have an official MSRP of $530 at launch. The 2.6GHz AMD Athlon 64 FX processor used in those benchmarks costs a little over $1000 today, though AMD roadmaps have revealed this product will not get a price cut until June at the earliest. AMD's top of the line DDR2 2.6GHz processor will ship for just under $700 at the end of Q2'06.

 

[Bona Fide]

Here's how it is:

E6300/E6400 = 2MBx2 = 4MB total L2 cache
E6600/E6700 = 4MBx2 = 8MB total L2 cache

It's multiplied by two because there are two cores...pretty straightforward. I don't think Intel has gotten around to having a "shared cache" yet.

 

[$BangforThe$]

Here's how it is:

E6300/E6400 = 2MBx2 = 4MB total L2 cache
E6600/E6700 = 4MBx2 = 8MB total L2 cache

It's multiplied by two because there are two cores...pretty straightforward. I don't think Intel has gotten around to having a "shared cache" yet.

Bona stop just stop it's 2mb x2=4mb and 1mb x2= 2mb. And it is shared cache.

Penryn is said to have 3mb x2=6mb. Now you have been given links that you either refuse to read or your playing stupid either way your wrong .

AMD down 26 cents @ $24

INTC up 29cents @ $18.34

 

[Donnie27]

Here's how it is:

E6300/E6400 = 2MBx2 = 4MB total L2 cache
E6600/E6700 = 4MBx2 = 8MB total L2 cache

It's multiplied by two because there are two cores...pretty straightforward. I don't think Intel has gotten around to having a "shared cache" yet.

NO! I'll write it just like you did but corrected.

E6300/E6400 = 1MBx2 = 2MB total L2 cache
E6600/E6700 = 2MBx2 = 4MB total L2 cache

 

[Bona Fide]

Bona stop just stop it's 2mb x2=4mb and 1mb x2= 2mb. And it is shared cache.

Penryn is said to have 3mb x2=6mb. Now you have been given links that you either refuse to read or your playing stupid either way your wrong .

AMD down 26 cents @ $24

INTC up 29cents @ $18.34

Yeah, I skipped the links...sorry

I retract my previous post.

 

[coldpower27]

NO! I'll write it just like you did but corrected.

E6300/E6400 = 1MBx2 = 2MB total L2 cache
E6600/E6700 = 2MBx2 = 4MB total L2 cache

Argggh.... people still are getting this wrong.

It's not x2 because it's shared cache architecture.

Remember Yonah that used shared cache is 2MB of cache not 1MBx2.... both cores can see the 2MB of cache. The total cache is 2MB for Yonah as well. It's more like this.

Yonah 1x2MB
Allendale 1x2MB
Conroe/Woodcrest 2x2MB

However in Conroe's case both cores can see each block of LV2 cache.

Core 1.
I can see and access Block 1 2MB and Block 2 2MB for a total of 4MB.

Core 2
I can see and access Block 1 2MB and Block 2 2MB for a total of 4MB.

The entire die has 4MB of cache on Conroe becasue both cores are viewing the same blocks. Get the idea?

 

[burningrave101]

Link for hardheaded folks.

http://www.dailytech.com/article.aspx?newsid=1619

What does that have to do with the amount of cache? You're just back onto the MSRP thing again. "Official MSRP" doesn't mean thats what the street price at vendors will be. Its nothing more then a suggested retail price and Intel has no control over how much they are sold for. Retail price will be dependent on the number shipped to vendors vs the number purchased by end users. One of those variables being much higher then the other will equate to whether or not prices are close to MSRP or higher then that. Its simple first grade economics.

 

[Kanaric]

Argggh.... people still are getting this wrong.

It's not x2 because it's shared cache architecture.

He knows that, he just said 1x2mb because some people refused to believe a dual core has a shared cache.

 

[coldpower27]

He knows that, he just said 1x2mb because some people refused to believe a dual core has a shared cache.

Really, we'll have to see on that, I don't think it's too hard to understand the concept of shadered cache, each core has access to the full amount of LV2 on the die but physically there is only the amount 1 core sees and not both.

 

[Donnie27]

Really, we'll have to see on that, I don't think it's too hard to understand the concept of shadered cache, each core has access to the full amount of LV2 on the die but physically there is only the amount 1 core sees and not both.

I know it is shared cache. I wrote it just like he did so he would understand that it was only 2 and 4MB's, not 4 and 8. One core could use either all of the cache or just 25% as needed. There is one single L2 Cache that can be smartly alocated between the two cores.

Back to what the Wulf was asking though. Only the 6400 ES has a 4MB L2, the retail model only has a 2MB L2. They didn't have to underclock because as you know, there were 4MB L2 ES chips at Xtreme systems IIRC.

Some folks were speculating that once the cache is filled Conroe isn't so great. IMHO that's bogus because it will then use Smart Memory Addressing to help here as well.

 

[coldpower27]

I know it is shared cache. I wrote it just like he did so he would understand that it was only 2 and 4MB's, not 4 and 8. One core could use either all of the cache or just 25% as needed. There is one single L2 Cache that can be smartly alocated between the two cores.

Back to what the Wulf was asking though. Only the 6400 ES has a 4MB L2, the retail model only has a 2MB L2. They didn't have to underclock because as you know, there were 4MB L2 ES chips at Xtreme systems IIRC.

Some folks were speculating that once the cache is filled Conroe isn't so great. IMHO that's bogus because it will then use Smart Memory Addressing to help here as well.

I am quite ware tht E6400 ES did indeed have 4MB of LV2.

I would say it depends, it looks like as the clockspeed ramps up and Conroe has better RAM the cache becomes less of a factor from the infromation I have seen.

http://www.behardware.com/articles/623-10/intel-core-2-duo-test.html

http://www.matbe.com/articles/lire/306/merom-et-conroe-test-des-core-2-duo/page13.php

That's mt take on it for now, though even with "only 2MB" Core 2 Duo E6300 is still around 4400+ level give or take depending on the benchmark being run. Though Socket AM2 suffers alot without DDR2-800 4-4-4-12.

 

[Donnie27]

I am quite ware tht E6400 ES did indeed have 4MB of LV2.

I would say it depends, it looks like as the clockspeed ramps up and Conroe has better RAM the cache becomes less of a factor from the infromation I have seen.

http://www.behardware.com/articles/623-10/intel-core-2-duo-test.html

http://www.matbe.com/articles/lire/306/merom-et-conroe-test-des-core-2-duo/page13.php

That's mt take on it for now, though even with "only 2MB" Core 2 Duo E6300 is still around 4400+ level give or take depending on the benchmark being run. Though Socket AM2 suffers alot without DDR2-800 4-4-4-12.

QFT! I wished those links were made a sticky! I saw them

 

[ToastMaster]

om/articles/lire/306/merom-et-conroe-test-des-core-2-duo/page13.php[/url]

That's mt take on it for now, though even with "only 2MB" Core 2 Duo E6300 is still around 4400+ level give or take depending on the benchmark being run. Though Socket AM2 suffers alot without DDR2-800 4-4-4-12.

I wouldn 't really call 87% truly "suffering" without DDR2-800. It is a difference, obviously. Remember however, this is the first official release of AMD's DDRII IMC. It'll take revisions to improve efficiency. Remember, Intel has been using DDRII for long before Conroe, and thus has much greater experience in maximizing the bandwidth difference.

That having been said, the links are impressive.

 

[coldpower27]

I wouldn 't really call 87% truly "suffering" without DDR2-800. It is a difference, obviously. Remember however, this is the first official release of AMD's DDRII IMC. It'll take revisions to improve efficiency. Remember, Intel has been using DDRII for long before Conroe, and thus has much greater experience in maximizing the bandwidth difference.

That having been said, the links are impressive.

A 15% average performance deficit is quite substantial, remember, this would place performance levels below that of DDR-400 2-2-2-5, since if I recall the improvement of DDR2-800 Cas 4 over DDR400 Cas 2 wasn't anymore then 5-8% on average.

I would say it's relative, it is unfortunate, that you need DDR2-667 Cas 4 to about match old DDR-400 Cas 2 performance. Though I guess that is the price you pay for the first revision of something.

 

[ToastMaster]

A 15% average performance deficit is quite substantial, remember, this would place performance levels below that of DDR-400 2-2-2-5, since if I recall the improvement of DDR2-800 Cas 4 over DDR400 Cas 2 wasn't anymore then 5-8% on average.

I would say it's relative, it is unfortunate, that you need DDR2-667 Cas 4 to about match old DDR-400 Cas 2 performance. Though I guess that is the price you pay for the first revision of something.

Do you happen to remember off hand the differences and comparisons for Intel's conversion to DDR-II? It would be interesting simply to see how their transition originally went, since I don't remember caring too much when they originally began switching over for the Pentium 4s.

 

[coldpower27]

Do you happen to remember off hand the differences and comparisons for Intel's conversion to DDR-II? It would be interesting simply to see how their transition originally went, since I don't remember caring too much when they originally began switching over for the Pentium 4s.

From what I remember for Intel since they aren't quite as sensitive to latency. DDR2-533 CAS 4 provides about the same level of performance as DDR-400 CAS2. Though remember at that time Intel was still on 800FSB mostly, so that is to be expected, both comapnies are now getting more memory bandwidth then needed. Intel because it's on a FSB, and AMD because they simply aren't as sensitive to bandwidth, and it was also very expected with the S754 and S939 transistion giving us a clue.