What would happen if AMD up the cache to 6mb in their CPU?

I read a while ago that AMD CPUs use their cache in a different way compared intel.

Would increasing the cache makes AMD processors as fast as Intels?

leave your comments and links to info you want to share with the NBR members.

Thanks

 

Wow, you're a long time member, I'm surprised you haven't been more active at this great forum.

To answer your post, I don't have any hard numbers, but I don't believe upping their cache would make them perform on par with intel, as clock-for-clock amd processors are slower (albeit not a whole lot, and amd is still a great value for the money, especially with systems using ati integrated graphics).

 

If it were that easy, they would have done it a long time ago.

 

AMD as part of their strategy have less L2. As said not that simple. It is too late to explain now. AMD has on board memory controller Intel does not? if you really need to know I will post back tomorrow. But trust us.

Edit: Who am I kidding? Not going to sleep.

AMD would do if they could, they can't. Having the onboard memory limits space. AMD has got up to 1GB per core. Also remember the AMD is the amount per core no sharing? Intel shares the entire amount. So many differences. AMD has no FSB? What replaced it runs between 1,600Mhz and 2,000Mhz? Intel just got up to 1066Mhz? Consider?

I am now tired gotta go.

 

1- I pop in the forum only between now and then as I don't and I don't want to have internet at home. I know what you may think, it's the 21st century but some people have adection problems with cigarets, me it's the internet.
2- Now I'm in holiday with a 24/7 access to the net.

one of the reasons I'm asking is at the moment AMD and Intel are different, but soon Intel will add the memory controller and the L3 cache to their CPU which sound that the gap between AMD and intel will be even greater.

I trust you guys but I want links and explaination to understand why intel need and can use a big cache, while amd feel that a small cache is better, in particular that it seems that a bigger cache boost the performance even in the case of AMD CPUs. Here I'm referring to the boost in performance (small or big) observed in similarly clocked Sempron and Athlon processors that have different cache size.

 

powerpack has summed it up quite well. AMD's architecture makes it less necessary to have a massive on-board cache. The incremental benefit of a bigger cache is unlikely to be worth the cost. And more cache is likely to increase the power consumption.

John

 

Does this means that the 2mb L3 cache in the Phenom is just a markiting talk?

 

There will benefit from the extra cache, but the benefits will steadily reduce for each increase in cache size. Here we are talking adding another 1MB of cache, not the 5MB in your original question.

John

 

AMD introduced the K8 architecture in 2003 which added many improvements, most notably the integrated memory controller, hypertransport and the x86-64 architecture that could also run legacy 32-bit applications.

One of the most expensive operations in software is accessing memory. Main memory latencies can run from 70 to 270 (roughly) cycles. CPU register operations typically take from 1 to 10 cycles with lots of instructions using only 1 cycle. L1 cache latencies are typically in the single digits. L2 latencies are typically in the low double digits. Previous architectures used an off-chip memory controller that added to main memory latencies. AMD moved the memory controller on-chip significantly reducing memory latencies. The K8 architecture leapfrogged Intel's then current Pentium 4 architecture. Other advantages that K8 had were the use of a shorter pipeline which reduced possible clockspeeds but enabled higher IPC (instructions per clock).

Intel's responded with Core 2 Duo in 2006 which had a number of IPC improvements. They cranked up L2 cache and added aggressive memory prefetchers that would speculatively fetch data from main memory to L2 in anticipation of it being used. They improved instruction latencies for vector instructions and added a fourth execution engine for instructions. So they made a bunch of improvement besides cranking up the cache.

The thing that allowed them to add transistors for additional cache and IPC improvements was their advantage in process. They were at 65 nm while AMD was at 90 nm which meant that they could pack twice as many transistors into the same amount of space. So AMD could increase cache but it would cost them a lot of money to do so. Wafers cost around $5,000 and you can charge a certain amount for a chip. If you make the chip bigger, then you will wind up with fewer chips and so each individual chip will cost you more to make.

Intel further increased their process lead with 45 nm and Hi-K/Metal Gate technology allowing them to reach crazy-high frequencies. AMD has been working on catching up on the process node but it appears that they don't have anything like Hi-K/Metal Gate.

Getting back to your main question though: AMD wouldn't see as much of a benefit from larger caches because the integrated memory controller already reduces latencies by quite a bit. Intel shouldn't get the huge bump that AMD got when they introduced the IMC because they already get the benefit of larger caches.

One other thing is that I've heard that larger caches typically mean longer cache latencies. So making the cache exponentially larger has performance costs as well as benefits. Some of Intels chips in their Itanium line have really huge caches but those chips are designed for different workloads and probably take advantage of huge caches in ways that aren't seen in desktop and mobile applications.

(this is to the best of my knowledge - if you see a mistake, let me know)

 

AMD runs very hot for laptops..even Turion TL56 is much hotter than Intel equivalent.

thats the main disadvantage ..nothing else

 

but AMD undervolte better which is not the case for Intel.

 

AMD has put their focus on desktop and server chips. They have some advantages on servers compared to Intel but those tend to be applications where a lot of bandwidth is required. It is hard to recommend a laptop using AMD chips unless you're taking advantage of their graphics solutions in a fusion environment (when it comes out).

BTW, I have an AMD Athlon 64 3200+ (Compaq Presario). It's a nice machine for its time and I still use it for web browsing. It's has an 89 watt processor. Modern Intel mobile processors use 25 to 45 watts. My old laptop is about an inch and 3/4s thick and weighs about 8 pounds. Lots of cooling capacity but no one would call it thin or light.

 

Absolutely correct. I have a quad opteron array in my Tyan Tomcat board, and it cranks. AMD has some of the best dual core server processors on the market. Actually their 2200 series Opterons are the fastest dual core processors made for servers.
AMD's quad core line is not very good, but they are really inexpensive, so they are affordable and practicle for a large amount of people.

For server apps amd is the way to go.
We use AMD processors at the architecture firm I work at, and the server is really fast, even with the 430 employees in the firm.
There is nothing like 64 dual core Opteron processors and a 18 terabyte array of 15,000rpm Hitachi Ultrastar's.

I love to launch 200mb cad files from it, cause they load in like 10 seconds, and it would take my system at home like 5 minutes to open.

K-TRON

 

I tought I was up to date; but it seems I not.

If I recall well, all CPUs benchmark in the special Processors edition of Custom PC showed that all AMD CPUs including the opetron 2000 and 5000 series was lugging behind Intel Core 2 Duo and Xeon

Who won't like something like this at home

 

I tried undervolting ..i couldnt find a sweet spot for my laptop & had BSODs also..

will try it again when free on a day then.

thanks.

 

I was able to get a sempron 3400+ an a Turion MK36 stable @ 0.8v. And someone in the forum was able to go below that with a Turion X2. I think he posted in one of the undervoling threads.

 

I think their 45NM cpus will have more cache. Shangai(server) will have 6MB L3 cache(shared). i think current notebook cpus have 1MB per core. With IMC u dont need huge amount of cache. I am sure intel's mainstream cpus will have less cache than penryn based counterparts.

 

Current notebook chips run 512K to 1 MB per core. Some might even run less in the Sempron line. Cache still helps - especially if your application's working set fits in cache. Intel can do whatever they want to on cache.

 

I doubt that intel will ever reduce their cache even after the integration of the memory controller in the cpu chip.

Intel is using the cache size as a markting argument because it knows that for most people bigger it better and more is better. I think this what helped them to sell a lot of P4 (higher clock speed... faster is better) although they were worst than AMD at the time.

 

Intel drops the cache size on their cheaper chips so it is marketing in a sense. If customers see cheaper chips/systems with less cache, then they think that more cache is better.

 

Go to bed, Imagine 1GB cache , I wonder if it will ever get a high as that!!

 

Leave him alone. It was late and he/she was tired

 

If we follow Moors low (correct me if wrong) the amount of memory should double every year.

Taking in consideration that an intel T9000 series has 6mb L2 cache, 1gb L2cache should be acheivable in 8 years.