Barcelona Architecture: AMD on the Counterattack
by Anand Lal Shimpi on March 1, 2007 12:05 AM EST- Posted in
- CPUs
Getting Spendy with Transistors - L3 cache
AMD lost the cache race to Intel long ago, but that's more of a result of manufacturing capacity than anything else. AMD knew it could not compete with Intel's ability to churn out more transistors on smaller processes faster, so it did the next best thing and integrated a memory controller. With the K8's on-die memory controller, AMD reduced the need for larger caches, which is why even current Athlon 64 X2s only have a 512KB L2 cache per core - a figure that Intel introduced back in 2002 with its Northwood core.
These days two Core 2 cores share up to 4MB of L2 cache, while the fastest offerings from AMD weigh in at half that. The gap will continue to widen with Barcelona, as each of its four cores will only have a 512KB L2 cache. While a quad-core Barcelona chip will have 2MB of total L2 cache for all four cores, a quad-core Kentsfield currently has 8MB of L2 cache for all four cores. By the end of this year, Intel's Penryn is expected to have 12MB of L2 cache for all of its cores.
In order to keep die sizes manageable, AMD constructed its quad-core Barcelona out of four cores each with a 128KB L1 and 512KB L2, much like most mainstream K8 based products today. However, the era of multithreaded applications demands that multi-core CPUs should have some common pool of high speed memory to keep them running at peak efficiency.
With four cores sharing a single die, AMD didn't want to complicate its design by introducing a large unified L2 cache. Instead, it took the K8 cache hierarchy and added a third level of cache to the mix - shared among all four cores. At 65nm, a quad-core Barcelona will have a 2MB L3 cache that is shared by all four cores.
The hierarchy in Barcelona works like this: the L2 caches are filled with victims from the L1 cache. When a cache gets full, data that was not recently used is evicted to make room for new data that the cache controller determines is good to keep in the cache. In a victim cache structure, the evicted data is placed in a storage area known as a victim cache instead of being removed from cache all together. If the data should become useful again, the cache controller simply has to fetch it from the victim cache rather than much slower main memory; victims from Barcelona's L1 are kicked out to the L2 cache.
The new L3 cache, acts as a victim for the L2 cache. So when the small L2 cache fills up, evicted data is sent to the larger L3 cache where it is kept until space is needed. The algorithms that govern the L3 cache's operation are designed to accommodate data that is likely to be needed by multiple cores. If the CPU fetches a bit of code, a copy is left in the L3 cache since the code is likely to be shared among the four cores. Pure data load requests however go through a separate process. The cache controller looks at history and if the data has been shared before, a copy will be left in the L3 cache; otherwise it will be invalidated.
Associativity hasn't been changed for the L1 and L2 caches; they are still 2-way and 16-way set associative, respectively. However, the new L3 cache is 32-way set associative. It has been designed to increase the hit rate of a relatively small cache compared to its competition.
AMD lost the cache race to Intel long ago, but that's more of a result of manufacturing capacity than anything else. AMD knew it could not compete with Intel's ability to churn out more transistors on smaller processes faster, so it did the next best thing and integrated a memory controller. With the K8's on-die memory controller, AMD reduced the need for larger caches, which is why even current Athlon 64 X2s only have a 512KB L2 cache per core - a figure that Intel introduced back in 2002 with its Northwood core.
These days two Core 2 cores share up to 4MB of L2 cache, while the fastest offerings from AMD weigh in at half that. The gap will continue to widen with Barcelona, as each of its four cores will only have a 512KB L2 cache. While a quad-core Barcelona chip will have 2MB of total L2 cache for all four cores, a quad-core Kentsfield currently has 8MB of L2 cache for all four cores. By the end of this year, Intel's Penryn is expected to have 12MB of L2 cache for all of its cores.
In order to keep die sizes manageable, AMD constructed its quad-core Barcelona out of four cores each with a 128KB L1 and 512KB L2, much like most mainstream K8 based products today. However, the era of multithreaded applications demands that multi-core CPUs should have some common pool of high speed memory to keep them running at peak efficiency.
With four cores sharing a single die, AMD didn't want to complicate its design by introducing a large unified L2 cache. Instead, it took the K8 cache hierarchy and added a third level of cache to the mix - shared among all four cores. At 65nm, a quad-core Barcelona will have a 2MB L3 cache that is shared by all four cores.
The hierarchy in Barcelona works like this: the L2 caches are filled with victims from the L1 cache. When a cache gets full, data that was not recently used is evicted to make room for new data that the cache controller determines is good to keep in the cache. In a victim cache structure, the evicted data is placed in a storage area known as a victim cache instead of being removed from cache all together. If the data should become useful again, the cache controller simply has to fetch it from the victim cache rather than much slower main memory; victims from Barcelona's L1 are kicked out to the L2 cache.
The new L3 cache, acts as a victim for the L2 cache. So when the small L2 cache fills up, evicted data is sent to the larger L3 cache where it is kept until space is needed. The algorithms that govern the L3 cache's operation are designed to accommodate data that is likely to be needed by multiple cores. If the CPU fetches a bit of code, a copy is left in the L3 cache since the code is likely to be shared among the four cores. Pure data load requests however go through a separate process. The cache controller looks at history and if the data has been shared before, a copy will be left in the L3 cache; otherwise it will be invalidated.
Associativity hasn't been changed for the L1 and L2 caches; they are still 2-way and 16-way set associative, respectively. However, the new L3 cache is 32-way set associative. It has been designed to increase the hit rate of a relatively small cache compared to its competition.
Facebook
Twitter
RSS
83 Comments
View All Comments
Amiteriver - Tuesday, March 27, 2007 - link
Sounds groooovyNow lets just hope they have something good to plug it into.
trisweb2 - Friday, March 16, 2007 - link
I just want to say how refreshing it is to read an article written by Anand. He is a master of the English language; he perfectly communicates and explains every technical detail and I come away with a better understanding of whatever he's talking about.Thank you, Anand, for being a good writer!
MrWizard6600 - Thursday, March 22, 2007 - link
I Agree, Outstanding.No other site I know of gives nearly as many in depth details, and while ill admit my knowlage of some of the terms is sketchy, I got through that one with a good understanding.
Sounds like AMD has something to fight Core 2 against.
I do have one criticism:
I would have loved to have heard what Intels equivilent to all of AMDs technologies would be, mind you this criticism corrects it self toward the end of the artical.
stance - Monday, March 5, 2007 - link
Remember AMD's old president and CEO Jerry Sanders with commentslike "We will see what we see" and "More bang for your buck" I
cannot wait to see duel socket motherboards with two four core
Barcelona's working their magic. reminds me of Carol shelby
when he brought the Cobra out for road test. exciting is not
the word, jaw droping performance? Don't take Richard's Statements
lightly
lordsnow - Sunday, March 4, 2007 - link
Does anyone have any idea how compatible the "Barcelona" CPU will be with current motherboards? When it comes out, does it need a new n-phase voltage regulator, for example?the reason I'm asking is, I want to upgrade and with the current state of affairs was going to go for a C2D CPU. But with these Barcelona CPU's due out I may stick with AMD - get a AM2 motherboard and cheap AM2 CPU and upgrade to the Barcelona CPU at a later date. But I have to be sure that whatever motherboard I buy now will be 100% Barcelona compatible.
Can anyone inform us about what the situation is in this regard?
coldpower27 - Sunday, March 4, 2007 - link
Barcelona being the server variant will be compatible with the Socket F infrastructure, while Agena will be a Socket AM2+ processor compatible with exisiting Socket AM2 infrastructure.lordsnow - Sunday, March 4, 2007 - link
Any ideas as to what kind of features a user will be missing by dropping a AM2+ "Agena" CPU into a AM2 socket? The enhanced Power Saving features, perhaps?chucky2 - Sunday, March 4, 2007 - link
I asked above and non-AnandTech folks like you and I said it would...but no one from AnandTech themselves jumped right in to give an affirmative.I asked for links from AMD's own website confirming that Agena and Kuma would work in current AM2 motherboards, and no one posted back.
Right now the AM2+ CPU's will work in current AM2 boards rumor is just that, a rumor...when AMD themselves confirm it, or a site such as AnandTech confirms it with AMD and reports on it, then I'll believe it.
Until then, it's <i>probable</i> that AM2+ will work in current AM2 motherboards...if you're willing to take the risk I say go for it, else, wait until we have an official answer one way or the other.
JMHO...
Chuck
Calin - Saturday, March 3, 2007 - link
"Intel regained the undisputed performance crown it hadn't seen ever since the debut of AMD's Athlon 64."Intel in fact lost the "undisputed performance king" title during the early lifetime of the K7 architecture. The Pentium !!! was faster at some tasks and slower at others (games) than the K7. Before that, the Pentium II was better than the K6-2 (the K6-3 had better IPC than Pentium3, but was slower in MHz)
coldpower27 - Sunday, March 4, 2007 - link
Intel had the undisputed performance crown again with the Athlon XP 3200+ vs the Pentium 4 3.0C/3.2C and higher processors.