Deciphering the New Cache Hierarchy

The cache hierarchy is a significant deviation from recent previous AMD designs, and most likely to its advantage.  The L1 data cache is both double in size and increased in associativity compared to Bulldozer, as well as being write-back rather than write-through. It also uses an asymmetric load/store implementation, identifying that loads happen more often than stores in the critical paths of most work flows. The instruction cache is no longer shared between two cores as well as doubling in associativity, which should decrease the proportion of cache misses. AMD states that both the L1-D and L1-I are low latency, with details to come.

The L2 cache sits at half a megabyte per core with 8-way associativity, which is double that of Intel’s Skylake which has 256 KB/core and is only 4-way. On the other hand, Intel’s L3/LLC on their high-end Skylake SKUs is at 2 MB/core or 8 MB/CPU, whereas Zen will feature 1 MB/core and both are at 16-way associativity.

Edit 7:18am: Actually, the slide above is being slightly evasive in its description. It doesn't say how many cores the L3 cache is stretched over, or if there is a common LLC between all cores in the chip. However, we have recieved information from a source (which can't be confirmed via public AMD documents) that states that Zen will feature two sets of 8MB L3 cache between two groups of four cores each, giving 16 MB of L3 total. This would means 2 MB/core, but it also implies that there is no last-level unified cache in silicon across all cores, which Intel has. The reasons behind something like this is typically to do with modularity, and being able to scale a core design from low core counts to high core counts. But it would still leave a Zen core with the same L3 cache per core as Intel.

Cache Levels
  Bulldozer
FX-8150
Zen Broadwell-E
i7-6950X
Skylake
i7-6700K
L1 Instruction 64 KB 2-way
per module
64 KB 4-way 32 KB 8-way 32 KB 8-way
L1 Data 16 KB 4-way
Write Through
32 KB 8-way
Write Back
32 KB 8-way
Write-Back
32 KB 8-way
Write-Back
L2 2 MB 16-way
per module
512 KB 8-way 256 KB 8-way 256 KB 4-way
L3 1 MB/core
64-way
1 or 2 MB/core ?
16-way
2.5 MB/core
16/20-way
2 MB/core
16-way

What this means, between the L2 and the L3, is that AMD is putting more lower level cache nearer the core than Intel, and as it is low level it becomes separate to each core which can potentially improve single thread performance. The downside of bigger and lower (but separate) caches is how each of the cores will perform snoop in each other’s large caches to ensure clean data is being passed around and that old data in L3 is not out-of-date. AMD’s big headline number overall is that Zen will offer up to 5x cache bandwidth to a core over previous designs.

Zen High Level Block Diagram Low Power, FinFET and Clock Gating
Comments Locked

216 Comments

View All Comments

  • Reww - Friday, August 19, 2016 - link

    Neither AMD or Intel invented the microprocessor, so they're both copying from someone. Now that we cleared that up, everyone can stfu about copying.
  • BillBear - Saturday, August 20, 2016 - link

    I will be thrilled to see AMD be competitive on more than price. If AMD is also competitive on performance it's a huge win for consumers.
  • SlyNine - Saturday, August 20, 2016 - link

    I almost expected Anand himself to come back and review this one.
  • FireSnake - Monday, August 22, 2016 - link

    Where did he go, anyway? Does anybody know?
  • patel21 - Tuesday, August 23, 2016 - link

    Some commenters say he is working for Apple now
  • Johan Steyn - Monday, August 22, 2016 - link

    So many people here are defending Intel. Yes AMD has floundered. They have been poor competition to Intel. They are are struggling and maybe even a dying company. It will be a miracle if this chip will be successful, yet I do believe in miracles. I just hate having an Intel CPU in my notebook.

    Why is this so? Intel is the bully in town and they bullied AMD to death (almost). I have been in this business at that time. Companies were basically forced not to sell AMD. Intel was found guilty of it and got a slap on the wrist for it. $1B is nothing for them. For this I would welcome the day Intel dies a slow (make it rather quicker) and painful death. But this will probably not happen.

    People say it it is just business, well it is in my books not ethical business even though it might be legal. It was even found to be illegal, yet with it they killed their opponent. These days many contractors do the same. When they build a building, the law requires a certain amount of parking space (in our country). If they do not do this, they are fined. Parking brings in little compensation and therefore they rather pay the fines, even if the fine are relatively high. This is what Intel did. They new they did wrong, but also new that the repercussions will be minimal. It was worth it for them to kill the competition by breaking the law and be fined. Intel might be your hero, not mine.

    This is sickening. Intel makes me sick. I really hope AMD has some success with Zen, even though I think Intel will find another devious way to curb AMD's success. I even hope ARM will eventually dethrone Intel.
  • Outlander_04 - Monday, August 22, 2016 - link

    AMD have surpassed intel in the past . Some of us are old enough to remember 1800 Mhz Athlon 64's smashing intel P4's running at 3000+ MHz .
    We also remember intels response that saw them bribe oems to continue using their crappy processors by sending back bags of cash to people still buying from them .
    We also remember the fines and penalties intel eventually paid for their price fixing. Price fixing that cost their fanboys because it kept the prricee of theeir processors aartificiaally high eveen though they were junk .

    A strong AMD is in everyones benefit . We will get more powerful processors and we will get them at a reasonable price . Lets hope ZEN is even better than it seems
  • sharath.naik - Tuesday, August 23, 2016 - link

    The Problem with AMD is that being a technical company, they should have realized lying repeatedly in the name of marketing about the performance of their products, is akin to crying wolf. For now, it does not matter if they actually have a good product or not. The General assumption is that this is going to be another falsehood, and likely their chip can match intel at 3 ghz only when one core is running (That too when turbo is disabled on intel). And will fall far behind both in single and multithread when there is no trubo restriction on the intel chip
  • slyronit - Tuesday, August 23, 2016 - link

    I agree with you, but if there's something that can kill Intel at this point, it would be ARM based chips, not AMD.
  • atomsymbol - Monday, August 22, 2016 - link

    Bulldozer and Piledriver have a write-through L1D cache. Pentium4 has a write-through L1D cache. Zen has a write-back L1D cache. Skylake has a write-back L1D cache.

Log in

Don't have an account? Sign up now