Depends on your definition of a "true core". Bulldozer had 8 physical cores but it wasn't a "true octa-core". Zen still uses the module system, it's just a lot more robust.

To be clear I didn't mean true 8 core as in it's all one die, it could be 2 4 core dies on one package. By true 8 core I meant each core is independent and doesn't need to share important components like Bulldozer did. BDs sharing of FPUs was crippling.

Clock for clock it probably won't be. But it'll likely be a true 8 core chip. I'm just hoping for high enough clocks for it to be comparable to Intel in single threaded performance even if it has lower IPC. So many things are still single threaded.

If they can pass even the lga 1366 xeon IPC, It would be enough. Even a xeon x56xx @ 4.4ghz is equivalent to 6700k@4.8ghz (cinebench r15)

To be clear I didn't mean true 8 core as in it's all one die, it could be 2 4 core dies on one package. By true 8 core I meant each core is independent and doesn't need to share important components like Bulldozer did. BDs sharing of FPUs was crippling.

Alright. But still, last I heard Zen still shares the fetch, decode, and L2 cache for each pair of cores. I don't think the FPUs are shared, though. I don't think AMD is stupid enough to make the same big mistake twice, so I think these new modules will probably perform fine. @Ryu5uzaku Yeah, if you've got an APU then single-channel is seriously crippling. But if you're not using the IGP then you don't really need dual channel. In real-world cases, you'd get more performance overclocking your RAM than adding another channel. Multi-channel memory for only CPUs tends to only look good on synthetic benchmarks.

Going to challenge the i7 5960X ey :P

All my CPUs before my Intel E6750 Core Due were AMD, and everything after was Intel. I still have the Athlon 64 boxes (had 2 of those), not to mention the old K7s, K6s etc. So yeah, I truly hope they will make a comeback. Fingers crossed. 🙂

Alright. But still, last I heard Zen still shares the fetch, decode, and L2 cache for each pair of cores. I don't think the FPUs are shared, though. I don't think AMD is stupid enough to make the same big mistake twice, so I think these new modules will probably perform fine. @Ryu5uzaku Yeah, if you've got an APU then single-channel is seriously crippling. But if you're not using the IGP then you don't really need dual channel. In real-world cases, you'd get more performance overclocking your RAM than adding another channel. Multi-channel memory for only CPUs tends to only look good on synthetic benchmarks.

Where did you read that Zen are using modules like bulldozer? Officially Zen is made with separates core with 6 pipes. Now is clear 6 pipes means 4 integer pipes and 2 AGU pipes. Why do you think that was a mistake for AMD to share the FPU? Intel does the same but resources of FPU are shared per thread. A quad core Intel has 4 FPUs shared by 8 threads and a quad module AMD (8 bulldozer integer cores) has the same amount of 4 FPUs, is just as simple .... AMD FPUs are weaker it doesn't have anything to do with sharing the FPU with 2 integer cores because intger and floating point pipe have in common only the frontend , decoders and schedulers. i7 with 4 FPUs is quicker than AMD with 4 modules and 4 FPUs , bulldozer. Anyway stars arhitecture (K10 )had also 6 pipes , but 3 integer ones and 3 AGU and the FPU was made by 3 pipes. In this case Zen will be 33% quicker than a K10 clock by clock . SHould be a nice bump (but we are talking about a very old arhitecture) , I don't know if is enough to compete with modern Intel processors.

Where did you read that Zen are using modules like bulldozer? Officially Zen is made with separates core with 6 pipes. Now is clear 6 pipes means 4 integer pipes and 2 AGU pipes.

http://www.extremetech.com/gaming/204523-new-leak-hints-at-amd-zens-architecture-organization It seems they aren't called modules anymore, but it's still "module-like".

Why do you think that was a mistake for AMD to share the FPU? Intel does the same but resources of FPU are shared per thread.

I meant more of a marketing mistake. Technically, AMD's design was superior. When you created software designed for Bulldozer, it was really fast. But the problem is the vast majority of software wasn't designed to work with AMD's module system. In some cases the octa core CPU would perform more like a quad core, because depending on the load, only half of each module could be utilized. AMD's design in theory was fantastic. If it were a fresh new architecture with a new OS and everything, it'd be very successful. It actually is still a decent architecture for some servers. But it otherwise wasn't all that practical. The reason Intel didn't get screwed over by sharing an FPU is because a hyper-threaded core isn't necessarily "parallelized". It's basically just allowing 2 threads to work sequentially, so there are no wasted resources. That's why HT is sometimes known to actually slow down certain tasks, because sometimes processes are actually expecting to run things in parallel, but a hyper-threaded quad core CPU can't literally complete 8 threads simultaneously. Bulldozer was the exact opposite - it could only do things in parallel, so when you've got 8 separate tasks expecting to be run sequentially, you're really only taking advantage of 4 cores instead of 8. In Intel's case, what they did was more practical for everyday use, and therefore it won.

SHould be a nice bump (but we are talking about a very old arhitecture) , I don't know if is enough to compete with modern Intel processors.

Poor AMD. They don't know, that Jim Keller sold them their old architecture 😉 Anyway, it's interesting to compare Pentium III and Core Architectures. http://img.tomshardware.com/us/2000/11/20/intel/p3-diag.gif http://images.anandtech.com/reviews/cpu/intel/core-vs-k8/CoreArchi.gif

http://www.extremetech.com/gaming/204523-new-leak-hints-at-amd-zens-architecture-organization It seems they aren't called modules anymore, but it's still "module-like".

On Excavator (Bulldozer) you can see 2 cores with shared FP scheduler. This was named module. On Zen we can see, not shared decode unit, Int scheduler and FP scheduler. Its full single "core" without any "sharing". With Zen CPU we will have 8 or 4 similar Zen cores in one CPU. edited: 4,8,16 or 32 Zen cores on single CPU.

If Zen gets relatively close to Intel's offerings per core, and the genuine 8-core chip costs as much as Intel's 4-core, it will certainly mean my current PC, built this March, is the lousiest PC I've ever built. But even if it renders my new PC obsolete, I still hope Zen will blow everybody's socks off and AMD really gets back into business. Anything that hurts the stupid Intel is excellent and most welcome news. So far I could only enjoy reading about Intel's suffering in the mobile market.

According to the slide, Summit ridge will have up to 8 "cores" And according to Guru3d, which says each core will have 4 integer units and 4 FP units.. it looks like we could have up to 32 threads on one CPU. That would be sweet!

Could be a potential beast after all.. Well 16threads sounds good, and if its 1.6x AMD FX single core perf. then it could be faster then Haswell/Skylake too. 🤓 sause: well not so good but still somethin http://s33.postimg.org/vgor6huv3/AMD_Zen_Performance_Double_FX_83501.jpg

Compared to AMD’s “Orochi” quad module, eight core die powering the FX 8350, the Zen based desktop Summit Ridge eight core CPU delivers double the performance in Cinebench R15. This means that a single Zen core is in effect equivalent to two Piledriver cores in performance, which is incredibly impressive. This dramatic performance difference comes from the significant architectural performance per clock improvements in addition to Zen’s simultaneous multithreading capability. it’s important to remember that AMD’s latest Orochi dies feature Piledriver cores rather than Excavator. Excavator cores are roughly 15% faster per clock than Piledriver This in turn puts Zen at a lead in excess of 60% vs Piledriver in terms of performance per clock. Doubling the performance of the FX 8350 puts Zen in direct competition with Intel’s eight core i7 5960X.

The article says the front end can decode 4 ops/cycle, which may not be true. Each of the four decoders can decode FastPath, which decodes into multiple ops. Most likely the average will be ~6 ops/cycle unless the decoders can only operate every other cycle. It is quite likely that the connections to the decoders can only handle 6 ops/cycle - even if the decoders could actually decode 8~12 ops/cycle (absolute peak).

Poor AMD. They don't know, that Jim Keller sold them their old architecture 😉 Anyway, it's interesting to compare Pentium III and Core Architectures. http://img.tomshardware.com/us/2000/11/20/intel/p3-diag.gif http://images.anandtech.com/reviews/cpu/intel/core-vs-k8/CoreArchi.gif

Well , Phenom (k10) was competitive only with Core arhitecture, Nehalem just got too faster . Bulldozer was barely able to compete with Nehalem ,not to mention Sandybridge . So , yes , I expect Zen to be somewhere between Sandybridge and Haswell , at most....

Well , Phenom (k10) was competitive only with Core arhitecture, Nehalem just got too faster . Bulldozer was barely able to compete with Nehalem ,not to mention Sandybridge . So , yes , I expect Zen to be somewhere between Sandybridge and Haswell , at most....

Between Sandybridge and Haswell is Ivy Bridge. Considering Skylake is amazing 10% faster than Ivy Bridge (I'd know since I just upgraded from Ivy to Sky), it would place Zen right next to Skylake's performance, especially assuming it overclocks as well as Intel CPUs so that a simple OC wouldn't allow Intel to beat it again. Coupled with double the amount of physical cores, it would be jolly good.

http://anandtech.com/bench/product/287?vs=698 Let me explain the picture and why I'm expecting that Zen should perform somewhere at a Sandybridge level. First I7 2600k has 3.4 Ghz base clock and 3.8 Turbo at 95W . FX 8320 has 3.5 Ghz base clock and 4.0 Turbo at 125W. Even if I7 has lower clock is able to outperform the FX in pure speed (like cinebench render) in both single and multi-thread , but by 50% in single thread . AMD says Zen will offer 1.5x performance per watt over bulldozer . Well that 1.5x performance per watt will put it just around the older Sandybridge . Don't understand me wrong, Sandybridge is still a good processor , between a Sandybridge and Skylake clock for clock is no more than 20% difference

According to the slide, Summit ridge will have up to 8 "cores" And according to Guru3d, which says each core will have 4 integer units and 4 FP units.. it looks like we could have up to 32 threads on one CPU. That would be sweet!

They are actually referring to the ALU, AGU, and various FPU pipelines. They do not get their own threads.

http://anandtech.com/bench/product/287?vs=698 Let me explain the picture and why I'm expecting that Zen should perform somewhere at a Sandybridge level. First I7 2600k has 3.4 Ghz base clock and 3.8 Turbo at 95W . FX 8320 has 3.5 Ghz base clock and 4.0 Turbo at 125W. Even if I7 has lower clock is able to outperform the FX in pure speed (like cinebench render) in both single and multi-thread , but by 50% in single thread . AMD says Zen will offer 1.5x performance per watt over bulldozer . Well that 1.5x performance per watt will put it just around the older Sandybridge . Don't understand me wrong, Sandybridge is still a good processor , between a Sandybridge and Skylake clock for clock is no more than 20% difference

I pretty much came to that same conclusion - after long expecting Haswell-like IPC. I based my expectations upon Excavator's performance and even created performance profiles and some process-derived benefits to create a very specific performance range. http://excavator.looncraz.net/

http://anandtech.com/bench/product/287?vs=698 Let me explain the picture and why I'm expecting that Zen should perform somewhere at a Sandybridge level. First I7 2600k has 3.4 Ghz base clock and 3.8 Turbo at 95W . FX 8320 has 3.5 Ghz base clock and 4.0 Turbo at 125W. Even if I7 has lower clock is able to outperform the FX in pure speed (like cinebench render) in both single and multi-thread , but by 50% in single thread . AMD says Zen will offer 1.5x performance per watt over bulldozer . Well that 1.5x performance per watt will put it just around the older Sandybridge . Don't understand me wrong, Sandybridge is still a good processor , between a Sandybridge and Skylake clock for clock is no more than 20% difference

If Zen offers only 1.5x performance per watt over bulldozer well damn. Jump from 32nm to 14nm was kind of crap then. They did talk about having 40+% faster ipc then excavator clock for clock. So that would bring Zen to 50-60% clock for clock boost. Which would bring it closer to haswel/skylake level. "Compared to AMD’s “Orochi” quad module, eight core die powering the FX 8350, the Zen based desktop Summit Ridge eight core CPU delivers double the performance in Cinebench R15. This means that a single Zen core is in effect equivalent to two Piledriver cores in performance, which is incredibly impressive. " If that is even close to truth AMD has one good processor incoming in Zen. Bristol ridge vs kaveri I can understand the 1.5x performance per watt, it's a 28nm vs 28nm other is just Excavator core.

Even if Zen comes with SB performance, with DDR4 and the latest chipsets and controllers it will still be a very viable option. I'm sure many have upgraded from SB not because Skylake or Haswell offer that much more performance but because of all the goodies you get when you upgrade to these modern chipsets.