AMD Zen will get 8 channel DDR4 support and SMT says CERN employee
An interesting snippet of news just before the weekend starts, AMD ZEN based processors will get up-to 8 channels of DDR4 support as well as SMT. The info was spilled by CERN engineer Liviu Valsan who in a recent presentation on datacenter hardware trends shared a thing or two about AMD's upcoming Zen processor architecture.
According to a slide from Liviu the upcoming x86 processors based on Zen will feature of up to 32 physical cores, however Valsan stated AMD will use two 16-core CPUs on a single die, so that's a bit of an old trick really. But it does confirm earlier findings. It however immediately places the 8 memory channels in debate, as you might look at it as two quad channel setups then.
Processor cores wise initially far more realistic would be four, eight and perhaps in the no too distant future 16 cores. Interesting is the mention of an SMT design. AMD already slightly hinted towards it, now it is confirmed. SMT means symmetrical multi-threading and it would be the equivalent to Intel's Hyper-threading. The Zen architecture will be built on a more efficient 14 nanometer FinFET process, rather than the 32 nm and 28 nm processes of previous AMD FX CPUs and AMD APUs, respectively.
The Zen family processors for consumers will feature a unified AM4 socket with its GPU-equipped "Bristol Ridge" APU counterparts, and feature DDR4 support and a 95W TDP. While newer roadmaps don't confirm the TDP for desktop products, they suggest a range for low-power mobile products with up to two Zen cores from 5 to 15W and 15 to 35W for performance-oriented mobile products with up to four Zen cores.
Each Zen core will have four integer units, two address generation units and four floating point units, and the decoder can decode four instructions per clock cycle. L1 data cache size is 32 KiB and L2 cache size 512 KiB per core. Two of the floating point units are adders, two are multipliers.
The completely new design will be 40% faster per core / instruction / clock cycle compared to the current Excavator cores, and that would be a serious increase alright. The throughput should increase significantly thanks to Simultaneous Multithreading, so yeah all things considered, this is looking good and sound with this architecture.
AMD Zen Can Have 32 Cores per Socket - 02/03/2016 09:45 AM
Can .. I wrote 'can have' .. Anyway a leaked Linux patch on LKML.org, first spotted by The New Citavia Blog, suggests that AMD Zen based processors will feature up to 32 physical cores. The patch a...
AMD Zen architecture processors are true quad-core CPUs - 04/29/2015 09:36 AM
A number of slides about AMDs upcoming ZEN APU architecture have been shared on AMDs Financial Analyst day. Zen is the code name for a new 14nm x86 micro-architecture being designed by AMD from the g...
Senior Member
Posts: 491
Joined: 2006-06-23
Would be great to see some super-duper APU's, especially if they use those for the next gen of consoles. Even if you don't play on consoles, better performance on that side should translate to better PC ports. Like The Division dev said they didn't want to create too big difference between platforms, ect.
Maybe.
Senior Member
Posts: 13599
Joined: 2005-08-13
Seeing as AMD's IGPs have always been crippled by memory, I think the extra channels will make a pretty big difference (though, 8 might be overkill...).
Without the IGP, 8 channels just sounds stupid, even for 32 cores. For the average CPU task, you can hardly ever notice a performance difference between 1 and 2 channels. I'm not talking about synthetic benchmarks here, I'm talking about real-world performance.
AM4 platform will almost certainly be dual channel DDR4 only. Server platform is another story.
Senior Member
Posts: 1443
Joined: 2014-07-22
Even if they deliver what they promise, a 40%+ increase in IPC, it's still not going to catch up to Intel as far as per core performance goes.
Bulldozer at 4GHz Vs Sky Lake at 4GHz - http://www.anandtech.com/bench/product/697?vs=1543
Just about everything still seems to be single or dual threaded. Single core performance matters too much and AMD still seem to be playing the moar coars game.
Remember that the 40% is simply from the architectural changes...has nothing to do with clocks. And the 40% comes from the implementation of SMT--this is something AMD has stated clearly several times in 2015. I don't think there's a question that at the very least it will bring AMD to performance parity with Intel. So just think of that while AMD undercuts Intel pricing by 10%-20%...

Big deal will the be the AM4 chipset, too, and how that performs. Also, remember that in UHD gaming and up, the cpu makes little to any difference at all as those resolutions are GPU-limited. No question but that the higher the resolution the less the cpu matters to the performance equation.
Also, benchmarks tend not to concentrate on the software most people who buy cpus actually use and so can be highly misleading for many reasons. Advising today I'd tell people who want to game at high resolutions to go cheap on the cpu and spend the big money on the GPU...

Senior Member
Posts: 3404
Joined: 2013-03-10
Even if they deliver what they promise, a 40%+ increase in IPC, it's still not going to catch up to Intel as far as per core performance goes.
Just about everything still seems to be single or dual threaded. Single core performance matters too much and AMD still seem to be playing the moar coars game.
Even if it's still slightly slower per core, I'd still take an AMD since I could buy a genuine 8-core for the same price that Intel is asking for a 4-core with the placebo hyperthreading function. The single core difference won't be that huge anymore, unlike right now when nobody serious gets an AMD CPU, diehard AMD fans aside. These days more than 4 cores might start to help even in the newest games, in non-game software they have been put to use for years already, despite Intel's best efforts to tell people nobody needs more than four cores.
Senior Member
Posts: 7425
Joined: 2012-11-10
Seeing as AMD's IGPs have always been crippled by memory, I think the extra channels will make a pretty big difference (though, 8 might be overkill...).
Without the IGP, 8 channels just sounds stupid, even for 32 cores. For the average CPU task, you can hardly ever notice a performance difference between 1 and 2 channels. I'm not talking about synthetic benchmarks here, I'm talking about real-world performance.