Monday, October 5th 2015
AMD Zen Features Double the Per-core Number Crunching Machinery to Predecessor
AMD "Zen" CPU micro-architecture has a design focus on significantly increasing per-core performance, particularly per-core number-crunching performance, according to a 3DCenter.org report. It sees a near doubling of the number of decoder, ALU, and floating-point units per-core, compared to its predecessor. In essence, the a Zen core is AMD's idea of "what if a Steamroller module of two cores was just one big core, and supported SMT instead."
In the micro-architectures following "Bulldozer," which debuted with the company's first FX-series socket AM3+ processors, and running up to "Excavator," which will debut with the company's "Carrizo" APUs, AMD's approach to CPU cores involved modules, which packed two physical cores, with a combination of dedicated and shared resources between them. It was intended to take Intel's Core 2 idea of combining two cores into an indivisible unit further.
AMD's approach was less than stellar, and was hit by implementation problems, where software sequentially loaded cores in a multi-module processor, resulting in a less than optimal scenario than if they were to load one core per module first, and then load additional cores across modules. AMD's workaround tricked software (particularly OS schedulers) into thinking that a "module" was a "core" which had two "threads" (eg: an eight-core FX-8350 would be seen by software as a 4-core processor with 8 threads).
In AMD's latest approach with "Zen," the company did away with the barriers that separated two cores within a module. It's one big monolithic core, with 4 decoders (parts which tell the core what to do), 4 ALUs ("Bulldozer" had two per core), and four 128-bit wide floating-point units, clubbed in two 256-bit FMACs. This approach nearly doubles the per-core number-crunching muscle. AMD implemented an Intel-like SMT technology, which works very similar to HyperThreading.
Source:
3DCenter.org
In the micro-architectures following "Bulldozer," which debuted with the company's first FX-series socket AM3+ processors, and running up to "Excavator," which will debut with the company's "Carrizo" APUs, AMD's approach to CPU cores involved modules, which packed two physical cores, with a combination of dedicated and shared resources between them. It was intended to take Intel's Core 2 idea of combining two cores into an indivisible unit further.
In AMD's latest approach with "Zen," the company did away with the barriers that separated two cores within a module. It's one big monolithic core, with 4 decoders (parts which tell the core what to do), 4 ALUs ("Bulldozer" had two per core), and four 128-bit wide floating-point units, clubbed in two 256-bit FMACs. This approach nearly doubles the per-core number-crunching muscle. AMD implemented an Intel-like SMT technology, which works very similar to HyperThreading.
85 Comments on AMD Zen Features Double the Per-core Number Crunching Machinery to Predecessor
SB level performance on a per core basis will be acceptable, assuming the power envelope for a quad core is around the same or lower than SKL, but they will need to be pushing at least 8 core parts and should consider 24-32 pcie lanes for higher end GPU setups.
Also, they need to be considering at least triple channel memory, especially if they want to put the APUs in the same socket, otherwise there will be 0 reason to not go with an intel part and dGPU. That said, intel should also be considering the same if they want to push their iGPU market forward. And this is all on top of the things skylake already brings to the table as far as chipset connectivity
HyperTransport is essentially the equivalent to DMI when AMD moves to an integrated northbridge (memory/PCIe controller) so to accommodate the faster chipset interfaces (PCIe3/SATAe/M.2, high sata 6g count, USB3.1, etc) they will NEED to make improvements there.
Of course, the APUs have been updated much more frequently, and have over time had the PCIe controller integrated into the CPU. Rumours say that they want to integrate the SouthBridge too at some point, which would be quite fun to see.
The DMI equivalent on AMD's side would be A-Link Express or UMI, basically a tweaked PCIe 4x interface.There's no Intel preference, just that AMD didn't put in as much FP hardware in Bulldozer (based on the bet that basically everyone would port FP code to GPUs). If you compare a 6-core Thuban to a 6-core Bulldozer, The older Thuban core beats the Bulldozer in Cinebench MT.
Intel has not been defeated in court, they've merely settled for 1 billion $. (a laughable sum, compared to the damage done)
There was hardly anything new as far as anti-competitive practices go in times of Prescott, AMD had the same problems prior to that.
Neither did nVidia lose much of the market share, despite Fermi fiasco.
PS
On a side note. Does any AMD CPU beat i5 750 (a 45nm product) IPC wise? Or have something with comparable performance but better perf/watt?
Fermi succeeded despite it's issues because of CUDA, even though nVidia lost a solid 10% marketshare (thanks to dbz over from beyond3d). In comparison, AMD's Stream API was hopeless, and OpenCL was just a pipe dream
On the i7-750, the FX-8150 beats it in some places, and loses out in others. An FX-8350 in comparison is either neck and neck or faster
CUDA argument is laughable.
As a bottom line, we see slower more expensive more power hungry products by AMD competitors only mildly affect the market share (10%-ish swing, if at all) while AMD losing one of the points can lead to it's market share to be cut in half, this looks particularly bad in graphic cards market, where AMD has more than competitive products.
.
What is a pipe dream is AMD thinking that they could make a pipeline as long as Netburst but not encounter the same problems as Netburst. That's what I call a pipe dream.
Zen, could make or break my investment future. Or, Not. I figure 1 to 2 Hundred of investment, now, could be a big deal or a minor loss.
Putting it on my Monday Calendar. I am gonna Bank on the new Zen chip. There's just too much to possibly lose, not doing it!
:toast:
www.investopedia.com/stock-analysis/032515/3-major-problems-facing-advanced-micro-devices-amd.aspx
"AMD competes with Intel in PC and server processors but has been bleeding share in both markets. In servers, AMD won as high as 25% market share in 2006, making it a major player in the industry. Today, AMD is essentially nonexistent in the segment, claiming a low single-digit share. Meanwhile, Intel has built a near-monopoly, allowing it to charge high prices and extract extremely high margins.
In PCs, the story is much the same. Intel has continued to steal market share from AMD, especially at the low end with its Atom chips, despite already controlling most of the segment. During the second quarter of 2014, Intel generated nearly 95% of PC processor revenue, shipping 84% of all desktop processors and 88% of all laptop processors."
The PC market continues to shrink as well. I read a report that that may change around 2019 as emerging markets come online but that is just speculation for now.
I will hold my judgement until I actually see this thing rolling.
I was hoping an eight core that performs what bulldozer should have been from the start. A real eight-core at a decent price. If Zen gets close to the performance of Skylake or cannonlake, I won't be upgrading my 3770k, especially with prices of DDR4. It feels CPU technology has been void and stuck for the last 5 years on performance and speed.
I do a lot of editing and buying an eight-core from Intel feels like a rip off, really a 1000 dollars for a chip? what is it made of pixie dust? Hope AMD gets the market stable on the price, but that also means I'll be waiting at least 2 or 3 years to have a eight-core at a good price, like the rest of us.
Intel charges what they want because they can, and thanks to mainstream desktop not needing more than 4 cores without HT, there just isn't enough demand for more cores on LGA115x for Intel to move it's slow ass and raise core counts. When they increse the core count again, I have a strong feeling that unlike 1->2 cores or 2->4cores, the performance boost of 4->6 or 4->8 will take a lot longer to materialize for most, if at all.
On the other end of the spectrum, more and more stuff wants a good GPU, and the popularity of phones and tablets demand ever more GPU to smooth animations as far as possible, so the iGPU keeps getting improved.
Go ahead and run a real app on one core of x86 and that pitiful arm chip. Guess what is going to happen? Synthetic benchmarks are more useless than ever.