A server APU can be used for GPU computing where you need fast communication. I doubt it will have more than 16 CPU cores, though (which would still be very good).
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
AMD Discussion
- Thread starter koyoot
- Start date
- Sort by reaction score
What does this have to do with Zen or APUs? its just a graphics card designed for use in servers.
[doublepost=1471892889][/doublepost]
Bandwidth between the CPU and the GPU is not usually a bottleneck. With zen topping out at 32 cores that is going to be a very big die. I doubt they have any room left to fit any sort of GPU on there, especially one that has even mainstream performance.
16 cores of die space for GPU is 4 times that as for desktop APU.
GPU computing on a card is only fit for big chunks of calculation.
It has everything. Because it is technology behind that. Architecture will be similar from compute side on Vega/Raven Ridge, Zen APU, as is here. This is exact use case for this technology.
Bandwidth is not the problem. Latency is. Integrated GPUs have crazy low latency. When they have direct access to RAM, that latency nears 0 as data does not need to be copied.
If anything, integrated GPUs actually decrease available bandwidth because if they're working directly from RAM, that RAM will be slower than the VRAM you'd typically find on a dedicated card. But for small tasks an integrated card can outbench a dedicated one simply because the latency is lower.
I'm still not convinced Zen is the right architecture for serious work, but for small stuff the performance benefits of an integrated GPU have been proven.
Yes, and that crazy low latency is what HSA is seeking for a dGPU too. That's AMD's "secret" weapon; to take the latency out of all GPUs (and other co-processors) operations. Cannot do that with Intel (except iGPU).
So, the outcome of HSA is less congested PCI-e lanes along with small overhead on CPU.
Last edited:
That Naples is quite a system... sound total overkill for MP. I think current OS X doesn't even support 64 cores/ 128 threads... maybe there are hidden changes under macOS Sierras hood?
Apple need to remake the Grand Central Dispatch in OS X anyway to make HSA happen... if Apple is ever going to adapt it.. Next gen GCD should be able to handle both parallel, serial and mixed computing in unified, coherent memory. Before that we wont see openCL 2.x. That would make macOS very interesting platform for HPC, live video broadcast and VR/AR.
Apple need to remake the Grand Central Dispatch in OS X anyway to make HSA happen... if Apple is ever going to adapt it.. Next gen GCD should be able to handle both parallel, serial and mixed computing in unified, coherent memory. Before that we wont see openCL 2.x. That would make macOS very interesting platform for HPC, live video broadcast and VR/AR.
Last edited:
http://www.anandtech.com/show/10578...rs-micro-op-cache-memory-hierarchy-revealed/2
This is best analysis of the Zen CPU. Now it is all down to core clocks on final silicon.
Both clusters need to have their own separate 16 PCIe lanes, for complete scalability.
4 cores, 8 MB L3 cache, 16 PCIe lanes.
8 cores, 16 MB L3 cache, 32 PCIe lanes.
This is best analysis of the Zen CPU. Now it is all down to core clocks on final silicon.
They will have 32 lanes. Zen is modular architecture. It has 2 clusters that are built from 4 cores, and 8 MB of L3 cache.Stacc said:
Both clusters need to have their own separate 16 PCIe lanes, for complete scalability.
4 cores, 8 MB L3 cache, 16 PCIe lanes.
8 cores, 16 MB L3 cache, 32 PCIe lanes.
Last edited:
It seems like it is so but let's wait to be sure.
An 8 core CPU with 32 lanes is great for the desktop.
Naples with 32 cores will go up to 128 lanes? Heard that rumor but seems overkill. Great if it does though.
Also 8 mem channels. Hopefully with ECC support, it would be non-sense otherwise.
However, only roughly a year from now.
It was also rumored to have support for 32 SATA/NVMe drives and a controller with support for 16 10GbE connections. The IO is also flexible.
But with all this integrated stuff, it will be a massive socket for sure.
An 8 core CPU with 32 lanes is great for the desktop.
Naples with 32 cores will go up to 128 lanes? Heard that rumor but seems overkill. Great if it does though.
Also 8 mem channels. Hopefully with ECC support, it would be non-sense otherwise.
However, only roughly a year from now.
It was also rumored to have support for 32 SATA/NVMe drives and a controller with support for 16 10GbE connections. The IO is also flexible.
But with all this integrated stuff, it will be a massive socket for sure.
Last edited:
Want to see how Haswell 4 core CPU/SoC performs?
https://browser.primatelabs.com/geekbench3/search?q=model:"MacBook Pro (15-inch Retina Mid 2015)" platform:"Mac" processor:"Intel Core i7-4980HQ" frequency:2800 bits:64
Even if Zen will be within 10% of this, with its GPUs, AMD might have a true mobile killer solution.
Biggest trouble however would be not for Intel, but... Nvidia.
https://browser.primatelabs.com/geekbench3/search?q=model:"MacBook Pro (15-inch Retina Mid 2015)" platform:"Mac" processor:"Intel Core i7-4980HQ" frequency:2800 bits:64
Even if Zen will be within 10% of this, with its GPUs, AMD might have a true mobile killer solution.Biggest trouble however would be not for Intel, but... Nvidia.
Server chips are due for 2017.
You ask why at the end of the year we will see first Zen chips? Because AMD is preparing their own gaming platform, which I already mentioned in another thread - Project Quantum. It will be based on Zen and Vega.
You ask why at the end of the year we will see first Zen chips? Because AMD is preparing their own gaming platform, which I already mentioned in another thread - Project Quantum. It will be based on Zen and Vega.
Q2'17 to be exact, if all goes well.
Quantum with Zen and Vega would look good. But depends on Vega's availability as well. Will it be October or next year as well?
AMD might be pushing Vega to October on account of this as well, if this is true.
Quantum with Zen and Vega would look good. But depends on Vega's availability as well. Will it be October or next year as well?
AMD might be pushing Vega to October on account of this as well, if this is true.
Vega is the name of architecture, not particular chips. There are 2 GPUs made with this. Everything points to situation where we will see Late Q4 launch - October for smaller Vega GPU, and Early 2017 for bigger Vega GPU.
RX 490 and RX Fury. Both with HBM2.