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
Compatibility is only good up to a point.
After that, it's just going backwards, or at least not going forward with newer tech.
After that, it's just going backwards, or at least not going forward with newer tech.
I'll be upgrading my system this fall after Zen comes out and I've been happy with my FX-6300 system. It has been a good homebuilt for me.
Have always been an Intel man (except for that one time I had a Cyrix 4x86), but have fond memories of the days when AMD Athlons made Intel sweat and the Athlon Thunderbird was the best overall CPU at the time.
Wow. 32 posts into this thread and practically zero specific relevancy to the Mac Pro at all.
Zen doesn't look to be viable given the current baseline Mac Pro design criteria
"... This would indicate that slot three has a full x16 lane connection for data, or in effect we have 64 lanes of PCIe bandwidth in the PCIe slots. That’s about as far as we can determine here – we have seen motherboards in the past that take PCIe lanes from both CPUs, so at best we can say that in this configuration that the Naples CPU has between 32 lanes and 64 lanes for a dual processor system. ..."
http://www.anandtech.com/show/10581...-naples-32cores-dual-socket-platforms-q2-2017
1. Not getting double socket Zen any more than getting a double socket Intel offering. [ the relevant number above is the likely 32 lanes per socket. ]
2. If Zen ( desktop and server) is 32 lanes per socket. Xeon E5 v1-4 is 40 and hints that version 5 will may crank that up to 48. (e.g., multiple TB v3 (or better) controllers and dual GPU slots would stretch a single socket's bandwidth ). [ AMD is in the game but .... not at level with v4 ( Broadwell-E) in terms of I/O bandwidth. [ selection of Blender likely driven by the somewhat better job of RAM channel I/O parity and other CPU function unit improvements. ]
3. If Southbridge is in-package that may have some negative offsets in routing complexity. [ geared toward single board implementations. Not what the Mac Pro internals look like. ]
Best case is that Zen Server options means that Intel doesn't kneecap the E5 v5 so much on performance and price ( e.g., kneecapped 4 core clock/turbo or set low cost floor at 6 cores without kneecaps on I/O. ).
Zen doesn't look to be viable given the current baseline Mac Pro design criteria
"... This would indicate that slot three has a full x16 lane connection for data, or in effect we have 64 lanes of PCIe bandwidth in the PCIe slots. That’s about as far as we can determine here – we have seen motherboards in the past that take PCIe lanes from both CPUs, so at best we can say that in this configuration that the Naples CPU has between 32 lanes and 64 lanes for a dual processor system. ..."
http://www.anandtech.com/show/10581...-naples-32cores-dual-socket-platforms-q2-2017
1. Not getting double socket Zen any more than getting a double socket Intel offering. [ the relevant number above is the likely 32 lanes per socket. ]
2. If Zen ( desktop and server) is 32 lanes per socket. Xeon E5 v1-4 is 40 and hints that version 5 will may crank that up to 48. (e.g., multiple TB v3 (or better) controllers and dual GPU slots would stretch a single socket's bandwidth ). [ AMD is in the game but .... not at level with v4 ( Broadwell-E) in terms of I/O bandwidth. [ selection of Blender likely driven by the somewhat better job of RAM channel I/O parity and other CPU function unit improvements. ]
3. If Southbridge is in-package that may have some negative offsets in routing complexity. [ geared toward single board implementations. Not what the Mac Pro internals look like. ]
Best case is that Zen Server options means that Intel doesn't kneecap the E5 v5 so much on performance and price ( e.g., kneecapped 4 core clock/turbo or set low cost floor at 6 cores without kneecaps on I/O. ).
If only 32 lanes are available in Zen then forget it. 40 is hardly enough now. 64 however would be sweet.
People are forgetting most important thing. 8 core, 95 W CPU is direct competitor to... Intel i7-6700K. The CPU that is iMac CPU. And that is quad core, 4.0 GHz CPU that has 16 lanes of PCIe, and Dual channel memory. Zen has 8 cores, core clocks of A0 revision of ES have 2.8 GHz, final model can have 3.2 GHz at base, will be extremely good if Black Edition will have 3.5 GHz base clock, dual channel memory and 32 PCIe lanes.
And yes, dec is right that it has nothing in relevance to Mac Pro. The thing is: Zen cores will be exactly the same in server versions. And here comes IPC, and throughput of the design. And in this point of view, Zen desktop version is interesting.
No, Zen is a direct competitor to i7-7700K. TDP is even same... and release time can differ just few weeks.
http://ocaholic.ch/modules/smartsection/item.php?itemid=4009
Kaby lake is Intel's' candidate for next iMac, Mac Mini and Macbook.
Last edited:
https://www.computerbase.de/2016-08/amd-zen-architektur/
https://www.computerbase.de/bildstrecke/74153/14/
Now all we need to know are the core clocks. From instruction design, this CPU appears to be on Haswell-EP levels. And this is Mainstream CPU. Not bad.
Or not. It is a hybrid of Haswell and Skylake architecture designs.
https://www.computerbase.de/bildstrecke/74153/14/
Now all we need to know are the core clocks. From instruction design, this CPU appears to be on Haswell-EP levels. And this is Mainstream CPU. Not bad.
Or not. It is a hybrid of Haswell and Skylake architecture designs.
Last edited:
All good points deconstruct. Especially when it comes to PCIe bandwidth, future versions of the mac pro will be bandwidth starved, given that SSDs and thunderbolt 3 controllers will consume much more bandwidth than the current version.
The other big limitation is the time table for release. If its not out for a year hopefully Apple is not willing to wait that long to release an update to the mac pro.
Remember, its unlikely the consumer version of Zen will have 32 PCIe lanes. They won't want to spend the silicon for systems that don't need it.
Another problem with Apple adopting an 8 core Zen CPU in something like the iMac is that it will likely be a step down (possibly a significant step down) in single threaded performance compared to skylake and kaby lake. I think most consumers benefit from fewer faster cores than more slower cores. Given the low core counts of most of the products apple ships (iPhones, iPads, Macs), Apple likely agrees.
But, let's look at it from a different perspective:
Zen APU has a built in GPU on the SoC. It could be as powerful as RX 470 or even 480. It will have HBM2 on chip. There's no need for 40 PCI-e lines. With one APU they could run the whole iMac, you need just storage and some external IO.
Let's build a Mac Pro around Zen APU.
- 8 core CPU, 4-5 TFLOPs GPU, HBM2 memory shared with CPU and GPU, fully HSA compatible.
- 16 lanes for second GPU. RX 470, RX 480 or RX 490.
- 4 lanes for M.2
- 12 lanes for other I/O
And there we have something, that Apple will call a next gen super computer. But...
In the tube, there's a place for a second APU. Now, this is a SupaDupa Mac Pro:
- 8 + 8 core CPU, 8-10 TFLOPs GPU, HBM2 memory shared between both APUs.
- 16 lanes for third GPU. RX 470, RX 480 or RX 490. Still everything running under HSA, using one universal memory space.
- 8 lanes for RAID M.2
- 24 lanes for whatever data you have
Now THAT's a super computer!
But what happens with Thunderbolt? Will Apple say bye bye, and create something new that utilizes USB type-c? TB is the reason I don't believe in Zen Mac Pro. But it's a fun possibility...
Zen APU has a built in GPU on the SoC. It could be as powerful as RX 470 or even 480. It will have HBM2 on chip. There's no need for 40 PCI-e lines. With one APU they could run the whole iMac, you need just storage and some external IO.
Let's build a Mac Pro around Zen APU.
- 8 core CPU, 4-5 TFLOPs GPU, HBM2 memory shared with CPU and GPU, fully HSA compatible.
- 16 lanes for second GPU. RX 470, RX 480 or RX 490.
- 4 lanes for M.2
- 12 lanes for other I/O
And there we have something, that Apple will call a next gen super computer. But...
In the tube, there's a place for a second APU. Now, this is a SupaDupa Mac Pro:
- 8 + 8 core CPU, 8-10 TFLOPs GPU, HBM2 memory shared between both APUs.
- 16 lanes for third GPU. RX 470, RX 480 or RX 490. Still everything running under HSA, using one universal memory space.
- 8 lanes for RAID M.2
- 24 lanes for whatever data you have
Now THAT's a super computer!
But what happens with Thunderbolt? Will Apple say bye bye, and create something new that utilizes USB type-c? TB is the reason I don't believe in Zen Mac Pro. But it's a fun possibility...
Last edited:
APU seems will stay at 4 cores. This is normal if desktop CPU stays at 8, as an APU dedicates about half of its die area to powerful graphics.
What they will have is double the FP units as before, as well as SMT.
Zarni, you are looking at it in the wrong way. Greenland, which was supposed to be template for upcoming Vega Architecture design and Raven Ridge APUs.
What this means: Vega and Raven Ridge are the same architecture, that has spawned from Greenland. They will differ in CU count designs. This architecture is scalable enough that can be used in mainstream APUs and high-end server APUs.
Biggest SOC's will have much more than 16 CU's, and much more than 4 cores. Mobile, high-end APU can have low-cost HBM(1 or 2 GB's), 4 cores, and 16 CU GPU that will be faster than RX 460(bacause of the architecture). There is no problems with HBM1/HBM2 because if you have HBM2 memory controller it is backwards compatible with HBM1. It will not have vastly improved compute capabilities, apart from FP16, but it will be "slightly" different on the graphics front.
How big can be server APU? Prototypes really have 32 cores, 16 GB of HBM2 and 4096 GCN cores. And over 600 mm2 die sizes.
Now about Vega. People thought that on this roadmap slide:
AMD touted that Vega could have 3 times higher performance per watt. Well the difference between Navi and Polaris is bigger than Polaris and 28 nm GPUs.
How is that? Because Vega had to be designed for Mobile and APU architecture. Fury X already have had lower power consumption than R9 390, for Example, because of HBM. What will happen if you will couple HBM2 with 14/16 nm process?
This is just general picture that I can share right now. Im sure in upcoming weeks things will get interesting.
Slight off- topic in second part of my post. Most important things are in first part, about Zen APUs.
What this means: Vega and Raven Ridge are the same architecture, that has spawned from Greenland. They will differ in CU count designs. This architecture is scalable enough that can be used in mainstream APUs and high-end server APUs.
Biggest SOC's will have much more than 16 CU's, and much more than 4 cores. Mobile, high-end APU can have low-cost HBM(1 or 2 GB's), 4 cores, and 16 CU GPU that will be faster than RX 460(bacause of the architecture). There is no problems with HBM1/HBM2 because if you have HBM2 memory controller it is backwards compatible with HBM1. It will not have vastly improved compute capabilities, apart from FP16, but it will be "slightly" different on the graphics front.
How big can be server APU? Prototypes really have 32 cores, 16 GB of HBM2 and 4096 GCN cores. And over 600 mm2 die sizes.
Now about Vega. People thought that on this roadmap slide:
How is that? Because Vega had to be designed for Mobile and APU architecture. Fury X already have had lower power consumption than R9 390, for Example, because of HBM. What will happen if you will couple HBM2 with 14/16 nm process?
This is just general picture that I can share right now. Im sure in upcoming weeks things will get interesting.
Slight off- topic in second part of my post. Most important things are in first part, about Zen APUs.
I think something like a supped up mac mini or iMac is a good fit for a zen APU, assuming that it hits its performance targets. Having a high powered GPU onboard the CPU package is certainly a benefit to a company that likes to make small, compact computers.
However, when it comes to the mac pro, Its a fun idea to envision one that is simply two APUs. However there are some flaws in this plan.
First, the 8 core consumer versions of zen will likely not support multiple cpu configurations, just like Intel's consumer chips do not.
Second, the 32 core version of zen will probably not be an APU, and have no GPU on board. If these are mostly destined for servers, there is no need to spend the spend silicon space on a GPU.
Last, dual CPUs still shares all the same problems that a dual CPU intel system has, in that you have to make more room for things like a second bank of memory and more power circuitry, which would require a significant form factor change.
If anything AMD provided for Apple Engineering Samples of Zen APU's were for next generation Apple TV.
Yes. Also, you forget that Zen APU's are mainstream mostly, that is first thing, Server APUs are also a thing, but... nobody knows will they ever exist, and with what configurations.
You can extrapolate possibilities, however.
Yes, there is. http://www.amd.com/en-us/press-releases/Pages/amd-reveals-worlds-2016feb01.aspx