New Zen microarchitecture details

Page 108 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

superstition

Platinum Member
Feb 2, 2008
2,219
221
101
sirmo said:
Why I mentioned Apple is because Apple more than any other manufacturer would be willing to pay the cost of the difference for the benefits HBM2 offers. Lower power, better bandwidth, and the biggest advantage to Apple in their pursuit of ever thinner devices the space savings.
Apple, though, is getting flack for using old parts to maximize profit in its Mac line. The thinness obsession is also form over function and someone willing to go there is often willing to sacrifice performance (function).

Apple felt it needed first-rate performance in its phones/tablets to combat Android but apparently feels OS X is enough to get people to use Macs even if they're underpowered and overpriced, something that has been the case for a very long time, generally. (I like my Macbook Pro because it's a mostly very well made machine, until you need to do anything that requires the Nvidia GPU or even the Iris. Then the fan noise is horrible — thanks to Apple's thinness gimmickry. With my 2008 model I could at least stop the noise by placing the computer on a large iron skillet. That trick doesn't work with the newer one I've got, sadly.)

The other question is Intel's relationship with Apple. Going to any AMD CPU might be enough to ruin any special deals it's getting, although my understanding is that days of Apple being the one to break out the newest Intel stuff have been in the past for quite a while now.

Even if AMD were to make a better product for Apple's needs I have to wonder what sort of deal it has with Intel to prevent it from using an AMD CPU. However, Apple's desire for using cheaper parts while charging premium prices may work in AMD's favor, since Zen should be priced lower.

As for HBM not being expensive, I thought the micro bumps are a big source of expense because of a lack of economy of scale. I suppose this is changing for HBM2? But, the articles I read a while back said GDDR5X was supposed to be cheaper (GDDR5 < GDDR5X < HBM2), in the context of GPU VRAM. The performance also has the same pattern.
 

sirmo

Golden Member
Oct 10, 2011
1,014
391
136
Apple, though, is getting flack for using old parts to maximize profit in its Mac line. The thinness obsession is also form over function and someone willing to go there is often willing to sacrifice performance (function).
...
Apple/Intel business relationship is an interesting dynamic there, and rumor is Jobs had a pretty good relationship with Intel. But the leadership has changed in both companies since then. I remember reading somewhere that Apple is worth $1.2B (or there abouts) worth of business to Intel per quarter, that's more than AMD churns in revenues from all their ventures.

Apple did consider Llano back in the day according to some pretty solid news sources. But we know the story of Llano and GloFo's failings at the time. Apple also has a pretty tight relationship with AMD.. as AMD keeps bending backwards to keep them happy. For instance the rx465 (the yet unannounced full Polaris 11 part, which is going exclusively to Apple) and the full Tonga 380X before then.

Apple could also use AMD's custom-semi service. Like Microsoft or Sony.. because at the end of the day AMD provides APUs for both the Xbox and PS. The added benefit to Apple would be having its own special snowflake APU that would be harder to poke fun at in BOM comparisons.

My work computers are all Macs. Mainly because I prefer MacOS to Windows or even Linux, as it fits my workflow the best. The key to getting the best bang for your buck on your Mac purchase is to buy at the product refresh. This resource is great for instance: http://buyersguide.macrumors.com/ it lists how long each product has been on the market and when its due for a refresh. I still build a new PC every few years, for gaming or as servers for my lab. I tried a Hackintosh a few times, but the risk of an update screwing things up is not worth the downtime for me in my line of work, time is money as they say.
 

jpiniero

Lifer
Oct 1, 2010
14,843
5,457
136
The cost is only part of the issue of HBM2... the latency of it makes it not so great for general purpose system memory.

Even if AMD were to make a better product for Apple's needs I have to wonder what sort of deal it has with Intel to prevent it from using an AMD CPU. However, Apple's desire for using cheaper parts while charging premium prices may work in AMD's favor, since Zen should be priced lower.

There is none AFAIK. Apple even considered using an AMD CPU using one of the Macbook Air models; but GloFo GloFo'd it up and they went back to Intel. That's how Intel got started with Iris. Between that and GloFo screwing up their attempt at getting in on the iPhone 6S processor I'm sure Apple wants nothing to do with GloFo now.

Apple's more likely to dump Intel for their own processors.
 

sirmo

Golden Member
Oct 10, 2011
1,014
391
136
The cost is only part of the issue of HBM2... the latency of it makes it not so great for general purpose system memory.
HBM2 could improve latency as well, as with the wider bus you need less cycles to strobe the data in and out. Little known fact is that over the years general purpose memory latency hasn't improved at all or very little:



Also Hynix's HBM(1) timings..
 
Last edited:
Reactions: superstition

Drazick

Member
May 27, 2009
53
70
91

NTMBK

Lifer
Nov 14, 2011
10,269
5,134
136
I'm curious to see how the new cache design scales to high core counts. It reminds me of Haswell EP's "Cluster on Die" mode, where the cache was statically partitioned between core groups:



If L3 cache is only shared between quad-core blocks, you will see similar results. I guess if you customers are running high numbers of VMs, where each VM doesn't have more than 8 threads, it would work pretty well- no widespread cache pollution from other VMs. Intel also added "Cache Allocation Technology" in Broadwell to tackle precisely this problem:



But this may be less well-suited to a HPC workload where you want to fit a big chunk of working set into L3 for all cores to work on, where a single massive 64MB cache might have worked better than 8x8MB caches. But conversely, you get better latency, instead of dealing with an enormous 32-core ringbus (or multiple ringbuses connected by switches).
 

Tuna-Fish

Golden Member
Mar 4, 2011
1,422
1,759
136
The cost is only part of the issue of HBM2... the latency of it makes it not so great for general purpose system memory.

HBM2 does not have high latency. If anything, it can have the lowest latency of any current DRAM.

Cool, so this is AMD's riff on Haswell.

There are differences. Notably the split 1-way i-scheduler (which is bad), and the 6-uop path from uop queue (which is good).
 

piesquared

Golden Member
Oct 16, 2006
1,651
473
136
Yeah good read theres a few more Hotchip slides around too. Im really curious why Lisa and Mark refered to Naples as an SoC. This is getting v. interesting.
 

Dresdenboy

Golden Member
Jul 28, 2003
1,730
554
136
citavia.blog.de
Hehe, cool. Yes, they're watching me closely.

Well, aside from any process induced implications, Zen's microarchitecture is as exciting as the patents and patches promised. Sure, it's not POWER8 or POWER9, but for a processor, which we might get to buy soon, this is nice.

Especially the already speculated SMT thread prioritization might lead to interesting results.

P.S.: Sorry for double posting the CB link. Busy times here! ^^
 

hrga225

Member
Jan 15, 2016
81
6
11
I'm quite suprised with shedulers.I think that Dresdenboy guesses well regarding powergating and extracting maximum ILP.Could reason also be SMT? From Hotchip slides threads are competing for resources eg. allocation is dinamic,only reteriment is static(which makes sense).Does this reduce complexity in implemenatation?
 

swilli89

Golden Member
Mar 23, 2010
1,558
1,181
136
Cool, so this is AMD's riff on Haswell.

What kind of conversation are you trying to engage in with a comment like this? If we're all being adults then we all can admit Intel probably found the best optimized way to do things and AMD is simply adopting the same techniques.

This thread has been wonderful in its focus on the technical side of things, you know, where we should be discussing things. Let's not try to start that typical partisan BS please.
 
Mar 10, 2006
11,715
2,012
126
What kind of conversation are you trying to engage in with a comment like this? If we're all being adults then we all can admit Intel probably found the best optimized way to do things and AMD is simply adopting the same techniques.

This thread has been wonderful in its focus on the technical side of things, you know, where we should be discussing things. Let's not try to start that typical partisan BS please.

I think you're reading too much in my comment. This architecture reminds me of Haswell, that's all.
 

Abwx

Lifer
Apr 2, 2011
11,172
3,869
136
I think you're reading too much in my comment. This architecture reminds me of Haswell, that's all.

It has nothing to do with HW, it looks like a XV whose 4 ALUs/module have been fusioned in a single core, the FP part being unchanged topology wise.

But hey, if it performs well it s because it s inspired by Intel, isnt it..?.
 
Last edited:

NTMBK

Lifer
Nov 14, 2011
10,269
5,134
136
To be honest, the overall chip hierarchy reminds me more of Nehalem/Phenom than Haswell. 4 cores with private L1/L2 connected together through a shared L3. Nothing like the post Sandy Bridge ringbus topology.
 
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |