Discussion AMD Ryzen MegaThread! FAQ and Resources

[Nope 1151]

Is it just me or is that a typo?
On the videocardz slide goldstone77 posted, it says 32vms/socket squared. On the intel one, it says 24 Vms/ Socket cubed???

 

[randomizer]

Those are references to the footnotes in the bottom right of the slide. Although I think the reference is wrong for VM density for Skylake.

 

[8350rocks]

It's only going to be particularly bad for Intel if AMD absorbs large amounts of the business market, which I hope they do, because AMD needs the money and Intel needs the competition.

I hope they do manage to get back up around 25-30% server market share like the old Thunderbird Opteron days. My god...given what they accomplished with Zen, to think what they could pull off with an R&D budget bigger than a trip to starbucks.

 

[goldstone77]

http://www.techspot.com/news/69331-amd-ryzen-9-threadripper-cpu-lineup-leaks-capped.html

 

[randomizer]

Awful naming convention.

 

[jdwii]

Awful naming convention.

That's how you know its real

 

[blazorthon]

As others have said TDP is *thermal* power. In theory a chip could draw 300w *electrical power* and only dissipate 100w thermally (heat is after all just wasted energy). It's all down to the type of transistor used, the design of the chip and so on.

As an FX 8320 owner- I can tell you to keep this thing cool, at stock settings, I've had to under volt it and switch to a closed loop water cooler (just to stop it thermal throttling under load). That is supposedly a 125w chip that can be kept cool with an air cooler (how?!).

On the other hand we can see Ryzen behaves itself using a 95w Wraith air cooler, so long as the clock speeds for all cores are kept below about 3.8ghz.

I think that suggests Ryzen is very *thermally efficient design* (anecdotally people like @jdwii have noted how cool it runs), however TDP has nothing to do with *electrical efficiency* which is a different metric. I understand that TDP and EDP are fairly directly linked it's worth keeping in mind that one a proportion of the other, and the factor that differentiates the two can and will be different between designs. Given that Ryzen is built on what is essentially a mobile process- and the issues that phone and tablet manufacturers have in keeping the modern parts cool (the 'tSkin' temp AMD was going on about with some of their tablet oriented offerings a while back), well I think it makes sense that Ryzen is a more thermally efficient design in terms of power used vs heat generated.

TL,DR; a Ryzen 7 1700 can to my mind only generate around 65w of *thermal energy* and still consume 100+ w of electrical energy. I don't think AMD's measurements are necessarily wrong and everything we are seeing suggests Ryzen runs cooler than equivalent Intel parts- albeit at similar electrical power consumption.

Conservaton of energy... The energy going in must equal the energy going out. The electrical energy gets completely dissipated as heat except for any small amount that might be given off as some form of light, but even that is mostly converted to heat because the CPU is opaque.

As for running cooler, what a CPU can do is a better job of getting the heat it generates out to the CPU cooler. Ryzen has the IHS connected via solder instead of inferior paste, so that helps it to better conduct heat away.

 

[$hawn]

http://www.techspot.com/news/69331-amd-ryzen-9-threadripper-cpu-lineup-leaks-capped.html

I might be totally wrong on this one, but the 14 and 10 core SKUs have me thinking that these chips all contain two 8core CCX's connected together by the infinity fabric [and not 4cored CCX like what we've seen so far].

Or possibly, this leak is fake.

 

[8350rocks]

http://www.techspot.com/news/69331-amd-ryzen-9-threadripper-cpu-lineup-leaks-capped.html

I might be totally wrong on this one, but the 14 and 10 core SKUs have me thinking that these chips all contain two 8core CCX's connected together by the infinity fabric [and not 4cored CCX like what we've seen so far].

Or possibly, this leak is fake.

4x 4 core CCXs with IF. The socket is 4K pins.

 

[-Fran-]

The obvious first question that I'd love to see answered in a review, is how the IF latency and bandwidth scales with more CCX'es interconnected.

Cheers!

 

[warmon6]

Awful naming convention.

Maybe... Just maybe... These model numbers are based on the year the engineers that work on zen were born. :lol:

Anyways... Agree that these are horrible and really doubt that this is the real naming convention (maybe something similar but not like this)

 

[goldstone77]

Ryzen will be able to use both 7nm at TSMC and/or GoFlo!!!!!!!

 

[juanrga]

It's only going to be particularly bad for Intel if AMD absorbs large amounts of the business market, which I hope they do, because AMD needs the money and Intel needs the competition.

I hope they do manage to get back up around 25-30% server market share like the old Thunderbird Opteron days. My god...given what they accomplished with Zen, to think what they could pull off with an R&D budget bigger than a trip to starbucks.

The more optimistic prediction on the low single digit percent: ~6%.

 

[jdwii]

Headphone users be warned as it seems like this site likes to play sound

https://www.forbes.com/sites/antonyleather/2017/05/18/an-interview-with-amd-the-latest-on-ryzen-memory-support-game-performance-and-ryzen-3s-launch/#1d4d42055ff4

Ryzen interview talks about future updates more software should be optimized soon.

 

[goldstone77]

Headphone users be warned as it seems like this site likes to play sound

https://www.forbes.com/sites/antonyleather/2017/05/18/an-interview-with-amd-the-latest-on-ryzen-memory-support-game-performance-and-ryzen-3s-launch/#1d4d42055ff4

Ryzen interview talks about future updates more software should be optimized soon.

"Ryzen owners and potential Ryzen owners are interested in knowing is will the new AGESA update offer better memory support?

James: Absolutely. The previous May update had this in mind and we continue to focus on this with the update we have planned for later this month and indeed throughout the rest of this year. It’s not just May 2017 and then we’re done though. We’re going to be increasing performance in many different ways and one of those is increasing memory compatibility – we’re also working with memory vendors as well as application and game developers."

I like it! I would love synthetic benchmarks to imitate reality!

 

[aldaia]

Maybe some of you have already read those rumors, but I post here since it maybe of interest.
Rumor 1: AMD is getting great yields with Ryzen. The rumor explicitly mentions that "About 80% of Ryzen dies have all 8 cores fully functional."
Rumor 2: A very big company is buying a lot of [strike]Naples[/strike] EPYC

 

[Gon Freecss]

@juanrga;

Could you elaborate on how the memory controller optimized for throughput would cause latency?

 

[juanrga]

@juanrga;

Could you elaborate on how the memory controller optimized for throughput would cause latency?

Latency and throughput have opposite requirements for a microarchitecture. A daily analogy would be the difference between a racing car and a travel bus. The racing can transports one or two persons but does at a very fast speed. The bus transport several dozens of persons but does at a lower speed. If you need to move fast a person between two points the option is the car. If you want move a lot of people together the option is the bus.

Same happens with microarchitectures. This is also the distinction between CPUs and GPUs. CPUs are optimized to crunch small amounts of data very fast, whereas GPUs are optimized for crunching lots of data on parallel, but at lower speeds. That is why CPUs have small number of cores clocked up to ~4GHz, whereas GPUs have a larger number of cores clocked low. CPUs and GPUs have different memory controllers. Do you need to transport little information very fast or a lot of information together? Do you need a memory controller with a small latency to access data or one does can transport a lot of adjacent info together with higher bandwidth?

The whole of RyZen CPU is more optimized for throughput rather than latency. It is not only the memory controller, but the dual CCX approach with a high-latency interconnect, the wide (10 pipe) muarch with higher SMT gains but lower ILP, etc. That is the reason why RyZen wins on throughput-oriented workloads as Blender, Handbrake, which have GPU ports.

 

[gamerk316]

@juanrga;

Could you elaborate on how the memory controller optimized for throughput would cause latency?

General rule: Throughput and Latency are conflicting requirements. Any attempt to increase one typically leads to a decrease in the other.

 

[Gon Freecss]

Thank you very much guys!

 

[jdwii]

Maybe some of you have already read those rumors, but I post here since it maybe of interest.
Rumor 1: AMD is getting great yields with Ryzen. The rumor explicitly mentions that "About 80% of Ryzen dies have all 8 cores fully functional."
Rumor 2: A very big company is buying a lot of [strike]Naples[/strike] EPYC

Was just reading that expect prices to drop on ryzen this is great and yes unusual news since we are talking about global foundries. This would make me think ryzen 3 will be in limited numbers however
..

 

[8350rocks]

https://videocardz.com/69755/asrock-confirms-x299-and-x399-motherboards

x399 platform confirmation.

 

[-Fran-]

https://videocardz.com/69755/asrock-confirms-x299-and-x399-motherboards

x399 platform confirmation.

I'm losing track with the names now.

Is that a "next gen chipset now" or just a really stupid naming convention from AMD? Also, what is the difference between X299 and X399 then? The only leaked info I could find about 399 was it could be a consumer dual socket. I don't mind that, but... Really? I don't think a dual socket for the consumer market matters that much... I'd say at least.

Anyway, weird stuff from AMD for the higher "prosumer" market.

Cheers!

 

[goldstone77]

https://videocardz.com/69755/asrock-confirms-x299-and-x399-motherboards

x399 platform confirmation.

HAHAHA, I like it!

 

[Nope 1151]

@yuka, I think, that in AMD's quest to "be as like to intel as possible", they've lost what makes them unique name wise.