AMD's Ryzen CPUs (Ryzen/TR/Epyc) & Vega/Polaris/Navi GPUs

I say, have faith, good sir. AMD is in a fight for its life, regardless of how it looks. Kaby is barely 20% IPC over BW, if that. This means next year's Zen on 14nm+, if that is the node they meant the 20% IPC gain on, and they deliver like this time, means they may survive another day. It doesn't yet address our issue of BGA crap, but it is actual competition (noting that coffee will still be ahead on IPC).

What we need is a revolution in materials science for cooling, to better deal with heat dissipation, or further node shrinks allowing for desktop chips to merge, much as laptop GPUs are in the process of doing. That is coming quicker than many think (7 & 10nm). So, we must hold the wall for the coming years, hoping that light is reinforcements, not the blaze of our demise.

Otherwise, It's time to get all the equipment to solder and bake every component, including re-balling and potentially buying new templates each generation, a very untenable position. Khenglish will have to start giving online classes! (I laugh so I don't cry)

Sent from my SM-G900P using Tapatalk

 

AMD Ryzen 14nm Wafer Yields Pass 80% - Threadripper CPUs on track-Guru3.com

"Good news for and from AMD, the yields for fabbing AMD Ryzen processors seem to be reaching strong values and are reaching numbers passing 80% for fully the functional 8-core parts. Ryzen is fabbed using a 14nm FinFet process. That fabrication node has been in use and refined for over a year now."

" Dang you guys, we have a wild Summer loaded with new technology ahead of us !"

 

The problem with getting Intel or AMD to give up BGA is you have to give them a financial reason, We here do not have the consumer power to do so. This being raise all the fit you want it means absolutely nothing. As far as mobile the sheepeople want thin and light not caring that much about upgradability. So we may not like it but we are stuck with it.

If personally as a consumer you care about upgrade ability then by all means purchase the desktop hybrid machine. I just can not in good faith drive even the sheepeople to it as then it is serving my wants at their expense. You have to recommend on a consumers needs not your own wants and feelings.

You have to look at it this way too. This is the reason Clevo and those others stay in business. If every system were gaming monster trucks looking out to kill the competition where would they be?



Back on topic;

https://www.techpowerup.com/233495/amd-ryzen-2000-series-processors-based-on-refined-14-nm-process

 

That is exactly why I'm irked when called a 'fanboy' about anything.

As I always state: I'm a fan of pure and unadulterated performance/productivity for the longest term of ownership possible. I don't have the patience, expertise or time to dabble in modding and overclocking to achieve my goals for the many dozens of workstations I am responsible for; I simply use by brain cells to pick the best that is available 'now' that will hopefully still be giving me the best in the foreseeable and emotionless future.

Having 2% or even a 20% advantage on a platform is easily wiped out with a day or worse; a few days of total and usually catastrophic downtime. The current quarter mile champion isn't who gets my money. I bank on the platform that is the most consistent and reliable over the long term. This isn't fanboism - it is something that is earned day in and day out actually using the systems in question.

No amount of marketing will shake/replace that level reliability/dependability and performance that has been my benchmark for the last decade or so now...

 

Please, stop with the swearing. Are you a juvenile having a temper tantrum when it doesn't get it's way?

NVMe isn't obsolete; SSD drives based on it are in most of the chassis/platforms (notebooks) that are available - if you want real work done with your storage subsystem... Why? They throttle and they throttle bad. My experience once again. IF YOUR EXPERIENCE DOESN"T MATCH MINE: ignore what I write; but don't state fallacies that you know nothing about.

I don't ask others about your abilities; I'm able to detect them fine on my own.

As for the people you think have blocked me? Who are you? Their mother?

I'm not being duplicitous. This topic is not as one sided as you're making it out to be.

Discussing Intel (or any other relevant competitor) in this thread is pertinent to the conversation.

At least; for those that want to discuss real info and not be spoon fed marketing and theoretical platitudes that may or may not be brought to fruition better than (or equal to...) their competitors in due time.

It is telling that I can seem to have an intelligent conversation with almost anyone else here except you? I've asked you before; let's try to bridge whatever artificial gap there is between us. Sadly; you only seem capable of making that gap bigger, instead.

 

OK, please stop the infighting. Discussing Intel where competition exists is fine, discusing non AMD tech where it is irrelevant is not. Swearing beyond the TOS has to be stopped by all as well. Next person here swearing, name calling etc. is getting BANNED no more playing, this is now enough.

So now who wants to test the waters and be my first disciplinary action as a MOD?

 

In other news, ancillary to the CPU, an interesting analysis of the 14nm+ process being used in the second Vega cards next year. Keep in mind the news that already AMD has achieved 80% yields per wafer with 14nm. Also, in the party that says smaller is cheaper and leaves more margin, remember newer node shrinks cost more until research has recouped and any retooling or changing production lines has occurred.

http://wccftech.com/amd-working-14nm-vega-2-0-compete-nvidias-volta-2018/

Edit: also, although not mentioned, saw an interesting article discussing waiting on HBM2 for consumer cards yesterday (which makes sense as, for about two quarters now, or slightly less), it has been known availability of HBM2 is limited until the second half of this year.

Further, unless Nvidia makes strides on minimums (FPS) to compete with the HBCC, Vega should still perform well, comparatively. Now, the question is how good 14nm+ will be, both on the CPU and GPU side, moving forward.

Sent from my SM-G900P using Tapatalk

 

Ryzen Buying Guide: R7 1700 vs. R5 1600 vs. R5 1400


ASRock X370 Killer SLi Motherboard Review + Linux Test

 

My only issue is he talks about towards the end that for "Main stream use" the cost of 8 cores does not make sense. Well for main stream use it does. It may not today for gamers but this is not the main reason for PC's. for Gamers the new ThreadRippers will probably make no sense at all.

 

Building on that, even the 1700X is seeing price sales putting it, at times, at the same cost as the 1700, while the 7700K, with its bug, is seeing sales below $300 at times.

The argument that anything over 6 for mainstream is too much is predicated on old software designed with artificial limitations around Intel designs. This will change with either software updates or new software. Instead, this is a copout for lack of parallel processing support/optimization than true information, and is often not given with proper caveats.

Sent from my SM-G900P using Tapatalk

 

http://www.sci-tech-today.com/news/...ata+Centers/story.xhtml?story_id=11100CE9ZACU

 

About time. Dropbox has always had weird bottlenecks when they had server load at certain dates.

 

some more;
http://pcworld.com/article/3197184/...ipper-prices-specs-release-date-and-more.html

 

From that thread below:

It is telling when even an established online presence is cautioning about 'rumors' about the new chip...

My point here too...

Btw, making an O/S and software work for more and more threads isn't something that is easy, nor is it something that all workloads can benefit/profit from. Intel isn't holding software developers back on that front - the same devolopers are doing it to themselves. Let's just try to keep things in perspective here; and place any 'blame', if there is any to place, on the proper parties or reasons.

 

As far as an app benefiting from multi thread first it needs to be saturating the initial thread. Then it can be looked at if multi thread would benefit the app by sharing the load. If the app never loads the single thread it is on it is doubtful multi threading will help.

Most of us barely even touch the full potential of out 4c,8t CPU's 98% of the time. I know I definitely fall in that category. This would be the same of an 8c,16t and even more so of a 16c,32t CPU. But there in lies the rub, it is not for the 98% or better of usage, it is for the other 2% of the time where the CPU can really shine.

 

Yeah and especially once you can allocate cores for specific threads without putting the remaining cores under stress!

 

TANWare, not necessarily. A thread (or core) doesn't need to be saturated to benefit from another thread/core. But it definitely depends on the workload if there will be a benefit at all (if it can't be parallelized: more cores or spawning more threads won't help one iota).

I too spec my systems for that 2% of the time I need them. The only daily users of near 100% workloads are data centers, cloud providers and the like. In other words: not enthusiasts like almost everyone here.

When we have so many cores on tap, having the ability to allocate cores will be a nice option for 0.00000001% of users that would have a workload that needed personal attention like that. For the rest of us? Having the O/S handle it (and if possible; the processor itself) will be much more efficient and beneficial.

Just like moving SpeedStep from the O/S to the processor gives a real and immediate improvement in how responsive a system feels; allocating cores at the system level will also gain improvements that can be exceeded only marginally and in only very specific workflows/workloads after hours and hours of testing and comparing.

When we let the machines control themselves within our parameters; they always end up doing a better job for the vast majority of use cases.

To throw that option in so we can play with it would be great for a minute or two (for me...) - but after that; it will be an option that will almost never be used again.

 

I do not mean it as daily drivers at 100%. I am more referring even to just a few years ago I can remember most of the time seeing CPU loads of 33% or greater 90% of the time and 50% or greater 50% of the time etc. With today's 4c,8t CPU's this is not really seen anymore.

I am not saying it will not benefit just that it may not be needed or even noticed but this is of course a case by case basis.

 

https://www.bing.com/videos/search?...6C923802FEEEB904DD916C923802FEEEB&FORM=VRDGAR

 

https://imgur.com/t/funny/C8cTF
https://www.google.com/search?q=thr...yoTUAhXHwVQKHadlDOAQ_AUICygC&biw=1246&bih=637

 

Have yoo seen all?

 

And in only 0.38 seconds

"About 55,700 results (0.38 seconds)"

 

not sure what is up with that link but;

 

I didn't have a problem with the original link either, but after the 'Full Disclosure - this is hardly proven at all' statement what is worth watching in this video?

Rumors, what ifs and possibly's... yawn... stretch... Good Morning everyone...

 

I can hear all the reviewers now "Yeah, yeah it is the most powerful PC on the planet, but can I play games?".

 

Intel still holds fastest single core performance... so technically they are still the fastest CPU's, on the other hand, AMD CPU's are significantly better at multitasking and extreme workloads.

 

I don't know why reviewers are giving Intel the edge in these tests, most people won't OC their CPU's, so running at out of the box settings would be more fair, most fair would be to also have runs at same clock speed for all CPU's.

Running one CPU at 4.8ghz-5.0ghz, one at 4.3ghz, and Ryzen at 4.0ghz isn't exactly a fair test.

If reviewers won't undervolt CPU's on laptops for a review, why are those same reviewers so keen to run Intel at max OC against Ryzen OC - they know Ryzen is limited to about 4.0ghz right now.

Are there any / many reviewers using professional workstation software for testing instead of games? I think that games is all these reviewers have in their repertorier.

 

There is only one or two that do heavy testing aside from games. ( http://www.anandtech.com/show/11244/the-amd-ryzen-5-1600x-vs-core-i5-review-twelve-threads-vs-four/4)

Even then, they are about half games. I've complained, but...

As to settings, what is fair is stock and full overclock once ipc is understood. You have to have both though. Then, caveats about Intel's heat, AMD memory, etc. At the end, you get a good comparison. But, like Tom's focused on gaming and someone in this thread used that to show performance overall, including productivity. That is improper conclusions. But I digress.

Sent from my SM-G900P using Tapatalk

 

Right now, answering questions live on Ryzen 5 and 7 builds. Sorry I don't have the link.

Sent from my SM-G900P using Tapatalk

 

That's a shortsightedness as for price AMD is less to purchase and setup. I just update to R5 1500X and so far performance on a new Win10 install has been going good so far beyond the custom software glitches. For what I bought my i5 that price I purchased I was able to get a MSI GTX1060 6G under $900 which I consider a valued upgrade. Plus my iNtel board no longer offered BIOS update and worst cause - I hate when you remove the CPU that the pin on the socket gets mangled something. I thought iNtel would've consider such when making pin Sockets. Their way of sockets s**** if you upgrade you will one way or another damage the socket pins where you need a magnifying glass to see the pins to fix them. This is why I what something like AMD and now they have something that will not bend the pins before even start I am now happy should I want a higher CPU later I can be rest assured I won't damage the pins unlik the iNtel the socket pins.

 

latest leak;
http://hexus.net/tech/news/graphics/105295-rx-vega-gpu-specs-uncovered-via-amd-linux-patch/

 

Here's more news on Zen 2, Navi and 7nm tapeout coming at the end of this year:
https://hothardware.com/news/amd-confirms-7nm-tape-out-2h-2017-navi-zen-2

AMD seems at least 'on track' on the CPU front and upcoming GPU replacement for Vega even.

I just hope that if AMD sticks to HBM, availability of HBM won't be an issue in the future like it is right now.

 

Thank you. I'll need to update my timeline to market, but from tape out to market, you usually see 9-12 months, with 14-16 if issues on yields, or revisions, or other issues arose.

This places 7nm in H2 2018 (likely Q4) to the start of 2019, so not far off my prior estimate. It also means 14nm+ is likely being used for the silicon revision already referenced for TR and Epyc, but not confirmed. That could place a Ryzen mainstream refresh Q4 too early 2018, with Zen 2 hitting a year later in Q1 2019. But, this news makes a refresh before 7nm on TR or Epyc less likely, allowing the jump from those chips (released, at latest, likely August/September) in approximately 18 months, give or take, after their release. That seems reasonable, considering.

In other words, I'm predicting yearly updates on mainstream, 18 month cadence on workstation and servers (give or take a quarter), and haven't worked out the details on the graphics side yet, which vary depending on the Vega refresh and Navi release (need to dig more).

With a yearly mainstream release, we may see 7nm on the workstation/server side first. This could be a better way to go with Intel's accelerated release on server products after cannonlake-X next year, moving it to first to release. As such, it would allow them to get to market with 7nm before Intel releases its 10nm+ with FIVR on icelake-X.

But, this is all in flux until more direction comes out from AMD.

Sent from my SM-G900P using Tapatalk

 

Those slides actually peaked my interest about something.
Mobile Ryzen Pro was also stated that it will be released in the first half of 2018.
Could that mean that it will be done on 7nm with Ryzen 2 cores powering it?

More importantly... for whom is the Ryzen Pro aimed at?
We know that APU's can be quite powerful on both the CPU and iGP end... and if AMD isn't planning on releasing strong APU's this time around, could that mean Ryzen Pro would address that?

Current mobile Ryzen APU's tend to point into the direction of 50% more powerful CPU and 40% more powerful iGP with 50% lower TDP.
That's all fine and well, but 50% more CPU power is not enough to catch up to 7700QH... and it would also means the CPU portion of mobile Ryzen would be clocked very low.
Someone mentioned that unless AMD plans to clock their mobile Ryzen under 2GHz, the CPU portion should be able to perform much better.

Is the Asus laptop with AMD Ryzen/Vega APU just a lower end version?
I don't like the idea of AMD pandering to OEM's so they can release low performing solutions on the CPU end.
Ryzen can be comparable if not better than Intel when utilized properly... so I was at least hoping for a similarly performing CPU portion for content creation (mind you I do plan on getting a desktop Ryzen, but it would also be useful to have a strong laptop Ryzen as well).

Guess right now we can only wait and see.

 

No way on Ryzen pro. It takes 9-12 months from tape out, putting 7nm out of the first half of 2018. Not only that, risk production for 7nm is scheduled for GloFo that half, not mad production for consumer products. So, those are 14nm products.

Ryzen pro is for corporate, primarily, not consumers or gamers. So it can be ignored for many in this thread.

The APUs come in Q3-Q4. So different product. As to mobile, they are getting into 3ghz range without really hurting tdp. So, that can wait, but I'm not expecting great things this first round.

I haven't guessed on mobile because not enough info yet. So just wait and see for the moment. Remember, these are BW equivalent chips. Expect 6C variant, which helps. But coffee will be impressive

Sent from my SM-G900P using Tapatalk

 

https://www.facebook.com/video.php?v=10155392242861473

It should be the recorded Q&A from this afternoon. Turns out, I need to watch Papermaster's presentation from a couple days ago for his section discussing Infinity Fabric.

@hmscott , thought you may want to know.

 

It's the ROG version. Not the utter worst from Asus.

 

I think it fascinating. Bach not too long ago, while Intel had the undisputed performance crown, When the media would complain about low ramping of CPU speed increases they just ate up the response the future lies in further cores and threads.

Now that that exact same future is here with AMD it is dismissed summarily. Prior to Ryzen if the 8 core Intel offering were reasonable it would have been eaten up like syrup on pancakes.

 

http://www.tomshardware.com/news/intel-skylake-x-kaby-lake-x-cpu,34496.html
http://www.thinkcomputers.org/amd-confirms-radeon-rx-vega-soft-launch-at-computex/
http://www.tweaktown.com/news/57693/amd-delayed-radeon-rx-vega-until-july/index.html

Edit; My issue with Ryzen for mobile is Vega. While it sounds great that it will be offered on die what about the HBM2? Since this will not be available about the only way to have super speed memory is to make an on die video memory like a large L3 cache, if that is even possible.

 

Further links;
https://www.techpowerup.com/233644/...ounce-the-ryzen-ready-flare-x-memory-giveaway
http://www.fudzilla.com/news/processors/43728-amd-manages-high-yields-on-ryzen
https://www.back2gaming.com/reviews...bo/amd-ryzen-7-1800x-8-core-processor-review/

 

ASRock AMD X399 and Intel X299 Motherboard Confirmed via Listing-Guru3d.com

 

It is supposed to have HBM2 on PCB, according to early rumors. Now, right now, HBM2 is in short supply, which Nvidia and AMD likely took the entire supply for the cards v100 and the commercial Vega. Q3 is when availability widens. Now, examine the timeline of products with it, from Vega to the APUs to mobile. That is a lot of HBM2, with no access until H2 of this year.

Now, the question of 1 or 2 stacks is still unknown, but I'd bet on it having it. Just expect delays if hardware partners don't come through. Then again, rumors have been wrong before...

Sent from my SM-G900P using Tapatalk

 

https://segmentnext.com/2017/05/23/amd-zen-2-technology/
http://www.fudzilla.com/news/processors/43730-amd-7nm-server-plans-detailed
I really need a video or the transcript of this event to watch to see the final comment in context. But, if they are targeting Q3 with Zen 2 next year, with a rollout over the course of multiple quarters throughout the lineup, we could see Zen 2 going against the Cannonlake-X platform.

So, that needs clarified as it could affect decisions on purchase. If someone sees the event video or a FULL transcript, link it and me in a post please!!!

 

That is because you are assuming AMD cores/threads and Intel core threads are otherwise identical. They're not.

The platform and the backend design is what makes one better than the other (depending on the workload being considered).

For example; the higher latency of the AMD cache(s) are not in the same playing field as Intel's excellent/world class caching algorithms at all.

To scrutinize the minutia and ignore the way everything works together (aka; as a 'platform') is to possibly come to wrong conclusions and very likely make very expensive mistakes. Expensive in 'performance', 'productivity' or 'time' (these are all one and the same to me...).

See:
http://www.tomshardware.com/news/intel-skylake-x-kaby-lake-x-cpu,34496.html


The above link shows the current 'rumors' from both Intel and AMD in a very neutral light...

Keeping in mind that Tomshardware states right in the article:

My take away from the impartial info above? Let the fight begin... (but do note that it is a fight for actual products - nobody is going to be handed a crown to anything on simple marketing dribble...).

 

I meant would the APU used in ROG be on the lower end of Ryzen APU's?
Its rather difficult to imagine this APU scoring lower than 7700HQ in Physics while being clocked at over 2 GhZ per core.

 

OMG, fanboyism is just ooozing from your comments here. TBH, there even with the slight increase of efficiency of the cores at same frequency of 13% it is doubful at 16 over 12 cores Intel will be able to make up the grunt of 33% more cores. This will get even worse as 32c,64t Epyc CPU's in possible dual CPU systems. Now I am not saying Intel will be trumped but I will admit the performance may eat Intel for breakfast, lunch and dinner.

 

This is why I always discuss the need for higher IPC and the speed bonus of Intel versus the higher core count and great speed and IPC on Ryzen based products.

As software is redesigned to use higher multithreading, which is already there on the server side, but needs developed in the prosumer sector for workstations, you will see AMD pull away. With the announcement of tapeouts for 7nm by the end of 2017, and those few articles suggesting the earliest 7nm products by Q3 next year, potentially, which should include a boost of IPC and up to a potential 50% more cores (meaning mainstream up to 12C/24T max, if they follow the 50% increase in each product category like starship for the Epyc, and potentially 24C/48T on HEDT) per market segment, software designers will feel more pressure to find optimizations, where possible, to MT workloads. But, the added core count is possible, but pure speculation, especially if the 4 core CCX is kept, unless they redesign to be able to do a 4-core block instead of 8-core block. So, lots to find out, but I agree. If it is heavily multithreaded, AMD eats it alive. For certain other tasks, Intel still will rock.

But, as the video from pcmag (I think it was) showed, people are building Ryzen rigs that never would have before, while software developers developed inside the envelope of what Intel had on the market, which is now changing because more cores are available. They have trouble saturating the cores, but we may see changes with that in the future as software designers up their game. They also discuss the Intel gouging and artificially keeping cores from the masses as well, showing I'm not the only one saying this.

But, now you can deal with the fanboy (which I think I know who your comments were addressing).

 

No, no fanboyism... it is in the article I linked before:

Again; I'm not talking about theoretical what/ifs... I'm talking about known quantities that are reinforced by a factor of 4...

I'm also not talking about 10/12/16 cores being equal to 32 cores either (but we'll see what we'll see...)... What I said about that is that I'm very interested to see that comparison in real life workflows and not comparing minutia and wrongly extrapolating it to the whole platform.

More performance at any cost is not impossible. More performance at the same or lower energy usage that equates to actually more productivity is what is hard.

See:
http://www.tomshardware.com/reviews/amd-ryzen-5-1600x-cpu-review,5014-2.html


Of course, AMD disagrees with the testing methods for it's cache... even after the tools used to test it were updated to play nice with AMD architecture...

 

Where exactly does it say "world class caching algorithms"? That's something so ridiculously unfitting to say... I mean come on, how many competitors are there worldwide to make it a challenge?