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

Sorry, what's an MiT? Google was of little help for that acronym due to the college.


Sent from my iPhone using Tapatalk

 

More links;
http://segmentnext.com/2017/06/15/amd-ryzen-threadripper/
http://wccftech.com/amd-ryzen-threadripper-cpu-launch-august/

 

No, testing with SPi32M is the go to. You can really narrow down effects with that bench.

Edit: in other news, 40% processing power improvement with GloFo 7nm. http://www.tweaktown.com/news/58037/amds-next-gen-navi-40-faster-rx-vega/index.html

Sent from my SM-G900P using Tapatalk

 

In real news:

Notice the words 'up to'.

We're still a long way from reaching anything 7nm yet (expected second half of 2018)... when we get there, lets see where the performance actually is.

 

Yes sir!

 

I don't think we will see 7nm with AMD till at least some timer in 2019. Next scheduled update here is Zen+ which should be 14nm+ and that is 2018 some time, probably 2nd quarter. AMD has been in the crapper for so long it has to play catch up essentially. Not a bad thing but this does take time.

Links;
https://www.bing.com/videos/search?...578DAD5AAC10D48F40D7578DAD5AAC10D48&FORM=VIRE

https://hothardware.com/news/amd-ryzen-threadripper-1950x-16-core-visits-geekbench

 

Well, they promised 7nm+ by 2020 (many reporting before). My prediction is Q1 business pro processors and certain refreshes. Navi is still looking at 2019, but, the production timeline, the tapeout in Q4 this year, and Su suggesting Q3 setting 7nm next year all suggest Q4 2018 - H1 2019 for arrival. NOTHING HAS SAID THEY WILL USE 14nm+ for the FULL LINEUP. So, if they do a refresh on Ryzen, the pro chips scheduled for Q1, and maybe some mobile, on 14nm+, then do TR and Epyc on 7nm, that then gives enough time to move the other lines to 7nm without an issue. Then, it is working on staggering the 7nm+ rollout at the end of 2019 into 2020. At around 2020, PCIe 5.0 and DDR5 will be adopted, unless DDR5/4 dual compatibility is done with the move to 7nm, similar to what Intel did with DDR3/4. But that is setting the stage for 5nm.

Sent from my SM-G900P using Tapatalk

 

I'm not sure yet what to expect, but I'm a little pessimistic about a full 40%

 

Look, when companies talk about node improvements or process improvements, they give two numbers. One is the increase in processing power at a set energy consumption, the other is the energy consumption improvement (or energy efficiency amount) while maintaining the same processing power. Usually companies choose to take a mix between energy efficiency and processing gains. This is before discussing logic density improvements, which also allows for die size reduction. But, considering AMD gave 52%IPC this last round on a promised 40%, I don't know why jumping to a transistor that is 7nm from 14nm would not be able to do this, especially since skipping 10nm. So, why so dubious, unless you think they'll go with more energy efficiency and not hit that mark.

Sent from my SM-G900P using Tapatalk

 

AMD has had high yields with Ryzen.
Over 80% as it was reported.
I think that they could be well on track with their slides. But as usual, we will need to wait and see how it all plays out.
They delivered with Ryzen after all.

14nm+ might simply be a stop-over before 7nm (which has been mentioned by AMD already before).

14nm+ is likely going to hit late this year/early next year, and then 7nm and Ryzen 2 will likely follow at the end of 2018/early 2019 until Ryzen 3 is released on 7nm+ in 2020.

At least that's how I'm seeing those AMD slides unfolding.
Delays could occur, but seeing Ryzen's momentum right now, and their Threadripper... I don't think this might play a large role.
Vega seems to be delayed due to HBM2 - other than that, the GPU is ready to go (though AMD would be foolish not to take this waiting period for HBM2 and improve on Vega further).

 

It does seem 14nm+ or Zen+ is the stopover gap. But correct the entire lineup may not be subjected to the refresh. And all oif this can change date or even plan wise, this is strictly off topic and extremely forward looking. Even the 40% as you have to remember Intel has plans to get there as well. So unless there is also some development of higher efficiency cores hopefully then competition will still be ongoing.

 

I'm 'dubious' because I've been around for more than five minutes.

What did AMD base it's 52% IPC improvement on? Where did they promise (and again; on what) 40% IPC?

All I know is that AMD would have been the new Intel if it introduced a CPU that was so advanced over what Intel was already offering...

52% IPC improvements mean nothing when you're still behind the leader.

 

That's almost exactly my thinking, but I guess I can't always be casting doubts.

 

They could also be playing it safe, since Intel will also be improving over the same time likely, they don't want to make that comparison.

 

Yeah; definitely playing it safe.

Any idea where the 52% increase in IPC 'claims' came from though?

 

Every company compares IPC to their prior generation.

Sent from my SM-G900P using Tapatalk

 

I was referring to OP's comment about them being the new intel with improvements over what intel is currently offering. Likely Intel will be making similar improvements so it would not be something they'd compare directly. As you said, only to older generations.

 

I take that to mean that AMD is pretty far down the totem pole ( performance/productivity) if even with a 52% IPC increase they are still behind.

Aside:
I can't say I'm surprised though; a client recently needed a new notebook and was considering an A12-9700P with 8GB of RAM and an SSD inside... after using the system for less than a minute; I walked away and recommended that the client do so too (even with a huge ~45% discount). When the client asked how the battery life was? The manager of the store stated 'no comment'. ;o


More real news from Intel:

Intel has a new Mesh architecture for it's Skylake-X and Xeon platforms...

See:
http://www.tomshardware.com/news/intel-mesh-architecture-skylake-x-hedt,34806.html

The above link gives a good hint at where Intel is saving power in the new chips...

And also shows a chip shot of their XCC die with 18 cores... note that they did not 'whip this up' because AMD did something scary to them.

It is obvious that with each new bit of information, Intel is showing that 'reaction' is not in it's vocabulary. This isn't something that transpired overnight... rather; it is right in line with Intel's long term thinking of where to extract more performance with less power in each generation of it's platforms...

I'll concede (again) that Intel may have been forced to play it's cards a bit sooner than it may have wanted. But this isn't slapping together something for the tech rags to talk about, nor is it reacting to AMD's TR no show which is getting further away as we speak. This is Intel doing what it does best: deliver platforms that offer actual real world results for those that want the most performance and highest efficiency possible - regardless of cost.

No, Intel isn't a bargain because of the prices it charges. It is a bargain for the long term value (performance AND reliability AND compatibility) a user gets from any platform based at or above the 'magical' $2K point that has stood the test of time since ~1980 or so...

While you could get multiple $500/$600 systems over the course of the ownership period of a single good/great $2K setup - they will still pale in comparison to the productivity the single/better platform offered for the same time period will provide.

In any case; like the quote above suggests - I'm also eager to test actual hardware myself to see what (if any) performance gains are too.

If AMD and Intel don't change the date(s) they've made with us; August can't come soon enough.

 

Dude, everyone is stating Intel has had the tech as in their Xeon's. It is just now they are forced to bring it down to x299 and consumers where it previously was not planning to do so. It is so obvious they were caught off guard but quickly came up with a plan B. Even all the Intel people realize it was a reaction and was unplanned bringing the core count from 12 to 18. Even the 12 core variant is Xeon brought down to the consumer market, again the plan to bring them over just got expanded is all.

As far as productivity, unless AMD really blows this the 18 core Intel may be faster but shouldn't give a drastic increase in productivity. Now there may be cases where a rendering takes a day and maybe 30 minutes can be shaved with the Intel, in that case it may be worth the extra money. But in the end few people will realize a greater workflow for the money. We do still need to see the performance numbers but the above statements are again based on the hope AMD TR is not a huge disappointment.

Edit; I am pretty sure during the development of Ryzen with multiple CCX's the memories of their own Core 2 Quad and Bulldozer took hold. I am sure they figured even their own 12 core would squash 16 core Ryzen dual CPU with 4 CCX. So plans have changed and other than getting an 18 core to behave in a x299, all is good in the end.

 

Again; I'm highly doubting they were caught off guard as many state. But which 'Intel people' have realized it was a reaction? Just 'fans' or actual Intel employees?

How was the 'unplanned' part possible if there is a die shot, not only of the 18 core part... but also of the Mesh architecture too?

Myself; I don't care about 'rendering' speeds. I care about workflow(s) - overall.

If 30 minutes a day is saved - at little to no cost to efficiency - or any other aspect of my workflows - Sold! (A one time price means so little vs. savings of ~8 days a year in waiting for a machine to finish... 'now'...).

Otherwise? AMD will be a wanna be sports car; yeah - the Z06 with it's theoretical performance in stark contrast to it's real world performance.


Again; I really, really hope AMD doesn't blow this one either (because then we all know Intel can and probably will take the foot off the gas again...).

 

Pure speculation on my part below, bring your salt shaker.

As another note. I have seen where allot of the planned x399's are going too use 8 phase power for the CPU. I am wondering if this is enough. What really makes me suspect this is there were some whispers of needing even more than 8 phase and maybe even 10. What makes me more suspicious is there being (again if there is) a 1950x 16c,32t that I would bet is a lower powered variant of the 1998/1998x. I posted this separately as it is purely my own speculation.

Edit; as an example just this x370 Ryzen board uses a 10 phase;
https://play3r.net/reviews/motherboards/rog-x370-crosshair-vi-hero-am4-ryzen-preview/

 

Intel would never ever pushed out the new series +12 cores i9 if not AMD pushed out their 16 core!! Same tactics like this is also Nvidia's!! They will always have a backup plan.

 

I meant SOME rare cases may see that, not every one will see that. Also rendering was supposed to be just one example. Even the article you link too states for SkylakeX and Xeon.

 

MiT stands for Materials in Training, aka Materials and Training. Reps from various companies come to you, "train" you on their products/services, hold demos and more.

You noticed that too....I thought I was the only one. I said before something is up with these X399 boards compared to X299.

 

Thanks ill give that a shot when I get some time. Might have to bump the voltage up some but we will see. What would you all say would be a relatively save voltage for the ram. From what I have read anything under 1.45 volts.

 

Maybe more rumors but no 1950x, but with benchmarks too. And they support ECC for business people who need it too, does x299?

http://www.cpu-monkey.com/site.php?...annels=&memory_ecc=&pcie_version=&pcie_lanes=

So everyone, my digging has finally paid off!

Edit; Below a comparison of the 1800x and 1998x, almost perfect scaling in the synthetics for multi-thread.
http://www.cpu-monkey.com/en/compare_cpu-amd_ryzen_7_1800x-705-vs-amd_ryzen_threadripper_1998x-724

 

AMD is hardly 'behind the leader'.
Ryzen initially touted 40% increase in IPC over Carrizo (last iteration of Bulldozer/Piledriver family.
They then managed to surpass that by getting 58% or 60% ahead - and they did.

Most AMD's performance issues are due to lack of software optimizations (and the market had much longer period of time to invest and code for Intel exclusively).

The fact that Ryzen is getting the scores it does with a relatively unoptimized platform is incredible (and I already stated this twice before at least).
Imagine what might happen when software devs start optimizing for Ryzen's architecture as well.
It already demonstrated that it's capable of going toe to toe in games at 1080p (or very close to Intel) when they have proper optimizations behind them.
Look at 7700k.
That thing is clocked much higher than Ryzen is at both stock and boost... and yet, Ryzen is closing the gap with THAT cpu with software optimizations on STOCK clocks (microcode updates are yielding a bit smaller increases, but relevant regardless) in games, while vastly outpacing it in multithreading.

I think that IPC-wise, and with proper software optimizations, Ryzen is actually a BETTER performer than Intel's most current offerings.
Getting IPC increases on a hardware level is something AMD delivered... now its up to software devs to bring it out properly.

 

I think those TDP numbers might be off.
At least for the highest clocked TR.
Why?
Look at the core increase and how TDP apparently 'scales' from one to the other.
If those TDP numbers are accurate (and right now we don't have reason to think they are),I doubt that this would reflect the actual product.
180W?
170W or 175W seems more reasonable as far as TDP goes for 1998X.

 

Again these were based on a pre-released product. But finally info and we can compare iterations.

Edit; Based on those benchmarks the 1976x should give a good fight to a i9-7900 even overclocked at 4.5 GHz. This is so much better than being in the dark.

 

How Much Will AMD Vega Cost? | FE Cards Listed

Sapphire Pulse RX 570 and RX 580

 

Good question, and I'm not sure on the answer, but I'm sure it helps if your RAM sticks are remaining cool in terms of longevity at any given voltage, and you have your beastly NH-D15 pulling cold air over the top of the modules, so temperatures are probably good for you. My motherboard is using 1.36V under the XMP profile automatically (as opposed to the 1.35V I expected), so I'm sure that 1.36V is safe - without me doing any research I would guesstimate that 1.4V and under is likely to be safe given how close it is to the 1.36V I'm using. (which makes me now think, should I increase the voltage on my RAM & see if I can bump it up the Mhz and/or tighten timings some more!)

 

Impressive that R1800X single thread performance is matched and multicore perf is nearly doubled.

They really need that 14nm+ process to demolish Intel.

 

http://www.tweaktown.com/news/58052/amd-epyc-32c-64t-3-2ghz-2tb-ram-128-pcie-lanes/index.html
EPYC pricing, speeds! Bring it! Intel does have some worries!

Sent from my SM-G900P using Tapatalk

 

If it's not behind... it certainly isn't ahead...

See:
http://www.cpubenchmark.net/compare.php?cmp[]=2969&cmp[]=2874&cmp[]=2966


With the top AMD processor now giving ~26% more multicore performance over Intel's i7-7700K - and that with 200% more cores...) - it still comes with a ~37% increase in cost over the Intel offering.

It (Ryzen) is also less nimble on it's feet (i.e. 'snappiness') with a ~32% deficit in single core performance over that same, less expensive Intel offering.

This is not progress to me.

Again; I congratulate AMD for playing catchup so far... but no cigar from me. Like I've mentioned a few times already; Ryzen 8 or 9 (gen 2 or 3) may become interchangeable with Intel's then current processors, but right now? Ryzen as it sits is a long way from parity with Intel - let alone superior.

Even at 60% increase in IPC over Bulldozer/Piledriver; still is too little too late.

Why did I pick these CPU's to compare? Because they all sit at 95W TDP. Processor efficiency is a high priority for me (with multiple workstations running...). Gaming? Not so much.

Gaming? Like I've posted before (may have been deleted...); this isn't what drives the need for a high end computer today like it once may have. And where it does make a difference? In single core performance where AMD is still taking a back seat to Intel, just like it has for the past decade.

Balanced performance (single/multi) and power efficient is the name of the game in 2017 and has been for a very long time.

When/if AMD steps up to that plate and hits a home run, I'll be there to cheer them on too.

 

I would get out my salt shaker for that one. I mean even the lower skews compared are a spot on percentage less. Look at the wording too.

This means one sample could have been tested then the data extrapolated between the other variants. All Variants are 32mb L3 cache too. You have to be careful without official data but if even one did support ECC all should.

Again better than shooting in the dark but beware of rumors..............

 

I'm
Wondering if the 8 or 16 core Epyc will be overclockable. I may, if prices are right, build a home server monster if so...

Sent from my SM-G900P using Tapatalk

 

I would bet those server boards with 128 PCIe lanes cost $$$$$.

 

Supermicro has now made a gaming board, so, considering that, sometimes it is worth trying, especially for something that will last awhile.

Sent from my SM-G900P using Tapatalk

 

good find there mate! but tbh, those numbers dont surprise me at all...

think about it: TR 1998X is just two 1800X slapped onto one PCB, thus reaching almost identical clocks (3.7 vs 3.8 ghz all core boost) means almost exactly double the performance in multithreading

seeing as 1800x had 4ghz 1 core boost and 4ghz was the average oc one could achieve (with 4.1-4.2 ghz outliers golden chips), its safe to assume that 1998X could reach 3.9 ghz on average on all cores with golden chips at 4-4.1 ghz, especially since were talking identical silicon here

so....seriously? 4ghz+ on 16 cores, im totally sold! especially at 975 bucks, ZOMG! lets hope that price tag sticks....

Sent from my HUAWEI NXT-AL10 using Tapatalk

 

its 62% better multicore perf by 1800X over 7700K

 

rumor is that its $849.

 

That is supposedly non-X variant. We'll know more after Epyc drops.

Sent from my SM-G900P using Tapatalk

 

yep, 850 for 1998 and 975 for 1998x, the former having lower base and boost clocks. but everything still based on rumours and leaks, of course

Sent from my HUAWEI NXT-AL10 using Tapatalk

 

The problem is they all are perfect numbers from what looks like just calculations. There is no statistical error at all between skews, this is highly unlikely.

 

Nah... that's only true if your workload/workflows is just running Cinebench all day.

PM's 'scores' are a much more realistic measure of what I can expect from a processor in my day to day work (even if they're (also) not perfect).

 

PassMark to me does not seem to refer to my usage. I am sure though everyone is different.

 

Most people pay little attention to passmark, and I believe for good reason most of the time, if looking for people's workloads. That is why the SPEC numbers are so interesting:


Sent from my SM-G900P using Tapatalk

 

That e5-2699a V4 was relevant 3-6 months ago, it needs to go. as far now most of them now have to be changes over as Intel is changing them mostly all over to keep from getting spanked too much. Intels prices will come down and capability will increase.

 

Links;
https://www.custompcreview.com/news/amd-shows-off-ryzen-threadripper-cpu-e3-2017/42284/

Vega
http://www.anandtech.com/show/11548...retail-listings-appear-cards-at-1199-and-1799
http://wccftech.com/amd-radeon-vega-frontier-edition-air-liquid-cooled-pre-order-1800-usd/
http://www.zdnet.com/article/amd-radeon-vega-frontier-graphics-cards-make-an-appearance/

 

Heads up, that video is a hijack of a Bitwit release from Jan 2017 CES, here's the original listing:

AMD VEGA Shown Running 4K at Ultra!!

And, here's Paul's Hardware coverage of that demo:

AMD's VEGA Demo, Ryzen Builds and AM4 Motherboards