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

I think it is legit but it is just a rumor/leak for now. If true though, wow. This just roasts the competitions rump.

 

Furthering the leak;

https://www.tomshardware.com/news/amd-threadripper-3000-castle-peak-32-core-cpu,40151.html

 

Lots of more detailed OC'ing / Memory tuning suggestions for Ryzen 3 from LTT. Here's an example of the benefits to be seen in FPS improvement:


Here are the highest per CCX OC's for LTT's CPU samples, done in Ryzen Master:


3rd-gen Ryzen Overclocking - Everything you need to know!
Linus Tech Tips
Published on Aug 14, 2019
Overclocking AMD’s 3rd-gen Ryzen CPUs is pretty different from previous generations – So it’s time for an update to our Ryzen overclocking guide!

 

I am wondering with TR3 if the 16 core will not be made at all? I think this may be true as the power requirement and TDP may be lower than originally spec'd for the x399. Also if there is no Threadripper 16 core they need not compete with the am4 3950x and can start from 3960x for the 24 core etc..

Edit; I guess there will be a 16 core; https://www.userbenchmark.com/UserRun/18683463

 

well its was just like before with the highest ryzen and lowest TR being equal at 8 cores. difference for interested buyers would be the surrounding platform, pcie channels, ram channels, etc.

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

See, I want that 32-core, but don't know if I could justify it, whereas the 16-core replacement for my platform would be a fair uplift in performance! But I agree, the lowest core count TR and the highest core count mainstream.

Now, once 5nm gets here, who wants to take bets that the 45% area reduction compared to 7nm (only around 28% compared to 7nm+) winds up being used, along with the move of the I/O die to 7nm, to double core counts on Epyc again to 128 cores around 2021? Any takers on that action?

 

Timmy Joe keeps improving his Navi 5700XT AE OC, this time with a new water block + chilled (ice @ 10:00) water, and sees nice improvements in frequency and scores, plus actual 3-6+ FPS improvements in games @ 4K.

2.25Ghz with the 5700xt chilled waterblock FUN!
Timmy Joe PC Tech
Published on Aug 15, 2019
Overclocking the Radeon 5700 Xt Anniversary Edition using the Bykski full cover water block and going to the extreme with some chilled ice water gets us well over 2.2 GHZ sustained with the Soft Power Play Tables. FUN!
Igor's Lab More Power Tool:
https://www.igorslab.media/morepowe...x-5700-xt-tweaking-and-overclocking-software/

 

https://www.tomshardware.co.uk/tsmc-interposer-processor-hbm-moores-law-not-dead,news-61442.html

That is one large interposer. In fact, that can almost hold 16 core chiplets while still having the 8 HBM stacks on it. The current 7nm core chiplets are about 72-75mm^2. That really means up to 128 cores could fit with 64GB of HBM2 (if the base size isn't increased, it could be 96GB of HBM2E with 12GB/stack).

People have heard me talk about use of an active interposer for a long while, and to be fair, nothing in this article says that the interposer is an active interposer. But, if the I/O die can be integrated into the interposer, along with logic routers, thereby making it an active interposer, along with good topology for memory, I could see the HBM aligned in the center of the package with core dies on each side, without the IF controllers needed for the core dies (still may be used in conjunction with PCIe PHY, but this likely won't appear until 2021 with the N5P process detailed in the article, so new core uarch and CXL and gen Z for coherency will be incorporated, meaning I cannot predict where IF would be used).

This furthers another one of my possible predictions of CPUs with HBM on package, so long as it comes to fruition.

Edit: here is the original blog post from TSMC: https://www.tsmc.com/english/newsEvents/blog_article_20190814.htm

@ole!!! - this is what I've been talking about and waiting for! Imagine Zen 4 cores on something like this! It reduces the IF latency, greatly boosts memory bandwidth and can lower some latency there, still able to use DDR5 to feed the massive amount of HBM, etc.

If they do this in 2021, do you think Intel's 10nm or 7nm is really going to be nearly as impressive? Still this is speculative on when they will use it, but this shows it is no longer theory. They are making massive interposers for a reason (a client wants or needs it). And N5P is risk production in 2020, volume in 2021. Which of their clients would need this? Nvidia could use it, but they are moving to Samsung according to rumors, plus will likely ride the 7nm 2020 products until 2022. It could be for ARM server chips, but AMD is putting pressure there with Epyc. It could be AMD, which matches their patents and whitepapers going back to 2015 (remember, AMD had to teach a second company how to package their chips because the connections were so fine and only one company originally was capable of packaging their 7nm, which integration on an interposer by the fab creating the chips onto another wafer, with the proper underfill at 5nm may be preferred).

@D2 Ultima - remember my active interposer and HBM on package stuff? Well....

 

I just sold my Powercolor 5700xt.

The stock blower -> insane heat and noise just killed it for me. I used one for a few weeks. Performance was great. Better than my Vega 64 and able to play WoW @ 2560x1600 Ultra 10 settings 60fps everywhere. (Vega 64 I had to dial it down to 7-8 to keep a locked 60fps)

But that noise.....

First time I had to turn up audio when playing to drown it out in a LONG time. I had to let it go to someone who doesn't mind the noise and is comfortable with the heat (even though AMD says working as intended).

I have my eye on the limited edition Powercolor Red Devil 5700xt atm. It seems perfect for my needs in all aspects.

 

Sorry to hear you couldn't tune it to make it work in every day use, but everyone's use case is different. TimmyJoe also complained about the noise under heavy load and recommended people wait for the AIB cards. But, now with the water block it looks pretty sweet.

With the water blocks all of that noise is gone. It does cost more, but they always do. Even the ones that come with waterblocks as AIB water cooled usually cost more.

It will be interesting to see how the AIB vs AMD models OC / run better with waterblocks (like TimmyJoe), when the AIB GPU's get water blocks fitted for them too.

 

The fan curves weren’t good enough to compensate for the noise. The problem is at 40% fan speed, the noise was too much for me. I tried an under volt and lower fans, but the card would eventually throttle even with the UV.

The blower is just atrocious.

I looked into an after market cooler and the arctic extreme IV seemed the best idea, but that’s another 70USD after taxes and Powercolor said removing and installing an AMC voided the warranty.

I can’t remember such a horrific blower design in quite awhile.

Like you said, water blocks eliminate much of the noise (depending on setup) and the new wave of custom cards not only greatly reduce noise but cool the cards MUCH better than AMDs very poor design.

I was very pleased with the performance especially at @400USD. The red devil seems to get the cooling right @450USD. I’ll look into the other cards trickling out, but there really is no excuse for the cooler AMD slapped on the 5700xt.

 

Lisa Su gives an update, and say's she's sticking with AMD... why would she want to leave - that IBM rumor made no sense to me - AMD has a long fast ramp up to execute on, there's lots more to do.

AMD CEO Lisa Su Tells TheStreet Why Her C-Suite Journey Is Far From Over
TheStreet: Investing Strategies
Published on Aug 9, 2019
Lisa Su says she has no plans to leave AMD and that there's much that she still wants to accomplish as the CPU and GPU developer's CEO.


With Zen 3 coming out in 2020 and Zen 4 coming out in 2021, Lisa Su is gonna be very busy:

AMD Designing Zen 4 for 2021, Zen 3 Completes Design Phase, out in 2020
by btarunr Aug 8th, 2019 20:49 Discuss (64 Comments)
https://www.techpowerup.com/258099/...2021-zen-3-completes-design-phase-out-in-2020

Zen 3 is 2020, and you shouldn't Underestimate it...
Moore's Law Is Dead
Premiered Aug 13, 2019
I want to be clear - no info points to a "Zen 2+," and Zen 3 is coming in 2020. That soon, and it will be efficient!
TIMESTAMPS:
1) Zen 3 is Coming Soon, Really! 1:22
2) Zen 2 Refresh is not Zen 2+ 2:10
3) Why it actually isn’t that “soon” for Zen 3 in 2020 4:47
4) Recap of Zen 3 5:24
5) What a Zen 3 R9 4950X could look like 7:22
6) There is no reason for AMD to sit still! 9:57
7) What comes Next? 12:00


And, here's Level1Tech's onsite at the Epyc Launch with more on the floor coverage:

Here's...EPYC! - Launch Footage & Thoughts
Level1Techs
Published on Aug 13, 2019
Level1Techs went to the San Francisco EPYC launch! Here's what we saw.


EPYC Articles:
+ AnandTech - https://www.anandtech.com/show/14694/amd-rome-epyc-2nd-gen
+ ServeTheHome - https://www.servethehome.com/amd-epyc-7002-series-rome-delivers-a-knockout/
+ Phoronix - https://www.phoronix.com/scan.php?page=article&item=amd-epyc-7502-7742&num=1
+ Phoronix EPYC Benchmarks - https://www.phoronix.com/scan.php?page=article&item=amd-epyc-7502-7742&num=1

 

There are a bunch of reviews up on the new AIB 5700 / 5700XT GPU's, so far I haven't seen a native water cooled model...which one(s) are you thinking of checking out to replace the AMD blower version?

Here's a TechYes review showing the difference in stock vs OC cooling temps and noise at different load and fan speeds for the MSI 5700 XT MECH OC vs AMD Blower:


For the 7nm generation TSMC and AMD seem to think running the silicon at high temps isn't a problem. Per the TechYes chart above you could run the 5700XT @ 82c with 54% fans with noise of 51dB vs the MSI AIB 5700XT @ 66c 78% fans and get 52dB, probably neither noticeable inside of a case pointing away from you.

So the AMD GPU blower fan owner can get the same noise output as the AIB twin fan GPU if only they would listen to AMD / TSMC and allow their GPU to run at 16c higher temp than the twin fan AIB GPU's.

It's a new thing to get used to - running the temps higher than we normally strive to reduce - but based on the fab / vendor parameters, we can run the blower "just as quiet" as the AIB dual / triple fan setup's, but with higher temps.

AMD designed the blower to run at auto with 51dB and let the temps run higher but safe to use day to day.

Everyone that is trying to get the temps far lower than needed are running the blower up well above 54% on up to 100%, cooling far above the required temperature design range. That's why it's so noisy.

MSI 5700 XT MECH OC Review - Finally Custom AIB Cards Are Here
Tech YES City
Published on Aug 17, 2019
Today we take a look at MSI's MECH OC 5700 XT which brings the temps and noises down and gives a bit of extra performance vs the Reference 5700 XT AMD Model. We also compare this against the RTX 2070 Super and 2060 Super as well as the 5700.


Review: PowerColor 5700 RED DEVIL
By Hilbert Hagedoorn on: 08/15/2019 08:02 AM | 42 comment(s)
https://www.guru3d.com/articles-pages/powercolor-5700-red-devil-review,1.html

Review: MSI Radeon RX 5700 XT Evoke 8G
By Hilbert Hagedoorn on: 08/14/2019 01:01 PM | 24 comment(s)
https://www.guru3d.com/articles-pages/msi-radeon-rx-5700-xt-evoke-review,1.html

Review: ASUS Radeon RX 5700 XT ROG STRIX
By Hilbert Hagedoorn on: 08/12/2019 07:16 AM | 132 comment(s)
https://www.guru3d.com/articles-pages/asus-radeon-rx-5700-xt-rog-strix-review,1.html

Here's a list of custom cards with pointers to reviews...

AMD Radeon RX 5700 CUSTOM Series Review Roundup
Published: 15th Aug 2019, 11:28 GMT
https://videocardz.com/81803/amd-radeon-rx-5700-custom-series-review-roundup

Gamers Nexus has a Live RX 5700 XT LN2 OC session up right now:

Live: AMD RX 5700 XT Liquid Nitrogen Overclocking
Gamers Nexus
Started streaming 2 hours ago
Our first attempt at extreme overclocking the AMD RX 5700 XT with liquid nitrogen!
We're taking the AMD RX 5700 XT under LN2 to see how far we can push the clocks, using the MorePower and SoftPowerPlay table mods for additional overclocking headroom.

 

While designed to run at high temps (and like I said in a previous post "working as intended"), the fact you can slap on a ~70USD cooler and drastically reduce the temps with much lower noise is my problem.

I don't mind running things hot and heavy if they are working within spec.

I do mind having to run fans that are very loud and/or have a bad pitch or after 15-20 min of gameplay the card throttles. I tried a UV and pushing the fans to 40% and below and that was a no go.
There is no cure for that except a different cooler design or voiding your warranty to use the washer method at the minimum or installing another competent cooler.

It only throttled in raids if that helps where the GPU (and CPU for that matter) takes a hit with all the FX.

I do mind cutting corners when you could have engineered a much better cooler on numerous fronts. Yes, the cards can run within spec much hotter, but why do it at all? Why have to have
3rd parties step in to fix your mistakes or greatly enhance your cooling? I'm not saying it has to be the best cooling ever, but it could definitely be a step up from what was provided.

I'll reiterate; I was very pleased with the performance of the card. It was providing me with near 1080ti levels of performance when playing WoW @ Ultra 10 / 2560x1600 rock solid 60fps which was
my goal. I will end up most likely picking up a model with a 3rd party cooling design, but there's just no excuse on AMD's part for that piss poor blower. There's no explaining or justifying it.

 

And I still can't find the Red Devil in stock anywhere atm. I'm living on my laptop atm GPU less.

 

I don't think AMD will make 7nm I/O die on Zen 3 for one reason:
They still have leftover orders to fulfill with GlobalFoundries (which was part of AMD's agreement in leaving them) from 2019 to 2021.
That means that AMD is 'obligated' in ordering 14nm and/or 12nm products of some kind from GLOFO until 2021 to avoid penalty fines... and in all likelihood, that would include the APU's (ones produced on 12nm node - which will stop once AMD starts releasing 7nm Zen 2 APU's at the end of 2019) and finally, the I/O dies.
AMD doesn't have anything else (that I know of) which would tie them to GLOFO, so I can only see them buying chips from GLOFO that contain the I/O die until 2021 (or, Zen 3+/Zen4?).

Its possible AMD might wish to stay on 7nm EUV until then (until yields improve on 6nm/5nm TSMC process nodes) and simply use it to maximum... so when they DO switch to the lower node, they could use 7nm EUV for the I/O die (unless they replace it with something better by Zen 4).

 

Some of the low end "600" series (rebadged 500 series) are Polaris on glofo 14nm.

Maybe that, and continuing production of Zen 2 I/O dies will be enough to fulfil the agreement, maybe AMD are banking on another WSA renegotiation with lower numbers

They may just choose to pay the penalties if their financials keep improving, saddling future products with old processes that may one day be fighting Intel's 10nm would be crazy

 

You've reiterated the state of mind of someone that insists on reducing the temperatures below operational / design requirements by increasing the fan speed past their comfort zone of noise. As I said low GPU temps aren't required by AMD / TSMC for safe operation, so owners complaining about the blower noise of their AMD GPU are running their blower fan's at a higher speeds than is required for safe operation.

Even Gamers Nexus can't see it, it's how most of us have been operating for a long time. Bring the temps down at any cost. Re-paste laptop CPU's / GPU's no matter what to bring down temps even if the stock temps are fine, buy expensive and dangerous (LM) pastes and expensive thermal pads to fight for that last degree of temp - and run fans as high as possible even if the fan speed makes our computers way too noisy.

For the 7nm generation TSMC and AMD state that running the silicon at high temps isn't a problem. The TechYes chart shows that AMD designed the 5700XT to run stock auto @ 82c with 54% fans which has a noise figure of 51dB and the AIB MSI 5700XT stock auto @ 66c with 78% dual fans to get 52dB, neither GPU noise output should be noticeable inside of a case pointing away from you.

The difference at the same effective fan noise output is allowing the AMD GPU to run 16c hotter than the AIB MSI GPU. Knowing that the 82c temp is running well below the safe maximum, everyone with a blower fan can actually run their AMD GPU at the same noise output as the AIB GPU's simply by running the automatic fan curve.

It's a new thing to get used to - running the temps higher than we normally strive to reduce - but based on the TSMC / AMD design parameters, we can run the blower "just as quiet" as the AIB dual / triple fan setup's, but with higher temps that are within the safe operating range.

AMD designed the blower to run at auto with 51dB and let the temps run higher but safe to use day to day.

Spoiler: TechYes chart of AMD Blower vs Twin Fan Temps / Noise

Is the Red Devil the only model you are interested in? What about the others? Too expensive or some other reason? What is it about the Red Devil you like above the other AIB options? Would you care to elaborate?

 

So your reasoning why AMD won't innovate is because of their contract with Global Foundries for 12nm / 14nm production requirements?

I don't think after all of AMD's hard work to innovate that AMD is going to let Global Foundries hold them back from progressing forward. AMD won't let Global Foundries stop them from releasing 7nm / 6nm / 5nm products as early as possible.

Exactly, AMD will adjust their mix appropriately and pay any gaps required to allow them to produce the latest and best processes for their products. Even if Intel never gets competitive again I don't expect AMD to stop pushing forward with new processes and innovating - as Intel isn't the only competitor on the horizon.

 

That isn't what I'm saying or implying. It actually is the exact opposite. I want the fans to run quieter/lower even at the cost of higher temps. I am not a low temp zealot. I am an acceptable fan noise zealot. My desktop is plenty loud with the numerous case fans and Noctua D15 ramping up with all cores @ 5ghz previously with a Sapphire reference Vega 64 that even running full tilt wasn't as bad as the 5700xt. But the reference cooler on the 5700XT is a non starter. It stood out from everything in my case and room and announced its presence immediately

The heat/temps are a problem in only that they force the fans on the AMD blower to run at a level that produces food processor grade noise levels (hyperbole intended) that can not be adjusted down without the loss of performance / throttling due to temps.

I've had all types of reference cards in my systems over the years, and this 5700xt was the loudest by far. It is only the second piece of electronics I had to sell almost immediately because the noise level was on another level. The other being the Alienware m17 Core 2 duo w/ 3870's. It lasted *1* week in my house.

Again, one more time for clarity, the temps directly weren't the reason I had to let it go. It was the resulting, unmanageable, fan noise that couldn't be ramped down enough even with a UV to prevent throttling. If it ran ~3 below Tjunc MAX 24/7 but was relatively quiet and didn't throttle, I would have been happy and it would be in my case right now. Really like the 5700xt, but the reference cooler gets a big ole F from me.

I like the aesthetics of the red devil and triple fan design along with the results from the review, but I'm always open to other models. I won't go Sapphire again due to poor warranty support. I don't like the look of the Asrock. Gigabyte is an option.

But then again, if a model presents itself that doesn't sound like a VTOL, good performance (OC a plus) and fits in well with my current dB / harmonics envelope, I'm all ears (ba bump bump....TISH).

 

Well, if the 51dB out of the blower is too loud for you then the MSI AIB model putting out 52dB is also going to be too loud?

Did you run on auto fan when it was so noisy or were you running it on 100% or high rpm settings? Or maybe you had a faulty unit?

You see where I am going with this right? I'm quoting measured noise outputs from a review testing both the AMD and AIB models running auto fan curves that also show performance differences.

Please let us know what you find from running the same settings - auto fan - on the AIB or if you can lower the RPM / noise or need to raise it.

I realize that review is only one sample input on the noise differences, but it is what I would expect to see.

It's too bad you couldn't have kept the AMD GPU to compare side by side with the new AIB model.

 

Two schools of thought here:

A> I thought this too, which is why I was thinking of picking up a XFX reference 5700 from best buy as a comparison. If I found a test model to be much quieter, I would probably pick up an anniversary edition as I
really like the blue lighting of the reference card. If the XFX was just as nails on the chalkboard, that would settle that.

B> It isn't always the dB but a combination of dB and pitch/harmonics which can produce an unacceptable sound. For example, many are fine with Clevo P870TM/KM/DM2/DM3's, but to be
they sound like jet engines even at lower dB settings, the sound of the fans are piercing to me.

Either way, with Powercolor informing me my warranty would be voided if I opened it up that was that.

Searching online, I'm not the only one who thinks the 5700xt reference blower is loud (some used much more colorful language lol) , but it comes down to the individual too.

Some thought the Vega 64 reference blower was ridiculously loud, but I found it acceptable even under full load benching (plus your case lit up a pretty, pretty red )

The AMD reference 5700xt, in the end was a non starter, but I accept I'm one individual who met a auditory intolerance when trying to use it and was blocked from fixing the problem
without voiding my warranty. As always, YMMV.

 

Here's a review of the Sapphire 5700XT AIB GPU with temp and fan noise charts:

Sapphire RX 5700 XT Pulse Review - COOL and QUIET Navi!
bit-tech
Published on Aug 12, 2019

Spoiler: Details...

Head over to the website for more benchmarks, graphs, and analysis!

Custom Navi designs have finally arrived, and the first through our benchmarks is the Sapphire RX 5700 XT Pulse.

Cards tested in this review:

Sapphire Radeon RX 5700 XT Pulse

AMD Radeon RX 5700 XT
AMD Radeon RX 5700

Nvidia GeForce RTX 2070 Super Founders Edition
Gigabyte GeForce RTX 2060 Super Gaming OC
Nvidia GeForce RTX 2060 Super Founders Edition

Games benchmarked in this review:
Assassin's Creed Odyssey (DX11)
Far Cry New Dawn (DX11)
Metro Exodus (DX12)
Shadow of the Tomb Raider (DX12)
Tom Clancy's The Division 2 (DX12)
Total War: Three Kingdoms (DX11)

Thank you to Sapphire for supplying the review sample.

Our reviews are fully independent; no third party has ANY say over ANY of the content you see here.

Timmy Joe had problems with the Arctic Accelero Extreme III on the AMD 5700XT AE, he thinks the memory wasn't getting good enough cooling coverage, and gave up and went back to the stock blower - with the washer mod - while he waited for his water block to arrive (see above).

Comments in his video from people that also used the AAE III on their AMD 5700 said they didn't have problems, some used thermal pad / tape to connect the VRAM to the heatsink.

Perhaps a design update for previous gen coolers is needed to fit the new 5700 series GPU's?

DO NOT Aftermarket Cooler your Radeon 5700xt!
Timmy Joe PC Tech
Published on Aug 8, 2019
The Radeon 5700xt's blower isn't good but it turns out the arctic Accelero extreme iii made things worse. Do not change your blower cooler before watching this video! Thanks to Arctic for sending the Accelero out, even if it didn't work for this card


Will L 1 week ago
"Do not stop. Do no give up Timmy Joe. You, YES YOU TIMMY JOE one of the few TechTubers that still care about overclocking like it's still 1998, I love you for that. So please, keep it going!"

James Hammond 1 week ago
"Timmy, love your videos but I had the complete opposite experience with the Accelero III on my 5700xt. I purchased my card on launch day and couldn't get it to survive even stock clocks with the blower.

Here is my post on LTT forums explaining my issues then covering the Accelero installation with pictures + tips https://linustechtips.com/main/topic/1086542-accelero-xtreme-iii-and-iv-5700xt-tips/ My combo is an R5 3600x and 5700xt. You can check out my 3dmark scores all under user name CartridgeCrusadingDesh but my best Timespy run was: Total Score - 9284 Graphics - 9699 CPU - 7476"

armando1is1great 1 week ago
"i had the same problem on an rx 480 with the accelero. what helped me was thermal tape or whatever its called, thermal pad thats sticky on both sides. that way i was able to tape half the small heatsink onto the memory where it wouldnt interfere with the plate. gave beautiful temps and never any issues. thermal arhesive is definitely a pain inthe ass to work with"

paul taylor 1 week ago
"dude, you got the wrong cooler. Morphius Vega or Prolimatech mk-26... Yes, they both work. Ive tried and both hit 2100+"

Lots more helpful feedback from owners in the comments...

---> Sorry, my posts were glued together again:

Fair enough, thanks for sharing your 1st hand experience, it's very helpful.

Hopefully AMD will take the negative feedback seriously this time and move away from the blower cooler like Nvidia has already done.

 

I didn't say AMD would stop innovating.
I said that AMD would probably not be releasing the I/O die on anything other than GLOFO 14/12nm until their order contracts with GLOFO expire in 2021 (in this regard, I think AMD would be wise to at least use 12nm for I/O die as it allows greater density and translates to smaller use of space).
The other parts of new CPU's will probably still be released on newer/smaller dies if AMD decides to release them on those newer dies (which will not be dictated by GLOFO, but rather the yields of the nodes).
There has been some 'talk' that AMD may stay on 7nm EUV for a while though.

Now, we don't know if this will happen and/or if AMD will decide to march onward to smaller dies with Zen 4 (they probably will if the yields are as good as on 7nm)... but we do know right now that AMD will be going forward with 7nm EUV for Zen 3 in 2020 (but the I/O die will likely be on 14 or 12nm from GLOFO).
Zen 4 looks like it will be coming in 2021... which will probably still have a 14/12nm I/O die from GLOFO (though this may depend on WHEN in 2021 the contracts are about to expire... so Zen 4 could be using a 7nm EUV I/O).
https://www.techpowerup.com/258099/...2021-zen-3-completes-design-phase-out-in-2020

"The "Zen 4" architecture is being designed for a 2021 market debut, and will come out at a time when the 7 nm process will have matured and attained high enough volumes at TSMC for AMD to either build bigger dies (more cores per chiplet), or leverage the even more advanced 6 nm EUV node. The maturity and volumes of these sub-10 nm nodes could change the economics of the MCM approach AMD is undertaking for its EPYC processors."

AMD Ryzen 3000 "Matisse" I/O Controller Die 12nm, Not 14nm

https://www.techpowerup.com/256511/amd-ryzen-3000-matisse-i-o-controller-die-12nm-not-14nm

So, it seems the I/O was indeed made on 12nm

 

Lots of problems with these assumptions. 1) the wafer supply agreement only applies to building on any node that GF can provide.

2) 6nm is a port from 7nm and uses EUV on 5 layers, whereas 7nm EUV took a full redesign and is EUV on 4 layers. N7P is a refined N7 without EUV. 5nm is a redesign either way for the process. But, earliest 3nm gets used is 2022, with 2023 being more likely.

AMD suggested waiting for 3nm because 5nm didn't bring much benefit. But with 5nm volume in 2020 and N5P volume with an interposer being available in 2021, along with a 45% area reduction over 7nm, 27% over 7nm+ EUV, you can pack more into the chip quicker or drastically reduce the size of the chiplets. Considering Intel's 7nm is equivalent density and is a threat of being executed during 2021-22, AMD would need to adopt it and go wider for IPC gains with 5nm Zen 4. That is more likely the long lived node, if not 3nm, as 3nm/2nm is pushing the very limits of current theory, will require GAA, etc.

So 6nm makes no sense with 7nm+ EUV already requiring the redesign. If you have to do that work again, you wasn't something more tangible than from 7nm+ to 6nm, which means 5nm, likely the N5P which is a refinement of the N5 node, said to have higher frequencies and better efficiency than 5nm, but same design, meaning they wouldn't have to change anything.

N7P was just announced, which was too late to consider for Zen 3 because they had already been designing for 7nm+ EUV. Same with 6nm, which would have been easier to port from 7nm. Meanwhile, knowing N5P now, this far out, and likely having been using the 5nm dev kit out for awhile, it is an easy move to then use that refinement for volume in 2021, a year after 5nm goes into volume production.

That matches Intel's density right away for Intel 7nm, beats their 10nm refinements, and will last them for a couple years until they go to 3nm.

With AMD adopting the process node around a year after volume production starts, that means AMD would be on 5nm until 2023 or 2024. Waiting for 3nm is clearly not an option if staying with TSMC. So it may be better moving to the smaller node early.

3) why the hell do they need to increase the core count per chiplet. 8-16 cores per chiplet was the sweet spot for their 64-core study in 2014-15 using an interposer. Sure, maybe increase the CCX to include all 8 cores on the die, but adding any more cores to it increases size and risks of critical defects, thereby reducing yields. Instead, adopting an interposer to reduce latency, keeping the core count, and increasing number of core dies makes WAY more sense, unless you have something to show me otherwise. They went chiplets to get away from large core count dies. Why reverse that now? It makes no sense!

I'm sure there are other points I didn't address, but this is a good start.

 

More Boost clock investigation on x570 motherboards:

AMD 3rd Gen Ryzen Boost Clock Investigation
Hardware Unboxed
Published on Aug 21, 2019



Has anyone tried MorePowerTool for tuning their Radeon Navi's?

Igor has his own Youtube channel with details on tuning, reviews of the AIB 5700's, and it's German:
https://www.youtube.com/user/wallossek

Igor has reviews and articles in English as well:
https://www.igorslab.media/en/category/editorsdesk/reviews/
https://www.igorslab.media/en/home/

Making the Radeon RX 5700 and RX 5700 XT more energy-efficient and significantly quieter with the MorePowerTool | Tutorial
Igor Wallossek, 19. August 2019 06:00
https://www.igorslab.media/en/makin...ntly-quieter-with-the-morepowertool-tutorial/

-------------------------------------

Perhaps the 3800x is worth it's premium as it bins out according to SL a bit better in top frequency terms compared to the 3700x. Too bad the 3900x isn't quite so lucky, I think I'd hold off and see how the 3950x does in single core top frequency bins from SL to see if the %'s increase:

" Ryzen 3000 Binning Statistics
Silicon Lottery recently updated its statistics page to show numbers from their Ryzen 3000 binning process. These numbers can provide some meaningful insight for those looking to play the silicon lottery, and see what percentage of chips can hit the highest frequencies and where they fall on the volt/frequency curve. As expected, Ryzen 3000 binning is a bit underwhelming, which isn’t altogether bad, depending on your perspective.

The highest, stable frequency achieved for the Ryzen 9 3900X was 4.20 GHz at 1.250V, and only the top 6% of chips were able to hit that mark. Stepping down to the Ryzen 7 models, the top 20% of Ryzen 7 3800 X models managed a 4.30 GHz overclock at 1.300V, while the top 21% of Ryzen 7 3700X chips hit 4.15 GHz at 1.262V."

From:

HW News - Lots of Insecure BIOS & Drivers, Ryzen 3000 Binning Stats
By Eric Hamilton Published August 18, 2019 at 10:43 pm
https://www.gamersnexus.net/news-pc...nsecure-bios-drivers-ryzen-3000-binning#Ryzen

Also it seems that complex ideas filtered through marketing (or braggadocios CEO's) can mislead - whether on purpose or through inadvertent enthusiasm. It's happened with RTX, and now AMD boost clocks. I'm still hoping that the 4.75ghz boost is reached on the 3950x to substantiate that claim.

And, Windows 10 / Microsoft sucks, again with vulnerabilties...

HW News - 40 Driver Vulnerabilities Found, Ryzen 3000 Binning Stats
Gamers Nexus
Published on Aug 18, 2019
In this hardware news recap, we talk about vulnerabilities discovered in BIOS and drivers, Ryzen 3000 binning stats, and RTX updates.
Topics include the following:
- Insecure drivers found from every major vendor
- Assetto Corsa no longer confirming RTX support
- AMD updates product pages
- Ryzen 3000 binning stats posted by Silicon Lottery
- Rumors about Threadripper 3000

 

I'll have to take a look to optimize my card. I took a chance on the 5700xt AE after the Powercolor 5700xt and, yeah, it's loud. It really is a product of dB and pitch/tone.

But I love the blue lighting in the case, price was great and performance is good. I just need to replace my rear fan with a matching illuminated one before I install a AIO on the CPU (to replace the D15) to make it shine.

 

Out of curiosity, have you tried undervolting the 5700XT yet?
That could solve the problem of loudness.

 

I tried that with my last one (Powercolor) with Wattman, but it was still too loud and as I drove the fans downward to ~40%, the card started to throttle
and was hitting some pretty monster temps (100+, 114+).

Subjectively, @hmscott and I may have been onto something as the Powercolor was a bit more piercing in noise. I had mused about picking up a XFX 5700
from Best Buy to test blower vs blower, but a good deal on an AE popped up; and I decided to roll the dice.

But more importantly, the AE edition runs faster and cooler out of the box than the Powercolor and under full load, it isn't as eustachian crushing. A good UV
would probably yield quieter fans at a lower rotation than before.

I'll have to check with AMD if they allow removal of the blower to upgrade the thermal pads, washer mod and better compound. The illuminated blue
Radeon shroud is slick so I would prefer to stick with it for my case as it matches my white leds. If they're not as zero tolerance as Powercolor, optimizing
the cooling in tandem with MPT might just be the icing on the cake.

Either way, I'm rather pleased with the 5700xt. Good price, good performance and now I'll sit back and see what happens in 2020.

 

HWU's review of the 3800x has been a bit delayed, apparently the 3800x has been hard to purchase, but HW Unboxed now has one and did this comparison review:

AMD Ryzen 7 3800X vs. 3700X, What's The Difference?
Hardware Unboxed
Published on Aug 24, 2019



The AM4 3950x is supposed to be 3.5ghz base and 4.7ghz boost, so maybe it will bin out at 4.7ghz - with 16 cores / 32 threads...at a nice low voltage / power draw...

I wonder how much the Threadripper memory channel increase would make a 16c/32t Threadripper outperform an AM4 3950x?... wait for Threadripper 3, or snap up a nice AM4 3950x?

 

on the 3950x it counts on if you have a TR mainboard. An AM4 3950 will most likely perform the same as a 16 core TR or at least 95% of it.

 

Looking at the binning, if the active CCX configuration is 1+1+1+1 for the 32 core then we might see the better binning as it will best disperse heat. I just want to see the best performance.

 

AMD confirms that their Zen 3 Milan processors will not use DDR5

https://www.overclock3d.net/news/so...ir_zen_3_milan_processors_will_not_use_ddr5/1

So, this could also mean that Zen 3 will come out on existing AM4 (which will require a BIOS update).

 

great decision, backwards compatibility is way more important than DDR5
especially in the beginning, when DDR4 will still be faster (just like DDR3 was faster than DDR4 when the latter came out)

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

Right now, we only know that Zen 3 won't use DDR5 in 2020... so that could imply existing AM4 compatibility with Zen 3 via a simple BIOS update (and Zen 4 will actually be the one to include full DDR 5 support in 2021 - which is not a bad decision because as you say, there's a good chance the speeds will be lower initially, and there should only be 1 year between Zen 3 and Zen 4 giving enough time for DDR5 to improve and not be overtly expensive), still, we don't know if AMD will for certain use existing AM4 or if they will opt for a newer standard for Zen 3 (but this news DOES increase the possibility of existing AM4 being in use for at least Zen 3, whereas Zen 4/2021 would of course have everything new).

In that case, I'd love for AMD to force Acer to make a BIOS update for Helios 500 to support both Zen 2 and Zen 3 (so I can just wait until Zen 3 and upgrade to that instead)... or AMD alone could do it if Acer refuses.

Jumping from Zen+ to Zen 3 would yield quite good performance increase. The IPC differential alone would be quite big in that case (nevermind the clock increases for 65W part).

Having said that, I actually think that Zen 3 (4700x for example with 8c/16th at 65W) could/would be roughly equivalent to 3900x in multicore performance with obviously higher single threaded performance (which will depend on IPC and clock increases - 7nm EUV should have good yields by next year to allow higher clocks - but reports apparently alluded to 'moderate' increases in IPC with greater emphasis on power efficiency).

 

Yes, and I said that months ago. Direct statements from employees on Zen 3 regarding server chips said no DDR5, which also signalled using the exact same socket.

Ignore Deks on that. Numerous things say Zen 3 is the last on AM4. Then AMD is unchained from that socket, meaning a ground up socket redesign, including DDR5. It also may bring more PCIe lanes to mainstream platforms, hopefully.

That isn't true. Companies have already demoed like 4400MT DDR5, specifically cadence and micron with CL42 latencies, as of May 4, 2018.

Samsung is already producing 12Gbps LPDDR5, with a 16Gbps variant coming next year (announced July 22).

Further, at the same speeds, DDR5 will have 30% more performance and double the bandwidth. So once adopted, it will immediately be faster than DDR4, which is awesome news.

But it is 2021 before all that goodness happens! New socket, more flexibility in design of the chip and the socket. That is likely when Intel will feel pain if they slip on 7nm deliverables.

 

well, LP RAM is basically a parallel technology, cant compare performance and efficiency with regular RAM, especially since its just implemented in mobile / small form factors like phones.

yeah, 4400 at CL42...thats the thing, each new RAM gen brings with it a crazy jump in primary timings, just like before going from DDR3 to DDR4. the highest clocks of the older gen usually overlap the lowest clocks of the next gen RAM, but with far lower latencies.

however, that part about 30% performance boost i didnt know, i assume that is at identical clocks when comparing with DDR4?

still, ill reserve judgement about ddr5 being immediately faster than the highest clocking ddr4 when its out. until then, we still have a couple years to go and im sure that ddr4 wont just stand still in that time frame.

however, i am a bit worried that we wont see any more advances in ddr4 sodimms...its been a while since corsair presented their 4000 mhz kit....

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

At least there will be one more AM4 generation to justify getting that x590 + x3950, that should be good for a few generations of use + an extended lifetime of use for many others as I pass it on.

AM5 + DDR5 in 2021 is too long to wait...if the x590 + x3950 or x599 + ThreadRipper 3 test well, I think it's time for me to upgrade.

 

This is why I went with first gen, to get the second gen, then new socket. I was hopeful some things would get here sooner on features, but 2021 seems fine for another upgrade. That should be a hell of a socket!

But that is 5nm and possibly a future 3nm socket, depending how long lived for the socket this time.

 

just a few years isnt too long. totally worth the wait imo a full new socket that possibly will last another 3-4 yrs with latest hardware n stuff. can start off getting crappy ram then as ddr5 matures like ddr3/4 did then can upgrade to good ram. we might even get more pcie lanes and memory channel on consumer side never know.

 

Really, you are gonna sit there with old tech for 2 more years instead of upgrading now and having much better performance (I assume) just to get better performance 2 more years away?

With 16 cores or more in AM4 / TR4 Zen 2 you'd have much better performance now than anything else available for the same price from Intel, so why wait 2 more years for DDR5?

Well, we do have to "wait" another month or two to find out about the 3950x + x590 or TR4 + x599 performance, but at least that's not another month or two, plus 2 years.

 

AMD's marketing is busy putting out materials to bring to notice new products and features, here's some of it recently released:

The Bring Up: Episode 11: 2nd Gen AMD EPYC™ Processors
AMD
Published on Aug 29, 2019
We Bring Up: Incorporating 2nd Gen AMD EPYC™ into our video workflow, coverage of the EPYC Horizon event, and talk all things servers with Forrest Norrod!

Spoiler: Timestamps:

00:54 2nd Gen AMD EPYC™ launch
02:45 Using EPYC™ in our video workflow
05:08 Bridget and Cavin at the EPYC Horizon launch event
07:26 Cavin’s Blender demo on 2nd Gen AMD EPYC™
08:45 A kickass interview with AMD Exec Forrest Norrod
09:12 State of the server industry and looking forward
10:45 High Performance Computing and the Frontier Supercomputer
12:26 Infinity Architecture in 2nd Gen AMD EPYC™
14:03 Bridget and Cavin fact-check Forrest
15:42 2nd Gen AMD EPYC™ competitive performance
17:02 Reflecting on the AMD EPYC™ journey so far
18:22 Every single khaki joke there is

Dropbox and AMD EPYC™ – The Power of Innovation
AMD
Published on Aug 28, 2019
Akhil Gupta, VP of Engineering at Dropbox, discusses the technology partnership between Dropbox ad AMD to help drive “game-changing” compute density delivering up to 2X greater density for their leading online storage services.

Stay Ahead of the Curve with AMD Ryzen™ Pro Processors
AMD
Published on Aug 28, 2019
Throughout its 50 year history, AMD has pioneered some of the most significant advances in the PC industry. With these innovations, modern enterprises continue to trust AMD to drive their business forward.

Control the Fleet Without Constraints with AMD Ryzen™ Pro Processors
AMD
Published on Aug 28, 2019
AMD Ryzen™ Pro based laptops and desktops fit seamlessly into mixed environments and offer industry standard manageability tools across the entire product stack.

For this one, IDK if AMD screwed up or they have pulled ahead with another process improvement that allows their Pro CPU's to reach 5ghz... as this video clearly shows 5GHZ in the presentation:

Why Use AMD Ryzen™ PRO Processors Pro for Business?
AMD
Published on Aug 28, 2019
Get the rundown in two minutes. Security features, manageability and performance are at the heart of AMD Ryzen Pro Processors. Learn how modern enterprises can benefit from AMD’s industry leading processor technology. Discover more: https://www.whyamd.com



AMD Advertises Ryzen Pro As Capable Of Hitting 5GHz Clock Speeds
By Usman Pirzada, 7 hours ago
https://wccftech.com/amd-advertises-ryzen-pro-capable-of-hitting-5ghz-clock-speeds/

 

yep. due to the fierce competition, a lot of good stuff will come in 2 yrs no point upgrading now. if i go AMD right now it'll be buying entire new rig, and if i go intel now the most is 8 core which i benefit so little from it.

 

in the end the time frame doesnt matter, but your personal needs and preferences. one man's 2 yrs is another man's century or 2 weeks, depending on personal stance

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

Threadripper chipsets; https://www.thefpsreview.com/2019/08/29/amd-preps-three-new-chipsets-for-third-gen-threadripper/

Could the 40 be 4 channel and the 80's be for 8 channel memory?

 

yes i think thats the general consensus in the rumor mill atm. but nothing confirmed yet.

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

Threadripper conjecture;

 

AMD pulled the marketing video that showed "5ghz" in it, and replaced it with a new "slide" without the 5ghz label:


Why Use AMD Ryzen™ PRO Processors Pro for Business?
AMD
Published on Aug 30, 2019
750 views - Posted 45 minutes ago
Get the rundown in wwo minutes. Security features, manageability and performance are at the heart of AMD Ryzen Pro Processors. Learn how modern enterprises can benefit from AMD’s industry leading processor technology.

Glad to see AMD is listening, let's hope the word gets to everyone at AMD, let's start seeing better technical accuracy in the marketing slides.

 

Pretty lazy editing but at least they are listening

 

wheres muh 5ghz!!