AdoredTV: What Is Zen 2?

Brian_B

2[H]4U
Joined
Mar 23, 2012
Messages
3,356
I'd argue both generations of Ryzen and Threadripper that have been released so far have lived up to the hype pretty well.

I agree entirely. But AMD has a long history from before. Fury and Bulldozer weren’t so long ago...
 

Brian_B

2[H]4U
Joined
Mar 23, 2012
Messages
3,356
It's simple. Don't like youtube videos about tech? Don't watch them.

P.S. If you don't like this kind of stuff, coming here to do nothing but piss and moan about it is threacrapping and you should be shown the door.

I don’t have a problem with the post. I have no problem with the videos, and I don’t watch them.

People who are trying to paint this guy as a prophet and stating we should drink the koolaid and believe - that’s what I’m taking issue with.
 

notarat

2[H]4U
Joined
Mar 28, 2010
Messages
2,331
I agree entirely. But AMD has a long history from before. Fury and Bulldozer weren’t so long ago...

I don’t have a problem with the post. I have no problem with the videos, and I don’t watch them.

People who are trying to paint this guy as a prophet and stating we should drink the koolaid and believe - that’s what I’m taking issue with.

Responding to the second quote first (there's a reason...bear with me)

I don't see anyone tryiing to paint this guy as a CPU or GPU prophet. I do, however, see people defending him from the threadcrappers whose seemingly only reason for being in the thread to begin with is to make disparaging comments about him. People stick up for the people/things they like and from what I can see the responses defending the guy occur -after- someone talks crap about him.

If you look at my posts, you'll see I make no comment pro or con about his speculations. They are what they are and his accent is what it is. I've been all over the world and have talked with people from a very wide variety of cultures. Accents can be difficult to decipher sometimes but I don't let someone's accent get in the way of the message they're conveying. I wish others were the same but, alas, that's not the case.

That said, I'll respond to the first quote now...about a "long history"

While Fury/Bulldozer may not have performed as well as hoped for, at least you're not going to be affected by a recall like you would be if you bought Intel. (Fact. Not speculation.)

I can think of 4 major recalls (and a brewing one) Intel has had over the last 25 years so on average, there's 1 recall every 5 years if you go Intel. None in the last 25 years if you go AMD. (Again, that's just going on the "long history" of both companies)

P-IV Recall
1.13GHz recall
i820 recall
P67 recall
basis peakes recall (their smart watch that tended to catch fire...on your wrist)
Brewing - Intel Atom C2XXX CPUs used in a large variety of Routers/NAS/SAN products where the CPU just suddenly decides to brick itself at random
 
Joined
Mar 3, 2017
Messages
1,021
So if the infinity fabric is decoupled from memory speeds, does that mean I can use cheap RAM and still get decent results?
 

umeng2002

Gawd
Joined
May 23, 2008
Messages
923
So if the infinity fabric is decoupled from memory speeds, does that mean I can use cheap RAM and still get decent results?

Define decent. Stock performance on Ryzen is more than decent. It's just that memory speed has a larger impact on performance because of the coupling.
 

SixFootDuo

Supreme [H]ardness
Joined
Oct 5, 2004
Messages
5,789
Sadly, there are far more people like this, than people who like to indulge in intelligent, educated best guessing based upon speculation, leaks and rumours. I just think how this forum used to be full of intelligent conversation regarding the latest GPU architecture or CPU. Talk of new features in the silicon, and advancements in technology in general...

However, I for one, are still very interested when something new happens in the PC market, as it seldom comes often enough these days.

And regarding Jim’s accent... Well, I have no trouble understanding him, but I could concede that people who do not have English as their native language could have problems, but I think being able to understand context and comprehending the information itself is the bigger issue here.

A lot of the people that can't do the accent are actually the same people that you speak of. Myself included. It's just too hard for me to follow along . I can but it's taxing. I think this comes down to the people that don't mind and the people that do. Nothing more.
 

knowom

Limp Gawd
Joined
Aug 15, 2008
Messages
425
I figure about a 75% to 25% improve to core counts and boost on clock speeds compared to the initial original 14nm Ryzen design since 7nm is a big jump. Given that they are likely to do some IPC improvements, possibly further precision boost ones, and infinity fabric as well to round it all out I think it's a reasonably safe thing to speculate. We've already seen some improvements overall I'd expect about a reasonable estimate would be upward of about a 62.5% improvement over the original Ryzen, but it could be as low as 37.5% depending on the chip and software in question. Across the new 7nm product stack I'm thinking roughly 50% average improvement, but the most balanced design would be a touch higher at more like 56.25%. You'll probably see some designs with close to a 75% clock speed increase, but more incremental core count bump and some that are inverse of that and a design that's more balance well between both for a bit of the best of both worlds. Which works best obviously depends on the work load and how single threaded it is and multi thread capable it is. Luckily with precision and a die shrink more cores can be dialed back in terms of heat and power so that a single or fewer cores can be dialed up in turn. Think checkerboard clock speed/power requirements. By dynamically adjusting clock speeds between neighboring cores of high speed speed high power high temperature and low clock speed low power low temperature they can regulate them more readily and easily and even dynamically adjust that a bit more with additional cores and threads from the 7nm die shrink for that matter. Precision boost should be more flexible than ever just from the increased core count to dynamically scale and load balance a blend of performance and heat output. The I/O will make a bit off a difference infinity fabric as well. On top of that other actual IPC changes cache changes or new instruction sets will add up to help too.

I believe AMD will be aggressive due to the design win of Ryzen itself and the timing of a major die shrink and the fact that it's been a long time since they were in the lead. I don't see them coasting along, but rather attempting to lap Intel while they've got a chance at doing so because with their budget they'll probably come back with a vengeance eventually. To be honest AdoredTV's estimates/speculations/leaks don't appear too far fetched. The hardest to believe is perhaps the highest end chip achieving 5GHz or so alongside the massive core/thread count, but then again we shouldn't discredit such possibility with precision boost and more cores to work with at a smaller node. If anything that seems more plausible perhaps. Going off my figures of roughly 75% to 25% improvements in either or both directions to core count/clock speed from a 100% die size reduction over 14nm. I don't see AdoredTV's leaks or estimates too unreasonable and basically in line with what I'd theoretically expect as well. Granted 75 to 25% could be on the high side of course. but what's to say it isn't? What if Ryzen was kind of the cusp of a lot more greatness, but heat and power simply got in the way too much at 14nm and at 7nm would just be within it thanks to precision boost and other tweaks alongside the die shrink itself? I'm simply giving out vague estimations that I've thrown out and could be a bit higher or lower depending on many factors. I know die shrink improvements alone aren't as simple as I painted it, but you do have to keep in mind AMD is surely doing a bit more than that they are ideally looking at ways to improvements lots of various aspects to take into account all the features their CPU's are capable of performing. All I'm implying is if they get close to the targets I mentioned to core speed/core count and can see a lot of what I said coming close. Things don't always scale perfectly on the software or hardware side and I realize that my estimates in no way set in stone, but rather vague idea's of what could be in a best world scenario perhaps. You could ask two different people of what they guess will happen and probably none of them would align perfectly.

Speculating is easy really, but that's all it is only AMD knows what's in store and they themselves have to adapt to challenges as they come across them. Good luck finding a perfect estimate on what will happen across a wide product stack several years in advance as a human at least. I mean AI might be able to get closer perhaps, but us mere humans probably not the case there are a lot of area's that could go better or worse than originally thought or planned in the tech world. AdoredTV's I think speculations or leaks aside are probably relatively close. I might assume a 75 to 25% improve to core count or clock speed from 14nm down to 7nm. Perhaps that's off a fair amount I'm not keen on just how much of a leap you get in relative terms from such a die shrink, but power is squared so the clock speed boost could be close and precision across more cores should help eek out more single thread performance in some instances by being able to add further granularity to temperature and power regulation to neighboring and adjacent surrounding cores that can be adjusted dynamically. I know a die shrink isn't going to be perfect due to human error and just physics and technical restraints that get in the way of that from happening. Still think AMD should reasonably get anywhere from a 50-75% bump in core count over 14nm at 7nm and anywhere from about 12.5% to 25% boost in clock speed. Like I said I could be pretty off base though if I'm close and more in line with my own targets of what I'd hope for on the higher side with AMD pushing harder because I think they will be it should be closely in line with AdoredTV's estimates. AMD will obviously present a mix of chips higher core count, but a bit lower clock speeds, balanced clock speed and core count, and lower core count with higher clock speeds. I think that's safe to expect. Plus with the chiplet binning it increases that flexibility further than it would with a big monolith die to really give a wider spread to hit a broader audience and help satisfy across a larger swath of price points to help consumers feel like they are getting what they are paying for in turn.
 
Joined
Mar 3, 2017
Messages
1,021
Define decent. Stock performance on Ryzen is more than decent. It's just that memory speed has a larger impact on performance because of the coupling.

Let's say, less than 10% difference from "stock" to what folks are able to get with the RAM that costs 50% more.
 

DuronBurgerMan

[H]ard|Gawd
Joined
Mar 13, 2017
Messages
1,340
Not going to comment any more on AdoredTV's veracity. I don't know why, exactly - and surely I have not contributed positively to the situation - but it's become an oddly touchy subject.

Anyway, as to the topic... we saw the 12 core ES pop up recently, so 12 cores on AM4 is probably a thing regardless of what the rumor mill is saying. Fingers crossed it will drop in to my X370 board just fine and save me some hassle and a bit of the "I didn't wait for Threadripper" buyer's remorse.
 

Snowdog

[H]F Junkie
Joined
Apr 22, 2006
Messages
11,267
Not going to comment any more on AdoredTV's veracity. I don't know why, exactly - and surely I have not contributed positively to the situation - but it's become an oddly touchy subject.

Anyway, as to the topic... we saw the 12 core ES pop up recently, so 12 cores on AM4 is probably a thing regardless of what the rumor mill is saying. Fingers crossed it will drop in to my X370 board just fine and save me some hassle and a bit of the "I didn't wait for Threadripper" buyer's remorse.

We don't need rumors or video speculators to tell us that either.

The AMD CEO gave an interview where she basically said, AMD would be using the extra space on the Ryzen 3000 package to add more cores (obviously via a second chiplet). She smiled and said she wouldn't say how many more, just more that 8.

In reality that is two more options: 12 and 16 and you can be fairly certain both will happen. It's only a question of when.
 

elite.mafia

[H]F Junkie
Joined
Aug 23, 2004
Messages
14,961
We don't need rumors or video speculators to tell us that either.

The AMD CEO gave an interview where she basically said, AMD would be using the extra space on the Ryzen 3000 package to add more cores (obviously via a second chiplet). She smiled and said she wouldn't say how many more, just more that 8.

In reality that is two more options: 12 and 16 and you can be fairly certain both will happen. It's only a question of when.

What interview? She didn't say that during CES I don't think?
 

ZiggyDeath

Limp Gawd
Joined
Nov 3, 2017
Messages
222
Yeah the 1700 regular I can't attest to them I OC'd a few but never screwed with them enough to really really push it with that ram profile, initially on the 1700x the max ram speed I could get was around the 2900mhz mark with all cores OC'd, then after a couple BIOS updates I did a 3 day testing period/used Reddit subforum/and luck...….
Yeah, I am running 32gb in 2 dimms after all. There's really only 1 legitimate worse setup and that'd be 16gbx4.

At launch, my ram could only do 2667, 2933 wouldn't even post. So the fact it's running 2933 is already a massive improvement. I didn't bother testing 3200 too long, not after realizing that the computer quite literally could not find the USB2 ports. Even in bios, the graphical representation of the board didn't let me click the USB2 ports. It was quite weird. And while I suppose I could have migrated all my peripherals to the USB3 ports via hubs, aversion to doing that aside, the fact that such a huge error existed meant I really didn't feel like putzing around to see if I could make 3200 work.

I've had the opposite effect. I tested a new BIOS for my CH Hero VI hoping they'd add an option to disable Bluetooth, like they do on the Heri VII. When I upgraded the BIOS I was then able to run my Patriot Viper RAM at 3200 and overclock my 1600X to 4.19GHz without any problems (Used to lock up, probably due to the memory not even being on the QVL List when I built the system)

I'll probably give OCing another shot later on. It was much more important for me to "unlock" another 5-10% performance from getting faster ram clocks than actual overclocking.
 

extide

2[H]4U
Joined
Dec 19, 2008
Messages
3,494
AdoredTV is back with another video to fuel the Ryzen hype train with his “best guesses” on what Zen 2 will become. TL;DR(W): the Scot is convinced 5GHz (and a whole lot of cores) is a pretty sure thing, doubling down on the dual chiplet theory, custom I/O chips, and overall performance increases due to changes with Infinity Fabric.

I think he is mostly spot on here except for the part about custom I/O chips -- as in different ones between the CPU and APU products. Given AMD's thriftiness and the cost of spinning up individual dies, I would bet they both use the same I/O die. Even if the substrate is different, ie, has different connections (like for video, etc) and there are slightly different pins used in the two configurations, I would be willing to bet we see a single I/O die shared by these two products. They are going to be mostly the same anyways, so it makes much more sense to me for them to be the same die. I guess time will tell.


Exactly, speculating is fun. Everyone just needs to remember that Jim and other tech writers are doing their best to paint a picture of what might be coming down the line, enjoy it and relax.

Yeah, this is one of the most 'fun' parts leading up to a new product launch.

I can think of 4 major recalls (and a brewing one) Intel has had over the last 25 years so on average, there's 1 recall every 5 years if you go Intel. None in the last 25 years if you go AMD. (Again, that's just going on the "long history" of both companies)

P-IV Recall
1.13GHz recall
i820 recall
P67 recall
basis peakes recall (their smart watch that tended to catch fire...on your wrist)
Brewing - Intel Atom C2XXX CPUs used in a large variety of Routers/NAS/SAN products where the CPU just suddenly decides to brick itself at random

Well, the P4 recall happened before the parts shipped, but you also had fun stuff like the Pentium 60/66 FDIV bug. I had one of those, heh.

Don't forget the Phenom 1 TLB bug, but yeah AMD has had less issues, but at the same time has released a lot fewer chips as well.
 
Joined
Jan 27, 2015
Messages
520
Yeah, I am running 32gb in 2 dimms after all. There's really only 1 legitimate worse setup and that'd be 16gbx4.

At launch, my ram could only do 2667, 2933 wouldn't even post. So the fact it's running 2933 is already a massive improvement. I didn't bother testing 3200 too long, not after realizing that the computer quite literally could not find the USB2 ports. Even in bios, the graphical representation of the board didn't let me click the USB2 ports. It was quite weird. And while I suppose I could have migrated all my peripherals to the USB3 ports via hubs, aversion to doing that aside, the fact that such a huge error existed meant I really didn't feel like putzing around to see if I could make 3200 work.



I'll probably give OCing another shot later on. It was much more important for me to "unlock" another 5-10% performance from getting faster ram clocks than actual overclocking.

Another thing I remember was modules that were Samsung B-die 1 sided had better results, Reddit has a huge sub forum on Ocing, I did alot of research and tuning based on what people were suggesting my voltage is at 1.46 currently.
 

ZiggyDeath

Limp Gawd
Joined
Nov 3, 2017
Messages
222
Another thing I remember was modules that were Samsung B-die 1 sided had better results, Reddit has a huge sub forum on Ocing, I did alot of research and tuning based on what people were suggesting my voltage is at 1.46 currently.
I have dual dimm double sided Samsung-B die, which is better than double sided whatever else is available. Quad dimm single sided Samsung-B die is worse.

The only worse configuration you could do is quad dimm double sided Samsung-B die, because Hynix is off the table unless the goal is to cripple the system (or run it "at spec").

Basically getting 32gb is a crapshoot and I choose the least shitty way to get to 32gb.

Which is why I said there is only 1 legitimate configuration that is worse than my setup.
 
Top