Speculation: Ryzen 3000 series

[exquisitechar]

It's a poor choice on Summit Ridge and Pinnacle Ridge because AVX2. Matisse has full support for AVX2 with 256-bit FMACs. No problems there.



R5 3600 should wipe the floor with an 8300 in every possible metric, especially if you are willing to tune it (which is what you were supposed to do with FX 8300s, FX8320es, etc.). That's no moon sidegrade, it's a space station an upgrade!

Yeah, it's a sidegrade over a 2700x, not any FX. Better ST performance, MT performance won't be that far behind, and it's vastly more efficient.

 

[TheGiant]

Now Adored's on Twitter saying that he got told today that clocks will be much lower than he said they would be, but that the CPUs will still OC to 5GHz. Makes no sense to me, but let's wait and see.

Adored should be very quiet now, in fact hiding behind somewhere

 

[CHADBOGA]

Adored should be very quiet now, in fact hiding behind somewhere

As soon as he speaks, he will give himself away.

 

[piesquared]

Perhaps the 16 core will be a Castle Peak Threadripper product. And maybe Jim's 5GHz claim will be vindicated there.

 

[DrMrLordX]

Adored should be very quiet now, in fact hiding behind somewhere

No, he just needs to stay out in front, and keep it real. If he was wrong then he was wrong. Time to take the punishment. It's one of the risks of making public predictions.

 

[Arkaign]

Adored should be very quiet now, in fact hiding behind somewhere

It's so dumb. Guy already had a terrible track record, is an egotistical psycho on forums (look at him on HoCP forums and his infamous 'kill yourself' history), and now just an abysmal leak/rumor mill on Zen2 and Vega that was wrong on so many levels, and only in the same universe as correct in the vaguest possible terms.

Here in this very thread we had much saner and more accurate predictions based on logical analysis of the technology and the marketplace.

It was just so unnecessary, because Zen2 is looking like a stellar release in most ways. It's just not the silliness of his purported nonsense.

He was getting clicks and acting arrogant over this for months. I hope this removes any credibility he has left, but hell some people believe in lizard people overlords running the world, so I honestly have no idea, lol.

Thank God for Anand and more reasonable sites and personalities.

 

[exquisitechar]

Not really. They're back at 2017 levels.

R7 1800x: $499
R9 3900x: $499

R7 1700x: $399
R7 3800x: $399

R7 1700: $329
R7 3700x: $329

I would prefer they had an 8c SKU closer in price to the R7 2700 but maybe the 3700x will take that spot after 6 months or so. We'll see.

True, I thought the 1700 was a little cheaper at launch. But wait till you see the 16 core...

 

[mattiasnyc]

Then again, if it wasn't for Jim would you kids really feel as good about yourselves?

 

[DrMrLordX]

True, I thought the 1700 was a little cheaper at launch. But wait till you see the 16 core...

$599 would be the smart choice, but if they are aiming for Intel HEDT then it could be $799 or higher. Eeef. I mean really, who's gonna stop them?

Then again, if it wasn't for Jim would you kids really feel as good about yourselves?

Kids? Oh, if only.

 

[DisEnchantment]

Some thoughts and questions

Would the 12 core actually be better at overclocking if there is any (lets say100 or 200 Mhz tops) due to the heat being generated by 6 core/chiplet instead of a single 8core/chiplet for example.
Naturally with Threadripper also confirmed to be on very much on the horizon, would a TR4 package with 4 chiplets in the corners have a a little higher clock than the 3900X?
Probabaly yes I would suppose, curious to know.

What would the Cache layout look like, would the L3 on one chiplet be indexed on the other? This would mean a true 64 MB L3 instead of 2x32.

Is the UMC on the IO die or on the chiplet, so this would mean uniform latency throughout and in case of TR NUMA is gone. Which would be great.

DDR 3200 being standard I would suppose 3600+ would seem stable.

I am really looking for ward to the tests.

If a hypothetical 32 Core ThreadRipper Zen 2 is announced I would be really interested if not super costly. Otherwise I would go for the 3900X

Looking forward for more coverage from Ian on these topics.

 

[TheGiant]

No, he just needs to stay out in front, and keep it real. If he was wrong then he was wrong. Time to take the punishment. It's one of the risks of making public predictions.

Adored wasn't wrong, no problem about that
he knew exactly what he was doing...
not that anyone took him him seriously here..or yes?

 

[amd6502]

It won't happen.

AMD has hinted that 7nm and future generations need significant work just to make it not regress on the clock speed, nevermind gaining it!

Yes but significant work is to be expected. Both from from the chip designers and the foundries that are maturing and improving the node. Just like 1st gen Zen improved in frequency for 14LPP/12nm I would expect some improvements for 7nm+, even if not as great, still in the same ballpark. I think Zen3 definitely has a very good shot at 5GHz.

 

[DrMrLordX]

Adored wasn't wrong, no problem about that
he knew exactly what he was doing...
not that anyone took him him seriously here..or yes?

I thought his clockspeed guesses were off (they were). He was definitely wrong about clocks, and he knows it, hence the backpedaling. In the end he's just a mouth on YouTube that happens to be kinda pro-AMD so the Zoner-types like him, I guess. Not a big deal.

Zen2, that's a big deal.

@DisEnchantment

DDR4-3600 is achievable on some Pinnacle Ridge systems. I am thinking that DDR4-4000 may become fairly common on Matisse rigs for those that put in the time to do the tuning. Maybe AMD will do more work to help their systems boot with XMP2.0 settings more-reliably as well, to help out those that don't like to tune RAM so much.

Also, interesting thought on the 12c OC. I don't see that they would overclock any WORSE than the 8c parts, unless power becomes a concern.

 

[coercitiv]

Then again, if it wasn't for Jim would you kids really feel as good about yourselves?

I thought we gathered here to feel good about the products, but if feeling good about yourself is the new meta then sure, make the most of Jim and pat yourself on the back kiddo!

 

[TheGiant]

so my thought about the release

that cinebench score - R20 has an IPC party already CFL with ryzen 2 (best case for the zen architecture atm)
AMD says 5GHz 9900K single core has the same score as 4,5GHz 3800X, that means 11% IPC uplift, nice!

I don't get the tdp difference 65W for 3700X and 105W for 3800X with 100 MHz turbo difference- 3700X wont hold 4,4GHz all core at 65W IMO, that why the 3800X has 105W
3900X wont hold to its 105W (as the 2700X doesn't , no matter

so performance wise 3800X will not win with 9900K IMO (lets see gaming gimme numbers!), but with half the power and lower price

in general, that is what we can call an upgrade - 15% frequency, 10% IPC, lower power ..

hello Intel?

 

[Zucker2k]

Do I wait for a 16 core part or buy the 3900X as soon as possible?
This is the question my impulsive self already knows the answer to.

Yes. Buy, buy, buy!

Correct me if I am wrong but the 2700x is fully OC'd as is? Does it have anything to OC. I ask because I am wondering if these Zen 2 chips will have a good deal of headroom for OCing. What happened to 5ghz? Is that still in the cards?

This will depend on a number of factors; the silicon, the chiplet design and how well it can handle power and dissipate heat, binning. I think with XFR and PBO, AMD has found a way to push the limits more effectively than when manually overclocking so that topic is moot to me at this point.


Congratulations to AMD for knocking the numbers out of the ballpark! I was always skeptical of 16 cores and 5GHz, anyway. At least, not on day one. However, the combination of IPC and Boost in this release is enough to give them parity with Intel in single threaded situations and beat them in multithreaded. We won't know for sure until the benchmarks. Gaming seems to be out of reach, for now, since even in this 'best foot forward' moment they could only say there is no difference between Zen 2 and Coffeelake.
That 12 core beast of a cpu is absolutely bonkers!!! Did she say 70MB L3?

Edit: I suppose it's the sum of all the caches?

 

[exquisitechar]

I doubt we'll see much in terms of overclocking past 4.6GHz. V/f curve seems rough, look at the 3700x compared to the 3800x. AMD probably halved the L1i just to help them get this minor bump in clocks over Zen+. If it could clock higher the SC turbo would've been higher too.

The 16 core might have a 4.7GHz boost, I guess? TR 3rd gen almost certainly will.

 

[Zucker2k]

My first visit to New York City back in the 90's I was heading down to the subway at 42nd street. I was just about to head on into the station at street level when this guy walked up to me and flashed a watch and said: "Wanna buy a TAG? 50 bucks!"

Now, if I was "gullible" then I would have believed that it was a genuine Tag Heuer and bought it. That's what the word "gullible" means. It means I didn't suspect something was up when I should've.

The only way to actually know if Jim is gullible is we'd have to know who his sources actually are. If they're really sources from inside AMD and if that's clear to him, then why would he be gullible?

And if it is both true that the sources are from AMD and that he was gullible in believing them it also by definition means that they weren't to be trusted. So then what we're really saying is that people working within AMD aren't to be trusted. I mean, it just further begs the question "why aren't they to be trusted?".

I used the word gullible because he had reason to not repeat FUD from his sources, especially after his prediction failed at the AMD demo. He just rolled with it and kept on dishing more of the same. That's gullibility or he just made stuff up. Your choice.

 

[happy medium]

in general, that is what we can call an upgrade - 15% frequency, 10% IPC, lower power ..

Really? This is a "tock" last Ryzen was a "tick".
I was hoping for more. I thought they would crush a 9900k. It looks to me like a $400 3800 is on par with a 10 month old $485 9900k.
Well to be fair we haven't seen game benchmarks yet.

If the 12 core part can get close to 5ghz , that seems like a great buy at @500$

 

[tamz_msc]

It's interesting that the 15% IPC uplift comes from measuring SPECint_2006 performance.

Footnote 2
https://www.amd.com/en/press-releas...ion-leadership-products-computex-2019-keynote

 

[Arkaign]

I used the word gullible because he had reason to not repeat FUD from his sources, especially after his prediction failed at the AMD demo. He just rolled with it and kept on dishing more of the same. That's gullibility or he just made stuff up. Your choice.

He was profiteering from obvious FUD, and clouding the real great work done by AMD in the process.

 

[TheGiant]

Yes. Buy, buy, buy!

This will depend on a number of factors; the silicon, the chiplet design and how well it can handle power and dissipate heat, binning. I think with XFR and PBO, AMD has found a way to push the limits more effectively than when manually overclocking so that topic is moot to me at this point.


Congratulations to AMD for knocking the numbers out of the ballpark! I was always skeptical of 16 cores and 5GHz, anyway. At least, not on day one. However, the combination of IPC and Boost in this release is enough to give them parity with Intel in single threaded situations and beat them in multithreaded. We won't know for sure until the benchmarks. Gaming seems to be out of reach, for now, since even in this 'best foot forward' moment they could only say there is no difference between Zen 2 and Coffeelake.
That 12 core beast of a cpu is absolutely bonkers!!! Did she say 70MB L3?

we don't know about that single threaded parity yet (cinebench is best case), but yeah if its not parity, it wont be that 20+% that was before

thoughts #2 for me...with that 3900X they made the 2950X (and the entire intel/amd hedt up to 16C lineup ) obsolete in a day- with that IPC and freq uplift, and power down 3900X will have better ST, same MT , PCIE 4 connectivity, so only specific use cases with lots of IO needed or that 4CH bandwith, or AVX512 - and even the mem bandwitch will be better on the new zen 3K since it can handle higher freq RAM...so they say

I can't wait to see the gaming numbers!!!any leakz ?

 

[IntelUser2000]

I think Zen3 definitely has a very good shot at 5GHz.

I wouldn't be surprised, but it won't go higher than that.

True. We need a materials innovation like graphene or whatnot

It's more than just changing materials. Intel has said in the 90s, they used a handful of materials to craft CPUs. Post-2010 they said that increased to using 1/3rd of the entire periodic table.

They are talking about how sometime in the 7nm to 5nm generation, even EUV will need double patterning, so it would have been used barely a generation. At some point, cost and difficulty raises to a point its just not worth doing it anymore. Just like how in areas where newer isn't always better, same thing will likely happen to computers.

Servethehome said internally, the HPC guys have plans to make chips that have 1KW TDP! They are doing that because the want for performance is insatiable but process and even design isn't giving them enough.

Look to increased specialization in the future. Consoles are one of them, where a stable unchanging platform allows developers to absolutely maximize what they can get out of the hardware.

 

[Arkaign]

Gaming probably won't be mind blowing, but I'm thinking it will be close enough. In basically everything else, it's top class.

 

[BigDaveX]

What's interesting is that most of the leaks had been implying a comparatively modest improvement in single-thread performance, while completely wrecking Intel in multi-thread situations. If AMD's slides are to be believed - big if, of course - it's actually more the other way around.