Speculation: Ryzen 3000 series

Page 100 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

TheGiant

Senior member
Jun 12, 2017
748
353
106
well the road for intel from 4 to 5GHz while increasing features and IPC was quite long

I too don't expect ryzen just reach 5GHz because of some magic 7nm trick, in fact if they can reach 5GHz, it wont be on the standard process and with low power

I couldn't care less about frequency...if AMD makes a 8C chip at 4,2 GHz all 8C/16T with torture test power of 90W, IPC of CFL and oc potential to like 4,7GHz it will be enough

the chip exists now in 9900K, also with the power at 95W fixed it can perform very good but the price is killing it
 
Last edited:

NostaSeronx

Diamond Member
Sep 18, 2011
3,800
1,283
136
Cat-derived Vt selection is more orientated towards power/density for GOP(ALU/AGU)/FLOP(FPU)/MOP(LD/ST)/FOP(FE) per cycle increases. My expectation is AMD is going to release at the OC Max, no overclockability whatsoever.
 
Reactions: Tup3x

.vodka

Golden Member
Dec 5, 2014
1,203
1,537
136
Zen/+ itself can do >5GHz in both 14nm and 12nm iterations, it just needs cold to do it. Clockspeed here is not an architectural limitation, it has to do with the fact that these processes were meant for low power SoCs with the even more important drawback of these using a 20nm BEOL, this puts such speeds well out of these nodes' intended target. Zen's physical implementation on these probably isn't the best it can be either, too. It all adds to a ~4GHz wall for Zen and a ~4.3GHz wall for Zen+ at borderline dangerous voltages for longterm usage at ambient temperatures.

TSMC's 7nm should let Zen2 scale higher than ~4.3GHz as seen in 12nm+Zen+ at ambient temperatures, that's a given, especially if AMD has gone with the HP variant.

Expecting 4.5GHz tops from this transition to a one and a half node jump (leaving the 20nm BEOL behind) despite the ever decreasing clockspeed scaling is just being too damn negative IMO. ~5GHz is, after all, the current reasonable clockspeed limit under reasonable conditions until further advances are made in materials and finding ways to cheat physics out of unintended behaviors as we go forward.. this limit applies to both Intel and AMD, they're both fighting the same physics.

Sure, Intel got to 5GHz on *lake at ambient temperatures and safe voltages first (no, faildozer+32nm doesn't count) by fine tuning their 14nm process and physical implementation of Skylake to perfection by necessity. AMD can do the same (or get reasonably close to that number) with a more competent process available to them and more money to throw at the problem for the refinement and fine tuning required. They're not operating on a shoestring budget anymore as they did when designing Zen, there are more possibilities open for them to take.

5GHz would be ideal, yet I think it'll happen as 4.35GHz does on the 2700x via XFR2/PBO: high voltages to sustain that clockspeed for peak temporary ST performance in a way that's safe for long term usage, but not viable for safe manual OC/all core usage at such voltages as can be done with the 9900k, for example.
 
Last edited:

StinkyPinky

Diamond Member
Jul 6, 2002
6,936
1,236
126
I fully expect AMD's top CPU's to reach 5GHz. I don't see why people are so suspicious about this? Ryzen 2700X can boost to 4.3Ghz on 14nm GloFo process, 2950X at 12nm can boost to 4.4GHz, so I don't see why new and improved CPU's on 7nm can't reach 5GHz, its seems very likely to me. If anything we might even see a CPU that comes later on, something like a Ryzen 9 3900x that can boost up to 5.2GHz.

On first gen 7nm parts? No way. Intel only managed it after three attempts on an easier node to achieve it.
 
Reactions: ozzy702

coercitiv

Diamond Member
Jan 24, 2014
7,092
16,360
136
On first gen 7nm parts? No way. Intel only managed it after three attempts on an easier node to achieve it.
Intel had clockspeed regression going from 22nm to 14nm: 4790K had 4.4Ghz turbo, 6700K went down to 4.2Ghz turbo and overclocking wasn't stellar either. It was Kaby Lake and then Coffee Lake that showed significant & iterative improvements towards 5GHz.

Personally I expect something along the lines of 4.5Ghz turbo and 4.7Ghz XFR/PBO on the first 7nm version. We'll probably start seeing 5.1-5.2Ghz "thermal velocity" sprints from Intel too
 

FlanK3r

Senior member
Sep 15, 2009
321
84
101
Me too, I think some model will have XFR up to 4.5 GHz (quadcore+sixcore+12C) and some models up to 4.7 GHz (8C/16T)
 

TheGiant

Senior member
Jun 12, 2017
748
353
106
7% increase in speed on a full node jump?
you forgot IPC improvement
as said before, Intel too decreased turbo with skylake vs haswell with node improvement
and skylake i7 6700K was hotter and not using full potential thanks to 2400MHz DDR4 with high timings as standard
 

Guru

Senior member
May 5, 2017
830
361
106
that is nice, but nobody talking about here on this forum

everything seems too optimistic for me- 7nm is the miracle

toms posted that TR from ryzen 2 dropped from the roadmap...wondering what it means https://www.tomshardware.com/news/amd-third-gen-threadripper-roadmap,39254.html
Its eating in their server CPU's, especially with the 3rd gen, because where can they go after 32 cores? Adding more cores is going to start eating up in their professional CPU's, so I think they are removing it because it will essentially compete with their Epyc line and there is probably limited 7nm chips to be able to do Ryzen, Threadripper, Epyc AND navi!

I think it might come, but only late Q4 2019 or even 2020.
 

coercitiv

Diamond Member
Jan 24, 2014
7,092
16,360
136
Yeah, we might as well think of it this way: all TR chiplets for 2019 have already been sold
 

Guru

Senior member
May 5, 2017
830
361
106
The advantage for GPU is different. The 5GHz barrier isn't relevant for GPUs in the foreseeable future. Localized heat is also less of a problem.

I expect high single digits or around ~10% gains for 7nm at 4GHz+ over 12nm, which is a performance oriented version of 14LPP+

Shintel is using a highly refined mature 12nm-like node with high speed libraries. They've also gotten more aggressive with elite binning. That's how they are able to reach high frequencies.

I am guessing we'll beat 4.5ghz boosts on good binnings. 5ghz would be a really pleasant surprise, and a big surprise. I do think those speeds will be surpassed in 2020 with a more enhanced and mature 7nm node.
I mean 7nm in not a new node anymore, yeah it's still fresh, but considering TSMC have 6nm now and are working swiftly on a 5nm, I think their 7nm is probably much more mature right now as opposed to how some people are looking at it from the outside.

7nm was developed early last year, we had 7nm engineering samples from AMD in Q2 2018, so by the time Ryzen 3000 releases its going to be one and a half year since 7nm process.
 

DrMrLordX

Lifer
Apr 27, 2000
22,496
12,368
136
Yeah, we might as well think of it this way: all TR chiplets for 2019 have already been sold

Wouldn't surprise me in the least. If AMD only launches 12c chips on AM4 this summer, their existing TR2 products will still have some market value for people who need the high core count. That way they can prioritize top-bin chiplets to EPYC where the margins are higher.
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,800
1,283
136
Matisse is more likely to get cut than Castle Peak; Rome and Castle are extremely high ASPs for relatively small 7nm dies. AMD had Vega at >400 mm^2, so a big 14LPP chiplet isn't a huge concern.

2990WX => $1,799.00
2700X => $329 * 4 => $1316

Top-end has the lowest volume, but has most of the margin for AMD. If they can make more people buy Castle over Matisse, even better for them.
 

H T C

Senior member
Nov 7, 2018
600
432
136
BUT: TR and epyc don't need the same chiplets. Epcy needs low leakage low clocking chiplets while TR rather need high leakage / high clocking chiplets at the cost of higher TDP. TR competes with Ryzen for chiplets IMHO.

This is true. However, TR's performance would seriously compete with Epyc's performance if Zen 2 chiplets were to be introduced in the platform and that's a serious issue for AMD since Epyc's CPUs cost a whole lot more to customers than TR CPUs do.

To prevent this, they block Zen 2 chips from the socket. Remember: TR CPUs have much higher speeds than Epyc CPUs, which is why putting Zen 2 chips in them would have their performance dangerously close to Epyc's, @ a much lower cost.

Furthermore: AMD promised AM4 socket compatibility up to 2020 but i don't recall hearing the same regarding TR3 platform. Perhaps i'm remembering wrong?
 

DrMrLordX

Lifer
Apr 27, 2000
22,496
12,368
136
What's "Shintel"?

It's Japanese for "true Intel". Heh.

BUT: TR and epyc don't need the same chiplets.

If you look at the 2990WX, it has lower boost clocks than the 2950X or 2700x. And the dice used in the 2990WX (and 2950X) needed to be top 5% bins from their respective leakage segments. So 2990WX at least probably would have competed with a hypothetical 12nm EPYC for dice, as 12nm EPYC ever existed.

It stands to reason that a 32c TR3 would compete with Rome/EPYC 2 in a similar fashion.
 
Last edited:
Reactions: amd6502

TheGiant

Senior member
Jun 12, 2017
748
353
106
the next gen 7nm mobile APU - any timings known?

looking for a replacement for my i5 6300U, icelake 4C is one candidate but new ryzen is ofc too !
 

amd6502

Senior member
Apr 21, 2017
971
360
136
It stands to reason that a 32c TR3 would compete with Rome/EPYC 2 in a similar fashion.

Yup, so the highest core count I could see for a new threatripper is 24c. But most likely even quite a bit lower, maybe dual chiplet 12c and 16c config at most. Zero at worst.

They would also want to wait long enough until they've accumulated a pile of salvageable Rome IO hubs.

I believe the 2990WX has 16c/32t main cores and the same amount of compute cores which have higher latency to memory. Rome based TR wouldn't have this limitation, and given the freq, FPU, and IPC boost, a 24c TR 3000 would outperform this flagship TR at many/most tasks.
 
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |