• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Zen 3 (Ryzen 4000) already in the works

Soldato
Joined
26 Sep 2010
Posts
7,146
Location
Stoke-on-Trent
I get what you are saying. But I would have released say 100 or 1000 of them at a higher cost and a special name. Would do wonders for marketing imo. Think of them as titans or something.
I'd love for that too, if anything to just shut everybody up :p

Right now AMD needs as big a war chest as possible, does literally every cent count right now? Let's have a look...

1,000 5GHz 3950XE (yes, I'm stealing your suffix, Intel) is 2,000 golden sample chiplets. Sell those CPUs at £999, that's £999,000.

Those 2,000 chiplets can make 250 EPYC 7742, which sell at £5,499 each. That's £1,374,750.

Is the loss of 375 grand worth the ePeen and headlines? Actually, given how AMD do suck at marketing, I'd be inclined to say "yes".


AMD, you're failing at life again! :p
 
Associate
Joined
27 Apr 2007
Posts
963
Says who? I take it you missed the part in my comment about "aggressive binning".

What we have in our hands right now is arguably junk silicon. All of the best stuff was always dedicated to server anyway, but with demand for Rome much higher than AMD originally anticipated (again, they didn't project Intel being so woeful), all the 2nd best yields have been diverted to server as well. That leaves us peasants with 3rd or even 4th best silicon for our silly little home computers. Hell, the 3950X is evidence that desktop Ryzen is getting the lower quality yields. How can the 16 core part have a higher boost, lower temperatures and the same TDP as the 12 core part? BINNING.

I guarantee you if you take a bunch of golden sample 6 and 8 core chiplets going into top-end Rome and slap them under a 360 AIO or massive Noctua then you'll get 5GHz boost.
Looking at the reviews of ECO mode for the 39xx chips, it strongly suggests that even at stock settings they have been pushed significantly beyond the optimal power efficiency point.
They can do this as Intel’s power efficiency is so woeful that they still look good by comparison.
In Eco mode the chips offer astounding efficiency.
Overclocked they are pigs as you’d expect as by default they are already boosted outside the efficient operating range.
This is what AMD do, they bin very aggressively with GPUs also.
Looking at the increased TDP of the TR3 chips and other factors make me think that the current 7nm process struggles with high clock speeds due to power issues.
Not an issue for servers or for TR3, although TDP has gone up for TR3.
So 16C at 5GHz sustained seems like a pipe dream at stock settings where the PPT is 142W I think!
This is an interesting graph from here:
https://www.anandtech.com/show/15043/the-amd-ryzen-9-3950x-review-16-cores-on-7nm-with-pcie-40/2

3950X%20PowerLoading_575px.png


Personally, if I needed many more cores than I have, I don’t, I’d be looking at the 39xx series in ECO mode, which for me is the only place where they fully shine and offer what I hoped 7nm multi core would.
Zen 3 on 7+ nm also looks very interesting as there’s a larger potential for gains than I initially expected based on recent revelations from AMD. They’ve really got my attention now in away that they didn’t prior to Zen 2.
 
Soldato
Joined
26 Sep 2010
Posts
7,146
Location
Stoke-on-Trent
Fully agree with you there, smilingcrow, but all of that data is based on what we have available to us. This is what I've been suggesting: what we have in our hands is "junk" 3rd or 4th tier silicon because all the good stuff must go to EPYC, so you're trying to suggest Zen 2 can't go mega clocks and mega efficiency based on what the bottom end hardware can do. That skews the picture.

How about we find some soldering wizard, buy a EPYC 7742 and 4 3900X and swap the chiplets out :D let's see what they can do then.
 
Associate
Joined
27 Apr 2007
Posts
963
1,000 5GHz 3950XE (yes, I'm stealing your suffix, Intel) is 2,000 golden sample chiplets. Sell those CPUs at £999, that's £999,000.
Those 2,000 chiplets can make 250 EPYC 7742, which sell at £5,499 each. That's £1,374,750.
Is the loss of 375 grand worth the ePeen and headlines? Actually, given how AMD do suck at marketing, I'd be inclined to say "yes".
AMD, you're failing at life again! :p

Those EPYC chips list at $6950 so in the UK with VAT they will be a lot more than you suggest.
If they released a 3950XE at $1,000 which would equate to a RRP of £1,000, they would lose nearly £750,000 in terms of RRP. How much of a loss that would be in terms of gross turnover or profit is much harder to quantify as you’d need to know the wholesale pricing and the other costs of manufacturing.
Releasing a very limited edition part of only a thousand worldwide would mean the UK might get 50 or less.
Considering the frustration being expressed over the 3950X and the potential for price gouging, I can only imagine how much worse it would be for a 3950XE.
Probably bad PR and as well as loss of income and all for what!
Keep the good chips for EPYC as they really need to deliver there and it’s much more important than pleasing a few Youtube goons.
 
Associate
Joined
27 Apr 2007
Posts
963
Fully agree with you there, smilingcrow, but all of that data is based on what we have available to us. This is what I've been suggesting: what we have in our hands is "junk" 3rd or 4th tier silicon because all the good stuff must go to EPYC, so you're trying to suggest Zen 2 can't go mega clocks and mega efficiency based on what the bottom end hardware can do. That skews the picture.


How about we find some soldering wizard, buy a EPYC 7742 and 4 3900X and swap the chiplets out :D let's see what they can do then.

Assuming that TR3 gets very good chiplets, then how easy will it be to disable enough to run tests with only 16 active to see how far they can be pushed at sensible voltages?
That’s probably the only test that can get close to checking your theory.

Your speculation that even the 9 series are getting 3rd or 4th tier silicon is just that.
Plus we don’t know the difference between the various tiers in terms of actual performance when it comes to desktop configurations.
 
Soldato
Joined
26 Sep 2010
Posts
7,146
Location
Stoke-on-Trent
Those EPYC chips list at $6950 so in the UK with VAT they will be a lot more than you suggest.
Sorry, I misread the listing from our friends at the Bolton competitor. The 7702 is £5,5000, the 7742 is actually £8K.

So with those EPYC CPUs retailing for £2M, that further highlights my point in why the best silicon has to go to server. I wouldn't say it's pure speculation though because the 3950X is clear evidence of different levels of binning on desktop Ryzen. And that 16 core running at 5GHz at Computex (although I can't find the camera phone pic of Cinebench now).
 
Associate
Joined
27 Apr 2007
Posts
963
So with those EPYC CPUs retailing for £2M, that further highlights my point in why the best silicon has to go to server. I wouldn't say it's pure speculation though because the 3950X is clear evidence of different levels of binning on desktop Ryzen. And that 16 core running at 5GHz at Computex (although I can't find the camera phone pic of Cinebench now).
So it's a good idea for AMD to release a 5GHz all-core AM4 chip and it's a good idea for them to keep them for EPYC!
Heads you win, tails I lose. :eek: :D
Surely 5GHz all-core was with exotic cooling?
Based on this and other reviews to hit 5GHz all-cores on water wouldn't so much require a golden sample but an unobtanium one. :)

power-3950x-2a.png
 
Soldato
Joined
26 Sep 2010
Posts
7,146
Location
Stoke-on-Trent
I thought you were referring to all core as not sure that the single core boost clock on a 16C chip matters that much.
It clearly matters to the Intel crowd :p
It was 5ghz all cores using LN2 and over 1.6v so not something you'll be running at home.
The closed-doors press demo at Computex? Pretty sure that wasn't under LN2 because nobody in the tech press would've cared. But I can't find the Cinebench shot and article now.
 
Soldato
Joined
6 Feb 2019
Posts
17,464
So it's a good idea for AMD to release a 5GHz all-core AM4 chip and it's a good idea for them to keep them for EPYC!
Heads you win, tails I lose. :eek: :D
Surely 5GHz all-core was with exotic cooling?
Based on this and other reviews to hit 5GHz all-cores on water wouldn't so much require a golden sample but an unobtanium one. :)

power-3950x-2a.png

A all core 5ghz 3950x can be done with LN2, but the power consumption is somewhere between 500w and 900w just for the CPU
 
Soldato
Joined
13 Mar 2008
Posts
9,638
Location
Ireland
A all core 5ghz 3950x can be done with LN2, but the power consumption is somewhere between 500w and 900w just for the CPU

Fine, so it'll run as hot as my old GTX 480, I can handle that with the new Intel Stock cooler. :p

ZjMu45G.jpg

vLSKJDT.jpg


In all seriousness though, I don't care if the new Zen 3 stuff can't hit 5Ghz. If the IPC improves again, it's a win for us all.
 
Associate
Joined
11 Dec 2016
Posts
2,010
Location
Oxford
I dont care what model naming logic AMD has used so far, but they need to make sure Zen5 processors on 5nm process releases with AM5 socket motherboards and are named Ryzen 5000. Release must happen on 5th of May. It would be criminal not to do this

(And with boosts to 5GHz)
 
Back
Top Bottom