AMD Zen 2 (Ryzen 3000) - *** NO COMPETITOR HINTING ***

[Perfect_Chaos]

Anyone any guesses on the all core turbo of the 3700x? Probably/hopefully 4.6 at least? With the IPC boost i wonder how gaming will be, that CPU should easily perform better than the 9900k in other tasks as well providing it's correct.

Hope there's enough of this to go around on their chips

 

[DragonQ]

Definitely getting one of those R7s if these leaks end up being true. PCIe 4.0, 12c/24t, probably >60% IPC improvement from my Westmere chip, higher clock speeds, and from the looks of the TDPs listed they're probably power efficient up to 4.0 GHz or maybe even higher (as opposed to ~3.5 GHz of 14nm Zen). First chip in the last decade that feels like it might last as long as the X58 platform.

 

[dragonslayer]

I wouldn't at all be surprised if AMD made their low end 6/8 cores because that would mean intel can't even compete especially if the 3xxx is around the £1-200 marks

 

[Potatowithearsontheside]

That's probably more to do with the interconnect (the data fabric part of IF) *IIRC something like 50% or more of the power envelope goes towards that, the cores take up surprisingly little power, probably more so with the node shrink.

*I think Anandtech reported on it sometime back, although maybe I'm misremembering, i would try and dig it up but as you've already experienced some people are rather hostile when it comes to having an open and honest discussion.

EDIT: Found it: I did misremember, Infinity Fabric consumes an almost constant 55-60W of the total on a 20 core layout so as core count goes down or up it takes up a higher or lower percentage of the total.

2nd EDIT: Found a more detailed look at power requirements of interconnects.

Thanks for those links. I did vaguely remember seeing them in the past but couldn't recall where.
The problem is, they support my view that the TDP figures look wrong. If the uncore (Infinity Fabric) is taking up a pretty static amount of power, that barely rises with additional cores, then each additional 4c should only be adding incremental amounts of power, yet we see a 10w difference between the 3600X and the 3700X, followed by a 30w increase to the 3800X. Sure, there are marginal base clock increases to go with the core increases, muddying the numbers a little. However, it's the scale of the increased power relative to the clock bumps that is screaming out to me that the ACT really can't be too much higher than 0.3GHz above base, unless the 3600X and 3800X have conservative TDP ratings. The 3700X looks like it lines up nicely with the rest of the product stack.
4.6GHz ACT on 16c would be immense though, especially combined with the expected 9-13% IPC increase for Zen 2.
We'd be looking at a 20-25% better performing CPU (in MT tasks; ST would be even better - 31-38%), and that's before we consider the doubled core count for the 3800X.
If all of this is true, I don't see how Intel could possibly compete for a few years.
For me, it seems too good to be true, but from all the information that's available it does at least seem feasible.

 

[LePhuronn]

@Potatowithearsontheside

Why are you getting so hung up on TDP? Everybody measures their TDP differently so it's at best a very rough idea of how much cooling you need, and in the end it never really matters a whole deal. What's so weird about a 10W increase between 8 cores and 12 cores, and then 30W between 12 and 16 cores? You're assuming TDP scales linearly, maybe it doesn't when an AM4 package is loaded with 16 working cores. Maybe AMD are overstating the TDP of the monster Ryzen 9s to prevent luddites strapping a boggo cooler to it and expecting 5.1GHz boost.

If you're fixating on these Ryzen numbers making no sense to you, remember this: i9 9900K is 95W TDP.

 

[Journey]

The 2950X does 4.4GHz all core when over clocked with about 220w being used. Just putting it out there.

 

[d_brennen]

@Potatowithearsontheside

If you're fixating on these Ryzen numbers making no sense to you, remember this: i9 9900K is 95W TDP.

"95w" that's actually 160+w Ah but it's ok because Intel don't measure boost clocks, only base clocks under an unknown (presumably tame) load

 

[LePhuronn]

...only base clocks under an unknown (presumably tame) load

I believe the official term is "Intel-defined, high-complexity workload".

 

[Potatowithearsontheside]

The 9900K runs 24/7 at 4.2GHz within 95w.

***

The reason I'm interested in those figures is for what it potentially tells us about where the ACT will likely be at. Since AMD and Intel measure TDP in different ways, with AMD tending towards actual ACT being at the TDP limit, it gives us an idea of the real world performance of the CPU. Nowhere in the product spec for the 9900K does it say that it'll 24/7 at 4.2GHz within 95w, so this kind of information could/should be used as a selling point for AMD.
IMO, being open and clear with specs is a huge thing. We just don't see it in tech right now, when we deserve to.

 

[Murphy]

Thanks for those links. I did vaguely remember seeing them in the past but couldn't recall where.
The problem is, they support my view that the TDP figures look wrong. If the uncore (Infinity Fabric) is taking up a pretty static amount of power, that barely rises with additional cores, then each additional 4c should only be adding incremental amounts of power, yet we see a 10w difference between the 3600X and the 3700X, followed by a 30w increase to the 3800X. Sure, there are marginal base clock increases to go with the core increases, muddying the numbers a little. However, it's the scale of the increased power relative to the clock bumps that is screaming out to me that the ACT really can't be too much higher than 0.3GHz above base, unless the 3600X and 3800X have conservative TDP ratings. The 3700X looks like it lines up nicely with the rest of the product stack.
4.6GHz ACT on 16c would be immense though, especially combined with the expected 9-13% IPC increase for Zen 2.
We'd be looking at a 20-25% better performing CPU (in MT tasks; ST would be even better - 31-38%), and that's before we consider the doubled core count for the 3800X.
If all of this is true, I don't see how Intel could possibly compete for a few years.
For me, it seems too good to be true, but from all the information that's available it does at least seem feasible.

I get what you mean but as we don't know the layout of the consumer orientated chips it's difficult to guess at how much power is being used in the interconnect, take for example the 3600X, will all 8 cores be contained within a single CCX or split across two, a single CCX would reduce the power needed for the interconnect but also means you can only use perfect CCX's, if you split those 8 cores across two CCX's you can copy the physical layout of higher end chips that have to use more CCX's but you can disable defective cores within a selection of CCX and still reach your 8 core goal.

I suspect what they've done is design a chip around a two CCX physical layout and everything else is a derivative of that, that's great from a design and cost perspective but it also means that even when you're only using, say, 8 cores you have to light up the non-core stuff for what would be a 16 core chip.

EDIT: Also having just double checked isn't the increase in TDP more along the lines of 10W from the 3600X to the 3700X and 20w from the 3700X to the 3800X.

 

[peahead]

Could not see it posted so another leak from a Russian Etailer.

https://www.guru3d.com/news-story/a...her-7nm-procs-surface-at-russion-etailer.html

Ryzen 7 3700X 12C/24T 4.2 GHz 5.0 GHz 105W

 

[DragonQ]

Could not see it posted so another leak from a Russian Etailer.

https://www.guru3d.com/news-story/a...her-7nm-procs-surface-at-russion-etailer.html

Ryzen 7 3700X 12C/24T 4.2 GHz 5.0 GHz 105W

I thought that one was actually before the R9 3800X?

 

[Potatowithearsontheside]

It's the same Russian seller.

@Murphy, yes, you are correct that the 3800X is 20w higher. It seems as though I should have been listing the 3850X in my posts, as that is the one that gets a clock increase along with the +4c.

 

[humbug]

The 9900K runs 24/7 at 4.2GHz within 95w.

***

Doing what, browsing the net?

Depending on what its doing at stock its anything from 160 to 200 Watts, also look at the 12 core 14nm Threadripper at just 5% more power.

 

[pete910]

Is that a typo in the bottom chart, Or is it the manufacture ?

"Witch Industrial Chiller"

 

[humbug]

Is that a typo in the bottom chart, Or is it the manufacture ?

"Witch Industrial Chiller"

Probably "With" but defiantly a Pee take on Intel's 5Ghz 28 core Male Bovine Manure

 

[LePhuronn]

The 9900K runs 24/7 at 4.2GHz within 95w.

Nowhere in the product spec for the 9900K does it say that it'll 24/7 at 4.2GHz within 95w...

Um, OK...which is it then?

The reason I'm interested in those figures is for what it potentially tells us about where the ACT will likely be at. Since AMD and Intel measure TDP in different ways, with AMD tending towards actual ACT being at the TDP limit, it gives us an idea of the real world performance of the CPU.

So what's the problem with 16c/32t at 135W then? Why does it matter then that 12 to 16 cores is a 30W increase, when 8 to 12 is 10W? You've already said you're not a troll, which I believe, but there is a big difference between taking rumours at face value and going out of your way to utterly dismiss them. Focussing on a totally arbitrary number that means little really comes across as the latter...

 

[pete910]

Male Bovine excrement

FTFY

 

[humbug]

^^^^^^^

The 12nm 16 core 2950X run just within its TDP of 180 watts @Potatowithearsontheside

There is no reason why a 7nm 16 core Ryzen shouldn't run at 135 Watts.

 

[pieman109]

When the original Ryzen came out similar things were being said about the 6 core and 8 core parts and their leaked costs. They actually turned out to be pretty accurate.

We are just used to being drip fed the most basic of upgrades from Intel for the last 8 years or so and while I am not 100% sure myself that these prices and specs are accurate, I see no reason they couldn't be.

Oh I really hope they're right 'cos if they are, it means great performance for peanuts. I mean really, 4.8ghz boost for £230 or thereabouts, say £250-£275, still an absolute bargain without having to change motherboards (and the issues it'll bring windows 10 activation) with Intel.