Speculation: Ryzen 3000 series

[Fanatical Meat]

sure they do

but that 3600X=9900K, well.... I am more conservative about that one

sometimes I think that hype has gone too far...

wait what's the date? 23.6.2019 when is the f..nice NDA expiring?

I thought it was up around mid month however I appear to be wrong. Starting to think NDA runs until release.

 

[Asterox]

sure they do

but that 3600X=9900K, well.... I am more conservative about that one

sometimes I think that hype has gone too far...

wait what's the date? 23.6.2019 when is the f..nice NDA expiring?

There is no hype, if you now what is Geekbench or prety garbage usseless benchmark.AMD has no problem(allowed) with "Geekbench leaks breaking NDA", or is like shoting 22caliber bullet in military special forcess bulletproof vest.

There is good reason(except 16/32 Ryzen 3000/5ghz overlocking LN), that still in wild there is not a single one leaked Cinebench score.What can you do with Geekbench, compared to Cinebench or benchmark based on 3000$ Maxon Cinema4D software for professionals.

 

[DrMrLordX]

At $250 it looks like it might just pack a good punch. Guess the official reviews should be interesting to say the least.

I thought the R5 3600 was going to be $199? 3600x was spendier.

 

[Kenmitch]

I thought the R5 3600 was going to be $199? 3600x was spendier.

Yep. 3600x is $249

 

[maddogmcgee]

And (some) people were expecting the 6c/12t SKUs as R3 for close to $100 because of the god-given right to free beer.

Some rights are worth dying for.

 

[moinmoin]

Interesting. Although one would expect in practice the difference from a 3800X would be indistinguishable. Can't imagine there's a current or soon to be released game out there that really benefits from the 8 extra cores. I'm really curious how an OC'd 3800X will compare to an OC'd 3700X.

Ryzen 3xxx is not only segmented by amount of cores but also by binning/max boost speed. All other parts being the same (indeed indistinguishable) 3950X will beat both 3800X and 3700X in games due to the higher boost speed. (I accordingly believe the upcoming Threadripper 3 to be even ever so slightly better in games due to even higher boost speed and more TDP headroom.)

Yeah, however from working at big companies in my past I’m 100% sure intel already knows within a few percentage points exactly what Ryzen will do.
They have all the resources to run simulations, they have all the skill to estimate performance from published stuff and like all large companies I’m sure they own one share of AMD stock to request the full financial report that they then dig thru for data and timelines.

In that case I hope for Intel their predictions for AMD products are more spot on than for their own 10nm products.

 

[DarthKyrie]

Some rights are worth dying for.

I think the spelling in the 2nd Amendment is a little off, I think they meant to write that we have the right to beer arms.

 

[Ajay]

Yeah I was leaving that off the table because there really should be proof but don’t kid yourself and think intel or AMD for that matter doesn’t lean on big vendors for information or at minimal listen for hints.
I could easily see intel saying I’ll give you a licensing break to a motherboard vendor that “accidentally” leaks some Ryzen info.

That would be outright illegal, but it doesn’t matter. I worked in product development for a midsize company and one of my tasks was competitive analysis. 'Somehow' the sales guys were always sending me our competitor's products. Also, as alluded to, shared vendors would also cough up plenty on what others were doing. So, I am 99,9% certain that Intel not only has all the information that they need, but that they already have AMD products in their labs (even if they are only ES).

 

[Thunder 57]

That would be outright illegal, but it doesn’t matter. I worked in product development for a midsize company and one of my tasks was competitive analysis. 'Somehow' the sales guys were always sending me our competitor's products. Also, as alluded to, shared vendors would also cough up plenty on what others were doing. So, I am 99,9% certain that Intel not only has all the information that they need, but that they already have AMD products in their labs (even if they are only ES).

And I'm sure it goes both ways. Imagine the reaction Intel must've had when they first got some Bulldozer numbers/samples?

 

[DrMrLordX]

And I'm sure it goes both ways. Imagine the reaction Intel must've had when they first got some Bulldozer numbers/samples?

"Well boys, we won't need that 14nm process anytime soon. Take a break, have a beer".

 

[Atari2600]

I think the spelling in the 2nd Amendment is a little off, I think they meant to write that we have the right to beer arms.

Does that mean they would have to ban half-pints? Never gonna build a beer arm up drinking half pints.

I'm all for that ban!

 

[RTX2080]

http://tieba.baidu.com/p/6173322333?pid=126254261673&cid=0#126254261673
http://tieba.baidu.com/p/6172938978?pn=1

R5-3600 leak
R15 ST 197 MT 1569
CPUZ ST 508 MT 4114

NOTE:
original leaker said it's stock result
stock all core load 3.6Ghz

edit: ouch the original leak/thread was deleted lol

 

[inf64]

Effect of 20% faster DDR4 memory on GB4 results (albeit on a different motherboards)

https://browser.geekbench.com/v4/cpu/compare/13631495?baseline=13650304

Both CPUs seem to run at about the same clockspeed (add .gb4 at the end of each URL). ST difference ~5.2% and MT difference ~10.5%

 

[IntelUser2000]

Effect of 20% faster DDR4 memory on GB4 results (albeit on a different motherboards)

You need to test it yourself on the same platform. Geekbench can vary 10x with same hardware specs.

User submitted scores are a bad way to compare 5-10% differences. Also think of how badly memory bandwidth bound it has to be to have a 10% difference for a 20% memory boost! That's why it can't be trusted.

 

[realibrad]

So, I need everyone to buy as many chips as they can, test them, and post the results so I only have to buy one.

I'm looking at you @Markfw

With all the leaks, I'm now considering the 3900x. What I'm hoping is that the 3800x will have better binned chiplets and that the 3900x is less able to overclock, which would make me less inclined to spend more on the 3900x. I think its possible that with the disabled cores, they could actually clock higher though, given that the heat/surface area will be better. It feels so good to have options.

 

[Fanatical Meat]

Off topic, earlier I read the x590 stuff found in some bios updates was actually a real product however at some point AMD decided to merge most of its features with the X570 and discontinued development of the x590.
Sounded like the 570 wasn’t going to have all the pcie 4 lanes or might not have had any and the 590 was going to be a HEDT line with either the same amount or more pcie 4 lanes.

 

[realibrad]

Off topic, earlier I read the x590 stuff found in some bios updates was actually a real product however at some point AMD decided to merge most of its features with the X570 and discontinued development of the x590.
Sounded like the 570 wasn’t going to have all the pcie 4 lanes or might not have had any and the 590 was going to be a HEDT line with either the same amount or more pcie 4 lanes.

From what I understand, the major difference was supposed to be PCIe 4 in X590, but, because it was ready they put it in X570. At least, that was I remember from the GN video.

 

[Markfw]

So, I need everyone to buy as many chips as they can, test them, and post the results so I only have to buy one.

I'm looking at you @Markfw

With all the leaks, I'm now considering the 3900x. What I'm hoping is that the 3800x will have better binned chiplets and that the 3900x is less able to overclock, which would make me less inclined to spend more on the 3900x. I think its possible that with the disabled cores, they could actually clock higher though, given that the heat/surface area will be better. It feels so good to have options.

Well, I just put together my 2nd 2990wx, so I really wasn't planning on it. I now have 2 of my 14 core Xeons, 2 2700x's and an 1800x idle, so buying another one wasn't really in the plans.

But we will see. Sometimes I just can't control myself.

 

[lopri]

What do you think the chance of any new CPU hitting 5 GHz? Single or all-core.

Edit: overclocked ofc.

 

[realibrad]

Well, I just put together my 2nd 2990wx, so I really wasn't planning on it. I now have 2 of my 14 core Xeons, 2 2700x's and an 1800x idle, so buying another one wasn't really in the plans.

But we will see. Sometimes I just can't control myself.

Well, I figure you already do work for me by cleaning the forum, so what is one more thing to expect of you to do for me without payment.

 

[coercitiv]

What do you think the chance of any new CPU hitting 5 GHz? Single or all-core.

It's a lower chance than any new CPU beating a 5Ghz Skylake in both ST and MT using lower clocks.

 

[exquisitechar]

Looking forward to this.

3 options for overclocking, eh? Interested in what's the max "safe" voltage, too.

 

[Elfear]

What do you think the chance of any new CPU hitting 5 GHz? Single or all-core.

Edit: overclocked ofc.

It's a lower chance than any new CPU beating a 5Ghz Skylake in both ST and MT using lower clocks.

Agreed. Personally, I think the 3000 series will be clocked close to the max similar to the Zen 1. The top clocking Zen 2 is the 3950X with a boost speed of 4.7Ghz (single core). To reach 5Ghz would require a 6.4% overclock. On a single core the 2700X boosts to 4.35Ghz so a 6.4% increase would be ~4.63Ghz. How many 2700X's reach those speeds on non-exotic cooling?

Having said that, Zen 2 is on a revamped architecture and a new node so we may see better overclocks than Zen 1.

 

[Mopetar]

What do you think the chance of any new CPU hitting 5 GHz? Single or all-core.

Edit: overclocked ofc.

Really good. The top Ryzen already has a 4.7 GHz boost and that’s at the 105W TDP. Overclocking would have to be absolutely abysmal or you’d have to get an absolutely awful 3950X that couldn’t be pushed to hit a 5 GHz single core clock.

All core is much less likely. Assuming such chips exist at all, AMD would prioritize them for Threadripper because people will pay more for HEDT parts and that silicon will be too rare to sell at the typical consumer price points.

 

[jamescox]

For cooling, balance and packaging all SP3 socket EPYC's are going to be either 4 or 8 dies. That's why the 8 chiplets version looks like it does. They can remove a die from each pair. That said you can't have 8 cores on an 8 chiplet EPYC. The smallest of those will be 16c but I have a feeling the two configurations merge at 32c and anything less is always 4 chiplets.

For the 32 to 64 core parts, the L3 cache size changes with the core count, so they probably have different numbers of active CCX. If they did 12 active CCX across 8 chiplets, then that would be a bit asymmetric. It would be roughly the same if all communication does go to the IO die, but it would still be different bandwidth. For cache coherency, all memory accesses will have to send a probe to the IO die, but it might be possible to transfer cache lines directly between cores on the same chip.

Anyway, they may have 4, 6, and 8 chiplet devices with anything over 32 cores always using full 4 core CCXs. The 48 core is then just a 6 chiplet device. With 32 core and under, they have different amounts of cores with the same amount of cache, so those probably do the 2, 3, or 4 active cores per CCX. I don’t think they will have anything under 64 MB cache with an 8 or 16 core. With 4 chiplets, the 8 core with 64 MB L3 would be 4 active CCX, 1 per die. You would need 2 cores per active CCX. For 128 MB L3, the number of active CCX would probably be 8, so all CCX in 4 chiplets would be active. They only go down to 16 core for 128 MB of L3 on the Wccftech list, so that is still at least 2 cores per CCX.