Speculation: Ryzen 3000 series

[RaV666]

Similar but different for me.
My nearly 10 year old Q9659 machine has been trouble free until about 2 years ago, small failures then bigger mysterious failures...

Well i can tell you that mine old phenom x6 on foxconn A79-S is stil chugging along nicely with no problems oced with ddr2 mem at 1066Mhz .
Sorry to hear about your issues, as a rule of thumb i never use gigabyte, they always have some quirks or problems.

 

[Fanatical Meat]

We’re about 28 hours away from more info

Dr. Su should troll us and sing In-A-Gadda-Da-Vida for 2 hours at her presentation

 

[DarthKyrie]

We’re about 28 hours away from more info

Dr. Su should troll us and sing In-A-Gadda-Da-Vida for 2 hours at her presentation

Now that would be hilarious.

 

[Tryad]

I, like many others are waiting on this to decide what I will be building. For so many years its almost been a meme that AMD would launch a new product and it would be competitive at the top end for gamers.

This time at least feels different. Lots of smoke around the 12c chip actually reaching 5GHZ. Hard to believe, but I'm in for one if it manages to do so. Its been a long time though.... since the Athlon 64 days perhaps? Its hard to remember now...

 

[Hitman928]

We’re about 28 hours away from more info

Dr. Su should troll us and sing In-A-Gadda-Da-Vida for 2 hours at her presentation

I would prefer something like this

 

[Timmyotule]

I, like many others are waiting on this to decide what I will be building. For so many years its almost been a meme that AMD would launch a new product and it would be competitive at the top end for gamers.

This time at least feels different. Lots of smoke around the 12c chip actually reaching 5GHZ. Hard to believe, but I'm in for one if it manages to do so. Its been a long time though.... since the Athlon 64 days perhaps? Its hard to remember now...

Yep. Even if it's only an eight core part hitting 5.0 GHz with increased IPC it will be really tempting. If it actually hits the leaked numbers it the 3700X will be four times faster than my overclocked 3570k at multithreading and actually have increased single thread performance!

 

[.vodka]

Yep. Even if it's only an eight core part hitting 5.0 GHz with increased IPC it will be really tempting. If it actually hits the leaked numbers it the 3700X will be four times faster than my overclocked 3570k at multithreading and actually have increased single thread performance!

Zen (1xxx parts) already has higher single threaded performance over your 3570k. It just can't clock as high, but doesn't really need to.

3.8GHz Zen ~ 4.5GHz Sandy ~ 4.3GHz Ivy (~+5% over Sandy)

Zen+ (2xxx parts) is +3% over Zen. Zen2 seems to be +10-15% on average over Zen+ according to geekbench save for workloads that benefit from the upgraded SIMD hardware (similar to Ivy -> Haswell) where it's even faster.

Factor in meltdown/spectre/L1TF/MDS mitigations... then for a Sandy/Ivy 5GHz capable golden sample, its ~1000MHz clock margin over 1xxx parts and ~800MHz margin over 2xxx parts has just been made irrelevant.

Sandy was legendary and Ivy was a nice incremental improvement, but they're well past retirement for anything high performance thanks to the vulnerabilities and its mitigations. Sorry.


Anyway, whatever upgrade you do, most of the improvement will come not from higher ST performance, but by having a ton more threads available. 4 threads in 2019 can be painful.

 

[Timmyotule]

Zen (1xxx parts) already has higher single threaded performance over your 3570k. It just can't clock as high, but doesn't really need to.

Anyway, whatever upgrade you do, most of the improvement will come not from higher ST performance, but by having a ton more threads available. 4 threads in 2019 can be painful.

I'm not sure what your sorry for. According to Passmark for stock speeds (though I do have my CPU overclocked) the 1600x has a few percent lower single threaded performance and the 2600x has a few percent higher single threaded performance compared to the 3570k: https://www.cpubenchmark.net/compar...-5-1600X-vs-AMD-Ryzen-5-2600X/828vs3000vs3235 Maybe there are better performance benchmarks to use?

My hardest use scenarios is primarily limited by single thread performance though I don't use my system very hard at all these days. Upgrading and entire platform for more cores (that I'm not really using) and similar single threaded performance hasn't made sense. I'm hoping tomorrow I'll have a firm reason for upgrading.

 

[Mockingbird]

a

 

[tamz_msc]

It has both single-core and multi-score that are very close to those of the Ryzen 7 2700X

So what? You mean to say that a 6C/12T cannot match an 8C/16T if it's fast enough?

 

[Mockingbird]

So what? You mean to say that a 6C/12T cannot match an 8C/16T if it's fast enough?

How come single-thread performance barely went up while multi-thread performance went up by almost 20%?

 

[OTG]

http://browser.geekbench.com/v4/cpu/13241660

I think the processor has been misidentified as a 6-core and is actually an 8-core.

It has both single-core and multi-score that are very close to those of the Ryzen 7 2700X

If the +15% IPC gain is close to real, and since clocks should at minimum match the 2700x, 6 cores beating 8 isn't unrealistic.

 

[Mockingbird]

If the +15% IPC gain is close to real, and since clocks should at minimum match the 2700x, 6 cores beating 8 isn't unrealistic.

How come single-thread performance barely went up while multi-thread performance went up by almost 20%?

 

[OTG]

How come single-thread performance barely went up while multi-thread performance went up by almost 20%?

IDK, you may be right, and it's just clocked kinda low. Should find out soon enough though!

 

[piesquared]

How come single-thread performance barely went up while multi-thread performance went up by almost 20%?

Clock speed and scaling?

 

[scannall]

The thing I am most excited about is PCIe 4, NVME drives. 5GBPS would be really nice.

 

[tamz_msc]

How come single-thread performance barely went up while multi-thread performance went up by almost 20%?

Compare apples to apples - Pinnacle Ridge 6C/12T at 4GHz vs this sample, also at 4GHz.
http://browser.geekbench.com/v4/cpu/compare/13241660?baseline=13267824
ST up by ~11% MT up by ~15%

 

[PotatoWithEarsOnSide]

The benchmark shows +4% in ST over the 6c 2600X, and +18% in MT. That is consistent with what you'd expect from scaling if you consider that AMD are clearly also targetting MT scaling if the Zen 3 rumours are indeed true.
The benchmarks mean nothing for IPC without knowing what clocks they were at in any event.

 

[DrMrLordX]

You get FX9590 power consumption, and the performance to go along with it.

In all fairness to the 9900k, the FX9590 never sat at the top of the desktop performance heap. The 9900k is top dog until Monday, not counting HEDT.

And with the impending 10 core skylake on a yet another socket i guess they learned nothing.

I don't think Comet Lake will require a new socket. At least some of the Z390 boards should be able to support 10c/20t Comet Lake.

We’re about 28 hours away from more info

Dr. Su should troll us and sing In-A-Gadda-Da-Vida for 2 hours at her presentation

Not sure if she wants to get as stoned as is necessary to actually sing that song. Plus 2 hours requires a lot of endurance. She can do relay singing with William Hung, William Shatner, and . . . Will Ferrell?

 

[coercitiv]

I don't think Comet Lake will require a new socket. At least some of the Z390 boards should be able to support 10c/20t Comet Lake.

The rumor is new socket pinout, with more pins for power delivery. (similar move to what we had in Z270 --> Z370 transition).

I'm pretty sure I read about it from a better source than this, but for now it escapes me.

 

[beginner99]

Also, this one also points to the L1i cache being changed to 32KB, right?

What I found more interesting was the 2*8mb l3 cache. I though zen2 doubled it to 32mb? or maybe they use parts with broken cache for 6-cores. Point is a 8-core could have higer ST because of more cache

 

[DrMrLordX]

The rumor is new socket pinout, with more pins for power delivery. (similar move to what we had in Z270 --> Z370 transition).

I'm pretty sure I read about it from a better source than this, but for now it escapes me.

Eww. And yet we have Zen2 launching on AM4. Personally I wouldn't want to run Zen2 on X370, but it's possible.

 

[CHADBOGA]

Disagreeing with someone or expressing an opinion is not against the rules. Only if you are abusive.

Shintai suggests otherwise.

 

[itsmydamnation]

Shintai suggests otherwise.

Shintai run away because when e push came to shove his knowledge couldn't back his rhetoric, but he couldn't help himself and so people like me just kept asking over and over to justify his positions.

 

[CHADBOGA]

Shintai run away because when e push came to shove his knowledge couldn't back his rhetoric, but he couldn't help himself and so people like me just kept asking over and over to justify his positions.

That's not what happened.