Ryzen 3000-Series CPU Specifications Listed Online Ahead Of Official Launch

[WhtieRoseAlpha]

Our local tech and gaming website released this news just now. 3rd gen Ryzen processors are listed in the said e-commerce website. What do you guys think?

https://www.back2gaming.com/news/ryzen-3000-series-cpus-specifications-listed-online-ahead-of-official-launch/

[LukeSavenije]

I've seen that before. those are old leaks, and i still don't believe them because they seem too good to be true imo.

[WhtieRoseAlpha]

Just now, LukeSavenije said:

I've seen that before. those are old leaks, and i still don't believe them because they seem too good to be true imo.

how long ago were they listed on that website?

[LukeSavenije]

Just now, akira_ph said:

how long ago were they listed on that website?

no, not on that specific website, but here in the tech news someone shared that exact listing.

[Fnige]

[.Apex.]

well if the rumors are true, then my 8700K is equiveleant to a 100$ "low end" Ryzen 3... well shit, it hasn't even been a year since i bought it, i barely got to enjoy the glory of a high end part. fml

[Mihle]

I hope it's true. I will then upgrade from my 3570k, and get more than 3x the performance

[Doobeedoo]

Yeah I've seen this already.

Thing is, as far as clocks I can see them getting much higher due to architectural improvements and 7nm performance node and I can see possible core doubling with such a node shrink, we've seen 7nm Epyc dies.

What I find hard to believe is both od those together under that TDP envelope. It would mean these would be extremely efficient. 25x20 goal improvements ?! 

[SidM]

Ya, this was also seen on a ukranian website

[Taf the Ghost]

11 minutes ago, Doobeedoo said:

Yeah I've seen this already.

Thing is, as far as clocks I can see them getting much higher due to architectural improvements and 7nm performance node and I can see possible core doubling with such a node shrink, we've seen 7nm Epyc dies.

What I find hard to believe is both od those together under that TDP envelope. It would mean these would be extremely efficient. 25x20 goal improvements ?! 

Not really surprising. AMD took some of the efficiency improvements and some of the clock improvements to get a total performance improvement approach. The GloFo 14nm silicon hits a voltage wall between 4.1 and 4.3 Ghz. That's why it could never overcome the clocks issue against Intel.

 

For Base Clocks, AMD hasn't really increased it much. 

 

https://www.newegg.com/Product/Product.aspx?Item=N82E16819113499

 

Current 2700X is 8 cores, 105w TDP at 3.7 Ghz base and 4.35 Ghz Boost. The new 3700X is 12c (50% more) with .5 Ghz faster base clock (13%). Given how the Voltage/Power issues were with 14nm, that's really not surprising. Going from 3.7 all-core to 4.1 all-core would functionally double the power draw on Ryzen 2nd gen. Even a minor shift in the Efficiency Curve would yield these results for AMD. And we know Epyc can do 100% more cores at roughly 10% higher clocks in the same power target range between 1st and 2nd gen.

[Taf the Ghost]

The interesting implication is that Ryzen 3rd Gen might actually be earlier than thought. Maybe early March? AMD is going for the juggler, if true.

 

If 3800X will be available around launch, then I full expect AMD to show off the 3800X against the 2950X and 9900k at the same time. I think Cinebench R15 and C-Ray don't get hit by memory bandwidth issues, so the dual-channel shouldn't effect much. AMD is going to make those numbers murder Intel.

 

I'm also calling Intel coming out with a Liquid Nitrogen 6 Ghz 9900k in Cinebench.

[Taf the Ghost]

Noticed Ryzen 2nd Gen discounts were heavy during Christmas and even after. AMD probably sent out the message to get the channel cleared soon.

[porina]

21 minutes ago, Taf the Ghost said:

The interesting implication is that Ryzen 3rd Gen might actually be earlier than thought. Maybe early March? AMD is going for the juggler, if true.

I was kind hoping for spring availability to match previous gens. Cash waiting!

 

21 minutes ago, Taf the Ghost said:

If 3800X will be available around launch, then I full expect AMD to show off the 3800X against the 2950X and 9900k at the same time. I think Cinebench R15 and C-Ray don't get hit by memory bandwidth issues, so the dual-channel shouldn't effect much. AMD is going to make those numbers murder Intel.

Cinebench R15 is hardly affected by memory bandwidth, at least not on any system I've had my hands on. Makes it a good way to show off thread scaling and easy to predict scores if you know core count and clock, once you figure out the scaling value for a given architecture.

 

I'm still fearful for other uses, just how bad ram bandwidth limiting will be on a dual channel 16 core part... especially with the improved AVX performance on top.

 

21 minutes ago, Taf the Ghost said:

I'm also calling Intel coming out with a Liquid Nitrogen 6 Ghz 9900k in Cinebench.

At least aim for over 9000! Already at 7GHz ball park.

Rankings: http://hwbot.org/benchmark/cinebench_-_r15/rankings?hardwareTypeId=processor_5773&cores=8#start=0#interval=20

Current top: http://hwbot.org/submission/4000398_luumi_cinebench___r15_core_i9_9900k_3142_cb

 

Then again, they used water chiller last time, next logical step might be phase change cooling. Don't see that much these days.

18 minutes ago, Taf the Ghost said:

Noticed Ryzen 2nd Gen discounts were heavy during Christmas and even after. AMD probably sent out the message to get the channel cleared soon.

Maybe I missed them. 2700X not really changed much, but 2600 has dropped about 25% since I got mine around summer.

[Taf the Ghost]

20 minutes ago, porina said:

I was kind hoping for spring availability to match previous gens. Cash waiting!

 

Cinebench R15 is hardly affected by memory bandwidth, at least not on any system I've had my hands on. Makes it a good way to show off thread scaling and easy to predict scores if you know core count and clock, once you figure out the scaling value for a given architecture.

 

I'm still fearful for other uses, just how bad ram bandwidth limiting will be on a dual channel 16 core part... especially with the improved AVX performance on top.

 

At least aim for over 9000! Already at 7GHz ball park.

Rankings: http://hwbot.org/benchmark/cinebench_-_r15/rankings?hardwareTypeId=processor_5773&cores=8#start=0#interval=20

Current top: http://hwbot.org/submission/4000398_luumi_cinebench___r15_core_i9_9900k_3142_cb

 

Then again, they used water chiller last time, next logical step might be phase change cooling. Don't see that much these days.

Maybe I missed them. 2700X not really changed much, but 2600 has dropped about 25% since I got mine around summer.

Zen's cross-core memory management is amazing, but Epyc survives on 1 channel per 8 cores. However, it's after 6c/12t that a single channel starts to limit scaling. It's why at 4 cores, Single vs Dual channel means little, but at 6c it starts to be an issue. Plus, as AMD's argument has been since TR1 launched: if it's really an issue for you, they have a perfectly reasonably priced platform to cover that issue.

[Taf the Ghost]

@porina

 

Speaking of Channel Clearing, AMD ran some ridiculous discounts on Polaris GPUs during the holidays. RX570 4Gb in the 120-140USD range (+ games). That's the channel AMD is really trying to clear.

[rgb fan]

On 1/2/2019 at 11:03 AM, LukeSavenije said:

I've seen that before. those are old leaks, and i still don't believe them because they seem too good to be true imo.

liesssssssssss

[porina]

14 minutes ago, Taf the Ghost said:

Zen's cross-core memory management is amazing, but Epyc survives on 1 channel per 8 cores. However, it's after 6c/12t that a single channel starts to limit scaling. It's why at 4 cores, Single vs Dual channel means little, but at 6c it starts to be an issue. Plus, as AMD's argument has been since TR1 launched: if it's really an issue for you, they have a perfectly reasonably priced platform to cover that issue.

I haven't looked closely at existing Epyc since it isn't a surprise it doesn't suit the needs of most on this forum, myself included. I do recall seeing some supercomputing leaks for Rome, suggesting the clock is very low by consumer standards (~2GHz). That helps offset the demand significantly.

 

Of course, it then swings around to get the right CPU for the job. For consumer workloads, maybe ram bandwidth still isn't that important, but it hasn't really been put to the test. Suppose someone could get a TR and only use it with dual channel to see what the impact is as a preview. I don't recall details but when TR2 came out, multiple places were saying they felt the platform was limited by ram bandwidth. Those use cases may differ from consumer ones, but the line is getting blurred at what point does mainstream consumer end, and historic HEDT starts.

[Altruist]

2 hours ago, syn2112 said:

well if the rumors are true, then my 8700K is equiveleant to a 100$ "low end" Ryzen 3... well shit, it hasn't even been a year since i bought it, i barely got to enjoy the glory of a high end part. fml

...no

[Altruist]

2 hours ago, Mihle said:

I hope it's true. I will then upgrade from my 3570k, and get more than 3x the performance

....

What is this... you mean 3(cores)?

[TetraSky]

1 hour ago, Taf the Ghost said:

Noticed Ryzen 2nd Gen discounts were heavy during Christmas and even after. AMD probably sent out the message to get the channel cleared soon.

If anything, they raised the price right before the holidays and then "discounted" it back to somewhere near where it was before.
This is the 2600x. You can see the price being relatively stable through October and November, but then late November early December the price skyrocketed only to be bought down... to more than it was just a few weeks before. It's still more expensive today than it was in early November. (Nov 1 : 224.30$ ­| today : 259.99$, prices in CAD)

[Doobeedoo]

1 hour ago, Taf the Ghost said:

Not really surprising. AMD took some of the efficiency improvements and some of the clock improvements to get a total performance improvement approach. The GloFo 14nm silicon hits a voltage wall between 4.1 and 4.3 Ghz. That's why it could never overcome the clocks issue against Intel.

 

For Base Clocks, AMD hasn't really increased it much. 

 

https://www.newegg.com/Product/Product.aspx?Item=N82E16819113499

 

Current 2700X is 8 cores, 105w TDP at 3.7 Ghz base and 4.35 Ghz Boost. The new 3700X is 12c (50% more) with .5 Ghz faster base clock (13%). Given how the Voltage/Power issues were with 14nm, that's really not surprising. Going from 3.7 all-core to 4.1 all-core would functionally double the power draw on Ryzen 2nd gen. Even a minor shift in the Efficiency Curve would yield these results for AMD. And we know Epyc can do 100% more cores at roughly 10% higher clocks in the same power target range between 1st and 2nd gen.

Yeah true, both of each improvements, not fully one or the other way but good balance. The GloFo 14nm was a efficiency node thus the voltage and frequency wall. Also, Ryzen is/was still in 1./2. gen. too so. 

 

So I guess yeah after seeing more of 7nm Epyc it's looking really amazing actually. But the Ryzen 9 though, seeing such a chip on a AM4 platform and in such a socket, oof. 

[Taf the Ghost]

34 minutes ago, porina said:

I haven't looked closely at existing Epyc since it isn't a surprise it doesn't suit the needs of most on this forum, myself included. I do recall seeing some supercomputing leaks for Rome, suggesting the clock is very low by consumer standards (~2GHz). That helps offset the demand significantly.

 

Of course, it then swings around to get the right CPU for the job. For consumer workloads, maybe ram bandwidth still isn't that important, but it hasn't really been put to the test. Suppose someone could get a TR and only use it with dual channel to see what the impact is as a preview. I don't recall details but when TR2 came out, multiple places were saying they felt the platform was limited by ram bandwidth. Those use cases may differ from consumer ones, but the line is getting blurred at what point does mainstream consumer end, and historic HEDT starts.

The Rome installation as a super computer is running at 2.35 Ghz. Up from 2.00 Ghz with the few larger runs of Naples/Epyc 1. And at twice the core count. So that's still 100% more cores and ~18% faster at the same power budget. (That's deep within the process' efficiency curve, though.)

 

After reviewers went to Linux to test the 2990WX, it turns out the bandwidth scaling only dropped off over 24c/48t. So even 32 Cores with 2 leech dies on quad-channel really only stopped scaling properly after 6 cores per channel. But it still scales. So you're only hitting some max-power limitations if you can load all 32 threads on AM4. That's hard to do.

[.Apex.]

48 minutes ago, Altruist said:

...no

why no? IF the rumors are true then Zen 2 can reach 4.6ghz+ easily, and not to mention the increased IPC, which would put it next to the 8700K or even faster than it.

 

@VegetableStu

i mean we're talking about a few percentages here if the 3300X or 3400X doesn't reach 5ghz or something, even though you still have to keep in mind the increased IPC, that's still damn close enough considering it's a 100$-ish part 

[Taf the Ghost]

9 minutes ago, Doobeedoo said:

Yeah true, both of each improvements, not fully one or the other way but good balance. The GloFo 14nm was a efficiency node thus the voltage and frequency wall. Also, Ryzen is/was still in 1./2. gen. too so. 

 

So I guess yeah after seeing more of 7nm Epyc it's looking really amazing actually. But the Ryzen 9 though, seeing such a chip on a AM4 platform and in such a socket, oof. 

GloFo's 14nm somehow ended up one of the highest yielding nodes of all-time. Nothing like Intel's 22nm, which appears to be the best ever, but the yields have been amazing on the node. But it doesn't clock beyond 4.2 Ghz without having to ramp up massive voltage. Even with just that voltage ramp gone, AMD could normalize most of the clocks.

 

But, AMD will be on a better node than Intel, which is really why they're going to clean house until at least 2020 in Desktop.

[Taf the Ghost]

39 minutes ago, VegetableStu said:

not strictly true.... if the clocks of the 8700K is still higher than what the clocks of an R3 3300X would be, it'll still generally perform better in lightly-threaded programs or workloads.

 

that being said, those who bought their current CPU for the core count should be quivering at that R9 O_O

Intel will still have notable advantages due to all of the optimizations already for the Core Architecture. So, there is that.