RTG already received first Zen 2 Sample! (Rumour)

[Brooksie359]

2 hours ago, MMKing said:

You know you're not countering my point by stating that the 2700x can reach 4.2 with a relatively* easy OC. When my original point was that exceeding 4.1ghz all cores required manual overclocking.

 

 

I'm not particularly interested in how well Intel or AMD CPUs can overclock with expensive motherboards and cooling solutions. What matters is what guarantee AMD or Intel is willing to put on their product, with cooling NOT exceeding the rated TDP (Thermal Design Power). I'm very much against the press overclocking the product, given the premium components and cooling solutions they usually utilize as well as the issue of Intel and/or AMD cherry picking CPUs for the media.

 

As it stands. AMD guarantee that the 2700x will hit 3.7GHZ on all cores, with a rated TDP of 105W. It is expected that no matter how cheap the motherboard, as long as it is supported. And as long as the cooling is able to dissipate 105W, the CPU is guaranteed to hit 3.7GHZ. If AMD is willing to guarantee 5GHZ all cores on their Zen 2 CPUs, that is fantastic news and extremely impressive.

 

https://community.amd.com/community/gaming/blog/2017/11/27/asdasd

''Precision Boost allows an AMD Ryzen processor to dramatically increase clockspeeds when the CPU has electrical, thermal, and/or utilization headroom to spare.''

 

https://community.amd.com/community/gaming/blog/2018/08/13/understanding-precision-boost-overdrive-in-three-easy-steps

''Precision Boost Overdrive requires a 2nd Gen AMD Ryzen™ Threadripper processor with AMD X399 chipset motherboard. Because Precision Boost Overdrive enables operation of the processor outside of specifications and in excess of factory settings, use of the feature invalidates the AMD product warranty and may also void warranties offered by the system manufacturer or retailer.''

 

https://www.intel.com/content/www/us/en/products/docs/processors/core/8th-gen-core-family-datasheet-vol-1.html

''Intel Turbo Boost Technology 2.0 will increase the ratio of application power towards TDP and also allows to increase power above TDP as PL2 for short periods of time. Thus, thermal solutions and platform cooling that are designed to less than thermal design guidance might experience thermal and performance issues since more applications will tend to run at the maximum power limit for significant periods of time.''

 

 

I'm not swallowing marketing hogwash that try to pass off a 28 core running 5GHZ as normal product behavior, while they accidentally forget to mention certain features the end user may not have. The boost technologies are just differences in degree. AMD and Intel are not willing to guarantee that their products will perform at ''boost'' speeds, with ANY supported motherboards while being cooled using any cooler matching or exceeding the manufacturers rated TDP. The boost speeds represent the absolute maximum clock speed the CPU will clock itself to, provided the end user provides enough cooling, power and possibly an unreasonably expensive motherboard.

 

In the case of the 28 core farse. Intel used a 600USD cooler and a motherboard with 4 * 8 pin power connectors and what looks to be a 29 phase power design. Though the power phase design is more difficult to judge without a close inspection of the motherboard and the parts.

 

The 2700x is a 3.7GHZ 8 core 16 thread CPU, with a boost technology that MAY allow 4.3GHZ. Zen 2 8 core 16 thread CPU hitting 5GHZ would be a 35% improvement over the current flagship 8 core 16 thread. A 20% increase would be 4.44GHZ, and the rumored engineering sample of 4GHZ is a 10.8% improvement.

 

Hardware unboxed released a video a couple weeks ago, testing the 2990WX on 4 different motherboards.

 

Even using a closed loop water cooler. The Asrock Fatality X399 is NOT able to maintain a 4GHZ clock at 1.25v, bellow the AMD stated boost clock of 4.2GHZ. 

Are you even listening to your own argument? If we go by all core boost then the 8700k doesn't hit 5ghz either. Nobody expects these chips to hit 5ghz all core boost at stock because that would be insane. We are talking about once overclocked so it's only logical that overclocked values of the current cpu be used which is about 4.2 on average with 4.3 being possible if you get lucky. 

[Stefan Payne]

10 hours ago, MMKing said:

The Ryzen 2700x will NOT boost all cores to 4.3ghz on it's own. I am unfamiliar with the specifics, but without manual overclocking, all 8 cores will only boost to around 4 to 4.1GHZ. The 8700K will boost all of it's cores to 4.3GHZ (according to Intel).

Nobody was talking about all cores, just 5GHz in General.

And I don't really see an issue with that from a clocking perspective.

 

The Thermals are a whole different story, but that wasn't what we were talking about, was it?

 

And we already got ~10% from 14nm to 12nm, while that wasn't really a shrink and that was a low effort chip.

 

With Zen2 we get a redesign and also some serious modifications. So yeah, it is possible. But you'd want the 5GHz for only a couple of cores anyway...

[Doobeedoo]

Definitely an early ES though.

 

Give me 5GHz 8-core

[thorhammerz]

2 hours ago, Brooksie359 said:

Are you even listening to your own argument? If we go by all core boost then the 8700k doesn't hit 5ghz either. Nobody expects these chips to hit 5ghz all core boost at stock because that would be insane. We are talking about once overclocked so it's only logical that overclocked values of the current cpu be used which is about 4.2 on average with 4.3 being possible if you get lucky. 

Where exactly did he claim i7-8700k can do an all-core 5.0 GHz (without overclocking) ...? 

12 hours ago, MMKing said:

The Ryzen 2700x will NOT boost all cores to 4.3ghz on it's own. I am unfamiliar with the specifics, but without manual overclocking, all 8 cores will only boost to around 4 to 4.1GHZ. The 8700K will boost all of it's cores to 4.3GHZ (according to Intel).

 

[Brooksie359]

1 hour ago, thorhammerz said:

Where exactly did he claim i7-8700k can do an all-core 5.0 GHz (without overclocking) ...? 

Did you not read what I said? I will explain it in more detail to get my point across. Everyone is wanting these chips to hit 5ghz to be competitive with the 8700k and the like. That would mean we are talking about overclocking and not base clock boost for both chips. If all the new chips had to do was hit 4.3 on all core boost to be competitive with the 8700k then that would be easy. Nobody gets the 8700k and doesn't overclock unless they just want to waste money. So again let me reiterate my point. We are talking about overclocking not stock settings so anything else is pretty much irrelevant to this conversation. 

[thorhammerz]

4 hours ago, Brooksie359 said:

We are talking about overclocking not stock settings so anything else is pretty much irrelevant to this conversation. 

ORLY?

 

Seems like the other party was clearly talking about (and clearly delineated quite clearly in his post) about performance at stock settings (with a clear definition of said stock setting provided for your reading benefit). Perhaps request a rename of this topic to "only overclocking comments about Ryzen 2" before attempting to unilaterally claim this conversation to be about one specifc tangential sub-topic .

 

..

 

And chillax, whether or not Ryzen 2 performs at whatever performance benchmark you so emotionally expect it to is frankly independent of the sentiment expressed on the LTT forum - it will run/OC/stresstest/dance-at-a-circus at whatever jiggahertz it can run at, when it is released . 

[dalekphalm]

13 hours ago, VegetableStu said:

wait, Zen 2 and RTG? surely not Radeon Technologies Group? o_o

So I'm not the only one who read that and was like "WTF..."

 

I'm confused as to why RTG would be getting early Zen 2 samples? Maybe for working on the integrated GPU for upcoming Zen 2 APU's?

 

I mean, I guess the HardOCP "article" (if you can call 5 extremely short bullet points an Article) states it's for improvements to optimize the Radeon cores into Zen 2. The WCCFTech article does a better job of explaining this.

 

Apparently there's a new interconnect between the CPU and the GPU for Zen 2 APU's and they need to start working on integration this early already.

[Brooksie359]

1 hour ago, thorhammerz said:

ORLY?

 

Seems like the other party was clearly talking about (and clearly delineated quite clearly in his post) about performance at stock settings (with clear a clear definition of said stock setting provided for your reading benefit). Perhaps request a rename of this topic to "only overclocking comments about Ryzen 2" before attempting unilaterally claiming this conversation is about one specifc tangential sub-topic .

 

..

 

And chillax, whether or not Ryzen 2 performs at whatever performance benchmark you so emotionally expect it to is frankly independent of the sentiment expressed on the LTT forum - it will run/OC/stresstest/dance-at-a-circus at whatever jiggahertz it can run at, when it is released . 

Everyone was talking about it hitting 5ghz with the next iteration and were talking about overclocking obviously. This guy comes in saying there is no way that is true and starts talking about stock clocks. So yeah I can call the guy out for talking about something irrelevant to the conversation that was being had. Now if they wanted to simply state that Zen 2 is likely to have a stock clock much less than 5ghz then yeah they can have that conversation but they weren't. They were arguing against someone else's reasoning why Zen 2 should be able to achieve 5ghz with someone reasoning. 

[MMKing]

8 hours ago, GoldenLag said:

Whenna CPU doesnt meet the requirement of a higher end product it gets downgraded to the tier that suits it. 

 

The 2700 might qualify for XFR2, but i doesnt hold up to the standard of the 2700x and gets a different clock and voltage at boost.

I suggest you work on your linguistics a little bit.

 

You said

''Also the 2700x is actually guaranteed to hit X clockspeed at Y voltage hence the implementation of XFR2. Otherwise it would be an 2700.''

 

You're implying that the implementation of XFR2 (SenseMI) is somehow important when distinguishing the 2700 from the 2700x. Which i showed, it was not. Both the 2700 and the 2700x support XFR 2. Further more, none of your other sentences are backed up by any evidence. They are just claims thrown into the air, with the expectation (i assume) that other readers should find the information you're referring to, or worse, you assume the claims are self evident.

 

''7nm is aimed at 5ghz. Its density is much better and powerdraw is down.''

This would be an interesting article to read.

 

''Also im not talking 5ghz all core. Im just expecting them to hit somewhere near what the node is actually rated for.''

Once again, where is this information that TSMC 7nm is rated for 5ghz? or ''somewhere near'' it.

 

The only references to 5GHZ i can find on the 7nm node is from Gary Patton over at Global Foundries from a techpowerup article.

https://www.techpowerup.com/242148/globalfoundries-7-nm-to-enable-up-to-2-7x-smaller-dies-5-ghz-cpus

 

''Patton said that he expects this design to be able to scale pretty well to some 5 GHz operating frequencies.''

 

''The ability to scale up to 5 GHz frequencies will of course depend on the architecture's design being able to achieve that operating frequency stably, most of all.''

 

First of, these are words coming from the Chief Technical Officer at Global Foundries, not TSMC. Second, the wording offers no guarantee or certainty.

 

Actually, i found the direct answer on overclock3d.net with the question as well.

https://www.overclock3d.net/news/cpu_mainboard/globalfoundries_expects_great_things_from_7nm_-_clocks_in_the_5ghz_range/1

 

Q: Does the first generation of 7LP target higher frequency clocks than 14LPP?

 

Patton: ''Definitely. It is a big performance boost - we quoted around 40%. I don't know how that exactly will translate into frequency, but I would guess that it should be able to get up in the 5GHz range, I would expect''

 

''I don't know''

''guess''

''I would expect''

 

5GHZ is not promised, 5GHZ is not stated to be what the node is rated for and once again. As far as i am aware, TSMC has not revealed or even hinted at how well they expect their 7nm node to perform once it's released with a written guarantee on the box. The only certain answer given (Patton interview) is that the 7nm node will perform at the very least better than the 14nm node, which could mean 100mhz higher, 1ghz higher or even 10ghz higher. I hope you now understand the importance of fact checking, and how an interview remembered from months ago needs to be revisited to correct your memory.

[GoldenLag]

39 minutes ago, MMKing said:

I suggest you work on your linguistics a little bit.

phone typing + secondary language often doesnt add up well............. also im not working on my linguistics any more due to the lack of academic use i will have for it in the coming years.

41 minutes ago, MMKing said:

You said

''Also the 2700x is actually guaranteed to hit X clockspeed at Y voltage hence the implementation of XFR2. Otherwise it would be an 2700.''

yes the wording is a bit off. though the 2700x is guaranteed to hit X clockspeed at Y voltage. i should have mentioned that otherwise it wouldnt hit the XFR2 requirements for being an 2700x. therefore making it an 2700 which have different clockspeed targets and probably voltage targets.

43 minutes ago, MMKing said:

''7nm is aimed at 5ghz. Its density is much better and powerdraw is down.''

This would be an interesting article to read.

i dont have an article to quote, but i seem to remember a 40ish percentile decrease in powerdraw compared to X previuos node. though i dont recall if this was 14nm or 12nm or if it was TSMC or Glofo. 

45 minutes ago, MMKing said:

'Also im not talking 5ghz all core. Im just expecting them to hit somewhere near what the node is actually rated for.''

Once again, where is this information that TSMC 7nm is rated for 5ghz? or ''somewhere near'' it.

 

The only references to 5GHZ i can find on the 7nm node is from Gary Patton over at Global Foundries from a techpowerup article.

https://www.techpowerup.com/242148/globalfoundries-7-nm-to-enable-up-to-2-7x-smaller-dies-5-ghz-cpus

pretty much where i take it from yes. since that is Glofo speaking. a relativly honest company coming to improvements in node design, the number is probably not far off. also did i say it was rated for 5ghz? i choose my words somewhat carfully as it is "aimed" at 5ghz. There is also the general consensis that at the time 5ghz node of glofo was inferior  or behind TSMC in developments. also props for you for digging as i really often dont and do not intend to cite my sources as i never remember where i take things from. 

 

50 minutes ago, MMKing said:

''Patton said that he expects this design to be able to scale pretty well to some 5 GHz operating frequencies.''

 

''The ability to scale up to 5 GHz frequencies will of course depend on the architecture's design being able to achieve that operating frequency stably, most of all.''

hence my mention of architectural widening in my post as that can lower clocks significantly. just take a look at the fx series. they clock really high, but their IPC sucks. we allready know Zen clocks relativly well given its current node stance.

 

now using TSMC`s own numbers.  the 7nm node is about 38% faster than their 16nm node used on pascal. now we dont have any good comparison numbers, but ill let you draw your own conclusion from that. (if you wanna draw any at all as this is the Companieses own numbers afterall) .

 

1 hour ago, MMKing said:

I hope you now understand the importance of fact checking, and how an interview remembered from months ago needs to be revisited to correct your memory.

im very well aware of the importance of fact checkin. i dont like to mention things im uncertain of without either declaring i am uncertain of something or not. also industry people dont like to make any promises to the press, as if it is broken the whole company takes a hit. which more or less outlines any wording you see in interviews. now as much as i like to revisit old articles i dont have the time to do that. im not gonna claim to be some fact god and feel free to correct me. 

 

 

so for some real comparison numbers. using the A9 bionic chip an the 14nm samsung and 16nm as a baseline for this.

since the A9 was dualsourced from samsung and TSMC we have a product we can compare to. 

 

since the 16nm does have a couple of iterations on it, one being the pascal version for higher yields its not the best comparison, but its the best we have.

 

both A9 chips had a perfromance difference between them that was negligable (2-3%) . which is more or less ignorable.

 

since glofo liscensed the 14nm node from samsung to be able to produce 14nm chips. they made some iterations on the node design. though i dont have anything on what these iterations were. they appear to be Voltage oriented, but honestly ive got nothing. 

 

from this we can make a terribly innaccurate estimate of what 7nm should look like. since 14nm glofo is somewhat equal to 16nm TSMC and that 7nm is roughly judging by TSMC`s own numbers about 38% faster.  using the 1800x base clock you get to about 4,968ghz. now the question is that 5ghz all that unreasonable?

 

i would also just like to say everything below  the first mention of the A9 bionic chips is purely speculation. id just like to share my thoughts

[leadeater]

1 hour ago, MMKing said:

Which i showed, it was not. Both the 2700 and the 2700x support XFR 2.

Both have XFR but the X models have twice the XFR clock increase, there is a difference but not all that important.

[cj09beira]

3 hours ago, dalekphalm said:

So I'm not the only one who read that and was like "WTF..."

 

I'm confused as to why RTG would be getting early Zen 2 samples? Maybe for working on the integrated GPU for upcoming Zen 2 APU's?

 

I mean, I guess the HardOCP "article" (if you can call 5 extremely short bullet points an Article) states it's for improvements to optimize the Radeon cores into Zen 2. The WCCFTech article does a better job of explaining this.

 

Apparently there's a new interconnect between the CPU and the GPU for Zen 2 APU's and they need to start working on integration this early already.

super simple really, RTG will be launching a pcie 4.0 gpu very soon, with xgmi support, and guess what is the only cpu that will support both ?

[MMKing]

@GoldenLag

You are indeed correct that the estimate is terribly inaccurate.

 

TSMC 10nm

http://www.tsmc.com/english/dedicatedFoundry/technology/10nm.htm

''With a more aggressive geometric shrinkage, this process offers 2X logic density than its 16nm predecessor, along with ~15% faster speed and ~35% less power consumption.''

 

TSMC 7nm

http://www.tsmc.com/english/dedicatedFoundry/technology/7nm.htm

''Compared to its 10nm FinFET process, TSMC’s 7nm FinFET features 1.6X logic density, ~20% speed improvement, and ~40% power reduction.''

 

TSMC makes no direct performance, power consumption or clock speed reference between the 7nm and the 16/12nm process directly.

speed improvement = 100% (16nm) * 1.15 (10nm) * 1.20 (7nm) = 38%

 

So yes, the ''speed improvement'' is pretty much spot on what you cited in your last post. However, speed improvement must not be interchanged with clock speed.

[Tam3n]

I hope it will be good, finally an upgrade from Sandy Bridge? Daily OC clock speed improvements have been pretty minor since 32 nm SB, though performance has slowly improved. I think the difference between Sandy Bridge and Coffee Lake is something like "up to" 30%... Though maintream Coffee Lake still doesn't have even half of the number of pcie lines of SB-E. The new AMD offering could be attractive if the performance is there.

[BuckGup]

On 9/30/2018 at 3:42 PM, TetraSky said:

Was hoping AMD would increase the core count a little more for Zen 2, like, 12 core 24 threads.
But hey, this is just a rumour and may very well not be true for all we know.

Lol only about 3 games scale across more than 4 cores though. I would rather have higher clock speeds across all cores rather than slower with more cores

[Stefan Payne]

On 10/1/2018 at 8:43 PM, dalekphalm said:

I'm confused as to why RTG would be getting early Zen 2 samples? Maybe for working on the integrated GPU for upcoming Zen 2 APU's?

Driver Optimization and playing around with Infinite Fabric, maybe PCIe 4.0.

 

That would be my guesses...

[jmm4]

Maybe this is a R3 or R5 ES. 3400G Perhaps?

[xriqn]

On 9/30/2018 at 10:24 PM, GoldenLag said:

AMD needs to up the corecount. Or at least make all chips SMT enabled. AMD doesnt neccesarly have to up the corecount (i hope they honestly do), but they really need to smash intel at every market tier. 

 

7nm is the new hype and they need to make the bloody most of it. My hope is they invade the mobile scene with fully 7nm APUs. 

You also have the "fine wine" argument. Older GPUs from AMD have really aged well. Also freesync is just a killer deal for people on a "budget". Vega 64 + 1440p 144hz freesync for 1600 isnt bad (including monitor, keyboard, headset and mouse(all good quality and mechanical))

True, the entire Radeon HD series (from like HD 6xxx) still holds up really well in modern titles.

[cj09beira]

On 10/1/2018 at 10:17 PM, GoldenLag said:

 

pretty much where i take it from yes. since that is Glofo speaking. a relativly honest company coming to improvements in node design, the number is probably not far off. also did i say it was rated for 5ghz? i choose my words somewhat carfully as it is "aimed" at 5ghz. There is also the general consensis that at the time 5ghz node of glofo was inferior  or behind TSMC in developments. also props for you for digging as i really often dont and do not intend to cite my sources as i never remember where i take things from. 

 

pretty sure global foundries node would be better if we compare hpc vs hpc after all global foundries hpc had cobalt for some parts too which would reduce resistance quite a bit, plus the part there that node was being made to allow ibm's next gen 5ghz++ chips.

 

btw do we know how ibm will manage after all they have some pretty high requirements node wise that i have my thoughts tsmc can provide, someone pointed that they might only use their research fab  to do that but does it have enough capacity? and wont it slow down R&D

[Taf the Ghost]

18 minutes ago, cj09beira said:

pretty sure global foundries node would be better if we compare hpc vs hpc after all global foundries hpc had cobalt for some parts too which would reduce resistance quite a bit, plus the part there that node was being made to allow ibm's next gen 5ghz++ chips.

 

btw do we know how ibm will manage after all they have some pretty high requirements node wise that i have my thoughts tsmc can provide, someone pointed that they might only use their research fab  to do that but does it have enough capacity? and wont it slow down R&D

Haven't seen any reports on IBM's plans, but, at least for 7nm, the research node might be able to supply IBM's needs. I imagine they are exploring many options beyond that.

3 hours ago, Stefan Payne said:

Driver Optimization and playing around with Infinite Fabric, maybe PCIe 4.0.

 

That would be my guesses...

Gaming Workloads. We've found CS:GO to be extremely similar to many abusive Cache Heavy workloads. RTG is the best in the company for stress-testing real world applications in weird ways.

[leadeater]

46 minutes ago, Taf the Ghost said:

Haven't seen any reports on IBM's plans, but, at least for 7nm, the research node might be able to supply IBM's needs. I imagine they are exploring many options beyond that.

Power9 is still new for IBM and they don't actually node shift that quickly, it's likely they are fine with what they have (14nm) for the next few years. Power10 which is targeted for 2020 doesn't actually have a fixed node target yet, just a goal of 10nm. My guess is Power10 will be slightly delayed and on 7nm.

[Taf the Ghost]

1 hour ago, leadeater said:

Power9 is still new for IBM and they don't actually node shift that quickly, it's likely they are fine with what they have (14nm) for the next few years. Power10 which is targeted for 2020 doesn't actually have a fixed node target yet, just a goal of 10nm. My guess is Power10 will be slightly delayed and on 7nm.

IBM will probably pay for a semi-custom node from either Samsung or TSMC, would be my guess.

[leadeater]

1 hour ago, Taf the Ghost said:

IBM will probably pay for a semi-custom node from either Samsung or TSMC, would be my guess.

I would guess TSMC, not sure IBM would want to give business to Samsung. No reason behind that thinking other than not wanting to make Samsung any bigger than they already are.