*** AMD "Zen" thread (inc AM4/APU discussion) ***

[iakhtar]

Saw an interview with an amd guy saying that 4000mhz is real easy on 4xx chipsets with the new ryzens, how much extra performance do you guys think 4000mhz+ mem kits will bring?

here's the link btw https://www.youtube.com/watch?v=CZOIOX-Ncr0

 

[humbug]

Saw an interview with an amd guy saying that 4000mhz is real easy on 4xx chipsets with the new ryzens, how much extra performance do you guys think 4000mhz+ mem kits will bring?

here's the link btw https://www.youtube.com/watch?v=CZOIOX-Ncr0

No idea, its impossible to predict.

Not much longer to wait, 3 weeks

 

[DragonQ]

Saw an interview with an amd guy saying that 4000mhz is real easy on 4xx chipsets with the new ryzens, how much extra performance do you guys think 4000mhz+ mem kits will bring?

here's the link btw https://www.youtube.com/watch?v=CZOIOX-Ncr0

Nice find. 4000 MHz being "easy" with Samsung B-die chips would certainly be a big improvement!

 

[B12B6]

https://www.3dmark.com/spy/3366892

2400mhz memory. It would seem Cpu tests 2700X vs 1800X in Time Spy is 6-7% better with 2700X.

 

[C-Roy]

Could it finally be time to upgrade my system

 

[daniel2K16]

Very tempted to sell my 1060 / 4790K and get Ryzen 2400g setup for 6 months while I wait for a 2070.

 

[David Bisset]

Higher IPC means Higher performance throughout all workloads.
It's why I find the emphasis of mentioning single core when mentioning focusing on IPC to be fairly wrong.

Sort of, except it's an attempt at making a workload-agnostic metric which doesn't apply universally - one CPU may have higher IPC in one task but lower in another one. This includes IPC not necessarily scaling similarly across cores on different systems.

That said, I agree with the sentiment that it's odd to only refer to single-core differences when talking about IPC mattering.

 

[InfiniteNoise]

My PC parts arrived, proceeded to build it and overclocked to 3.7GHz right away (Ryzen 5 1600), booted and ran flawlessly (well except for RX 550 issue under Linux, but other than that, all great).

I'll leave it at that for now.

 

[4K8KW10]

New Zen-based CPUs leaked:


https://wccftech.com/amd-desktop-zen-roadmap-2018-2020/

 

[DragonQ]

I am still curious why they use PGA for mainstream and LGA for HEDT.

 

[4K8KW10]

I am still curious why they use PGA for mainstream and LGA for HEDT.

Maybe concerns about how expensive these LGA chips are and to try to prevent possible pins damages?

 

[humbug]

AMD Ryzen 3000: Globalfoundries expects 5.0 GHz for the 7nm process

Well, let's do one more AMD related item that will make your eyebrows frown. Globalfoundries Chief Technical Officer, Gary Patton, recently talked about the upcoming 7nm manufacturing generation. The advantages compared to 14nm have grown significantly. The die sizes could be more than halved, clock rates in a range of 5.0 GHz are realistic.

This could be interesting for AMD's Zen 2, presumably in the form of Ryzen 3000, as reported on PC Games hardware Germany. Next month, you guys will see Pinnacle Ridge released, a Zen Refresh ("Zen +") within the Ryzen 2000 series, which will leverage the enhanced 12LP over the older (last years) 14LPP process. It gets more exciting with Zen 2, which will be manufactured at 7nm. Although AMD has not yet confirmed that they will use Globalfoundries (or TSMC) as partner, GlowFlo (ex AMD) probably will be the partner. Late February, AnandTech spoke to Globalfoundries' CTO, Gary Patton about the 7nm generation. The article has been largely unnoticed but contains some interesting details

Patton switched from IBM to Globalfoundries in 2015 when the manufacturing division was sold to the partner. Since then, the development department has made significant progress in the first 7nm process. Originally, one would at best assume a halving of the required chip area compared to 14nm. Improvements in the wiring, in particular, mean that the area savings should now be around a factor of 2.7. A hypothetical zeppelin-die (Ryzen 1000) at 7nm would therefore only about 80 instead of 213 mm² large. This all allows much potential for additional or larger CPU Cores. In addition, Globalfoundries expects a performance increase of 40 percent. Clock speeds in the range of 5.0 GHz seem quite realistic according to Patton, whereby it will finally arrive on the chip design - CPUs must be designed on such frequencies in order to be able to use the theoretical process possibilities. Time will tell, Ryzen 3000 going for 5 GHz? Bring it on.

 

[opethdisciple]

New Zen-based CPUs leaked:


https://wccftech.com/amd-desktop-zen-roadmap-2018-2020/

To me this resembles Intel's old Tick Tock life cycle.

 

[4K8KW10]

To me this resembles Intel's old Tick Tock life cycle.

Yes, but technically more difficult for execution. Because Intel shrank their chips. AMD will try to move to a new process with more noticeable changes in the architecture.

 

[Zartar117]

Higher IPC means Higher performance throughout all workloads.
It's why I find the emphasis of mentioning single core when mentioning focusing on IPC to be fairly wrong.

Not really, IPC generally means instructions per clock which is obviously related to instruction latency, though the way people generally (incorrectly) use it, it should be called IPS (instructions per second). I would view IPS as a mixture of instruction and cache latencies combined with the predict rate of the branch predictor since all this effects how many instructions will be executed in a given time frame. So it doesn't make sense to not place the emphasis on single core performance when referring to IPC (or IPS for that matter).

 

[nashathedog]

AMD Ryzen 2000-series lineup leaks ahead of April debut

https://www.techspot.com/news/73609-amd-ryzen-2000-series-lineup-leaks-ahead-april.html

Sorry if it's already been posted.

 

[eddiew]

2700X turbo is nice at 4.35 - giving me some hope that they'll be overclockable to 4.3 all-cores. On the other hand, it suggests they aren't viable at 4.4 because who on earth ships a processor with a 0.05ghz speed rating if it's not bouncing off its ceiling?

Also, isn't that 105w TDP higher than last gen? On what's meant to be a die shrink and power efficiency pass?

 

[CAT-THE-FIFTH]

2700X turbo is nice at 4.35 - giving me some hope that they'll be overclockable to 4.3 all-cores. On the other hand, it suggests they aren't viable at 4.4 because who on earth ships a processor with a 0.05ghz speed rating if it's not bouncing off its ceiling?

Also, isn't that 105w TDP higher than last gen? On what's meant to be a die shrink and power efficiency pass?

It could be because they are leakier cores which can clock higher??

Look at the lower TDP processors though - at 65W TDP there is a 400MHZ max increase and at 95W TDP a 250MHZ max increase. Remember,its only been a year since Ryzen was released,and this is not a bad increase. If you factor in there will be an IPC increase,and a decent decrease in L2 and L3 cache latencies,AMD at least is moving more quickly than with BD and K10 in year to year improvements!

 

[DragonQ]

If the 65 W and 95 W parts are both receiving a clock bump then the optimal efficiency point has increased. However, if the leaked/faked slides end up being true then clearly they are going far beyond the optimal efficiency point for the top-end 105 W TDP product - something they couldn't even do with the first generation Ryzens (Threadripper did though).

 

[humbug]

AMD Ryzen 2000-series lineup leaks ahead of April debut

https://www.techspot.com/news/73609-amd-ryzen-2000-series-lineup-leaks-ahead-april.html

Sorry if it's already been posted.

WCCF posted the same slides, they are fake

Photoshop and last years dates.

https://i.imgur.com/dFABPDI.jpg

The originals newly hosted. can you guys see these?

https://i.imgur.com/e0ov8aC.jpg



https://i.imgur.com/ZEYf0rF.jpg