AMD's Future Chips & SoC's: News, Info & Rumours.

[juanrga]

Either >40% better performance at same power OR >55% power reduction at same performance.

Or some combination of the two. I'd bet something like 30% Power Reduction & 15% performance gain.

Of course, I was only mentioning the two limiting cases.

I think engineers will chose a point with a power reduction higher than 30% because they will have to accommodate about 50% moar cores in a similar package, if the rumors of 6-core CCX are true.

 

[goldstone77]

I think engineers will chose a point with a power reduction higher than 30% because they will have to accommodate about 50% moar cores in a similar package, if the rumors of 12-core CCX are true.

12-core CCX?

 

[juanrga]

12-core CCX?

mistake corrected

 

[Gon Freecss]

Either >40% better performance at same power OR >55% power reduction at same performance.

Or some combination of the two. I'd bet something like 30% Power Reduction & 15% performance gain.

Of course, I was only mentioning the two limiting cases.

I think engineers will chose a point with a power reduction higher than 30% because they will have to accommodate about 50% moar cores in a similar package, if the rumors of 6-core CCX are true.

Will TSMC's process handle that? Cause in all likelihood, AMD will use TSMC's 7nm process with their Zen 2 chips.

 

[goldstone77]

Will TSMC's process handle that? Cause in all likelihood, AMD will use TSMC's 7nm process with their Zen 2 chips.

We don't know yet if that is going to happen, but if AMD wanted too I don't see why not. They have already been working with TSMC, and I'm sure they have a way of porting over the process fairly easily. GloFlo implemented Samsung 14nm LPP and plugging in AMD designs.

 

[aldaia]

And the above ExtremeTech quote don't mention 14LPP in any part.

Also: For those of you curious about 12nm and how it compares with 14nm, GF confirmed its 12nm is a refinement and improvement to an existing process with some optimized layouts that offer up to 10 percent performance improvement or a 15 percent density increase.

So could you then please explain what other 14nm process are they talking about considering that this was during a visit to a GloFo FAB?

 

[Gon Freecss]

Will TSMC's process handle that? Cause in all likelihood, AMD will use TSMC's 7nm process with their Zen 2 chips.

We don't know yet if that is going to happen, but if AMD wanted too I don't see why not. They have already been working with TSMC, and I'm sure they have a way of porting over the process fairly easily. GloFlo implemented Samsung 14nm LPP and plugging in AMD designs.

I mean, their process is inferior to GloFo, on paper.

 

[jdwii]

https://arstechnica.com/gadgets/2018/02/amd-sending-out-free-processors-to-solve-firmware-flashing-catch-22/

No one posted this yet but this gives us all evidence that AMD is committed in the AM4 platform and AMD really is trying to show that they are trying to offer longevity with the AM4 platform.

Again Amd is doing what they did in the past supporting their platform for years not just 2 years.

 

[gamerk316]

https://arstechnica.com/gadgets/2018/02/amd-sending-out-free-processors-to-solve-firmware-flashing-catch-22/

No one posted this yet but this gives us all evidence that AMD is committed in the AM4 platform and AMD really is trying to show that they are trying to offer longevity with the AM4 platform.

Again Amd is doing what they did in the past supporting their platform for years not just 2 years.

But the downside becomes that performance increases are limited by the platform.

I remember back in the early i7 days, when AMD users upgraded two or three times on the same platform, and never reached i7 level performance. Sure, they saved some money initially, but lost money over the long run on the second CPU upgrade.

 

[juanrga]

And the above ExtremeTech quote don't mention 14LPP in any part.

Also: For those of you curious about 12nm and how it compares with 14nm, GF confirmed its 12nm is a refinement and improvement to an existing process with some optimized layouts that offer up to 10 percent performance improvement or a 15 percent density increase.

So could you then please explain what other 14nm process are they talking about considering that this was during a visit to a GloFo FAB?

Glofo refers to Industry nodes, not to its own nodes. That is why they mention 16nm as baseline for the claims of density and power despite there is no 16nm node at Glofo. I don't know what 14nm is similar/equivalent to 16nm: 14LPE? 14LPC? 14LPU? UMC 14nm?. What I know is that there is no slide, official document, interview or any other where Glofo mentions its 14LPP node as baseline for density and power claims.

I guess why. You cannot publish a slide claiming up to 15% better density over TSMC 16nm, and then publish another slide claiming the same over 14LPP, because 14LPP is 9--15% more dense than TSMC 16nm.

So we take the official slide that says that 12LP is 15% more dense than TSMC 16nm and we obtain that 12LP may be about 0--6% more dense than 14LPP. This agrees with 12LP being only an extension of 14LPP, not a new node. This agrees with 12LP being what was formerly named 14nm+.

 

[juanrga]

https://arstechnica.com/gadgets/2018/02/amd-sending-out-free-processors-to-solve-firmware-flashing-catch-22/

No one posted this yet but this gives us all evidence that AMD is committed in the AM4 platform and AMD really is trying to show that they are trying to offer longevity with the AM4 platform.

Again Amd is doing what they did in the past supporting their platform for years not just 2 years.

But the downside becomes that performance increases are limited by the platform.

Reason why AMD and partners are releasing 400-series mobos as "optimized" for Pinnacle Ridge.

And if a new mobo is already needed to extract the maximum performance of the new Pinnacle Ridge chips, which are only a minor refresh of Summit Ridge chips. What can we expect for Zen2 and Zen3 chips?

 

[juanrga]

Either >40% better performance at same power OR >55% power reduction at same performance.

Or some combination of the two. I'd bet something like 30% Power Reduction & 15% performance gain.

Of course, I was only mentioning the two limiting cases.

I think engineers will chose a point with a power reduction higher than 30% because they will have to accommodate about 50% moar cores in a similar package, if the rumors of 6-core CCX are true.

Will TSMC's process handle that? Cause in all likelihood, AMD will use TSMC's 7nm process with their Zen 2 chips.

Where is this confirmed?

I guess AMD will use TSMC 7nm for GPUs and custom SoCs (consoles and so). Then Glofo 7nm for CPUs and APUs.

 

[-Fran-]

And if a new mobo is already needed to extract the maximum performance of the new Pinnacle Ridge chips, which are only a minor refresh of Summit Ridge chips. What can we expect for Zen2 and Zen3 chips?

Lol. Well, for one thing, better things than on the Intel side with forced upgrades for peanuts increases in performance?

You're really pushing the misleading boundaries there with that sentence. So, what is this "maximum performance" you're talking about vs an older platform? 10%? 20%? 100%?

You should find a job at the Sun or the Onion and write tech news there.

 

[Gon Freecss]

Where is this confirmed?

I guess AMD will use TSMC 7nm for GPUs and custom SoCs (consoles and so). Then Glofo 7nm for CPUs and APUs.

I read somewhere that GloFo's process won't be ready until the second half of 2020, and Lisa has been talking about using different 7nm processes for their products.

 

[juanrga]

I read somewhere that GloFo's process won't be ready until the second half of 2020, and Lisa has been talking about using different 7nm processes for their products.

Zen2 products are scheduled for late 2019 launch, but a possible delay to early 2020 wouldn't be unexpected considering Glofo historic record. Zen products were also originally scheduled for 2016 launch

and then delayed to 2017.

 

[juanrga]

And if a new mobo is already needed to extract the maximum performance of the new Pinnacle Ridge chips, which are only a minor refresh of Summit Ridge chips. What can we expect for Zen2 and Zen3 chips?

Lol. Well, for one thing, better things than on the Intel side with forced upgrades for peanuts increases in performance?

You're really pushing the misleading boundaries there with that sentence. So, what is this "maximum performance" you're talking about vs an older platform? 10%? 20%? 100%?

I am pretty sure the improvement will not be 1% neither 4%. Engineers wouldn't waste time designing new chipsets and new mobos for getting unnoticeable performance gains.

And my question remains. If AMD and motherboard engineers are introducing a new chipset for a tiny Zen refresh, what can we expect for Zen2 and Zen3 chips?

 

[8350rocks]

@juanrga:

You can expect that X370 will still work with Zen2 and Zen3...

 

[juanrga]

@juanrga:

You can expect that X370 will still work with Zen2 and Zen3...

That wasn't the point.

 

[8350rocks]

@juanrga: Regardless of the point, that is the answer 😉

 

[goldstone77]

AMD Ryzen 5 2400G Delidded: Solder vs. Thermal Paste vs. Liquid Metal
by Igor Wallossek February 19, 2018 at 6:00 AM
http://www.tomshardware.com/reviews/amd-ryzen-5-2400g-delidded,5475-2.html

Summary & Conclusion
Yes, replacing AMD's paste with liquid metal works well, facilitating improved thermal performance compared to a stock Ryzen 5 2400G. It's just too bad that you won't see much of a practical difference. There's not much added value, and AMD's processor is neither faster nor more power-friendly in the end.

The only measurable benefit is lower temperatures. Unfortunately the boxed cooler is not really able to take full advantage of lower temps without changes to its fan profile. Hours of work, notable risks, and minimal benefits just don't balance out. Thus, this modification is only worth considering if you're building a seriously compact PC.

Our findings may give you reason to complain about AMD's cost-cutting or praise the company for its execution. At a financial level, AMD made a reasonable decision. But when it comes to enthusiast mind share, shifting over to thermal paste is bound to raise eyebrows. On the other hand, are gamers buying eSports-ready PCs likely to argue about what's between their dies and heat spreaders?

So, no running room at the top end for these processor despite better cooling.

 

[Rogue Leader]

@juanrga:

You can expect that X370 will still work with Zen2 and Zen3...

@juanrga:

You can expect that X370 will still work with Zen2 and Zen3...

That wasn't the point.

@juanrga: Regardless of the point, that is the answer 😉

I'm still here, I don't go away..... tread lightly gentlemen.......

 

[goldstone77]

A side by side clock for clock comparison@3.4GHz. 1600X@3.4GHz 4,118 vs. 2600@3.4GHz 4,269 offering a clock for clock gain of 3.5%. 1600X@3.4GHz 18,906 vs. 2600@3.4GHz 20,102 offering a clock for clock gain of 5.9%. http://browser.geekbench.com/v4/cpu/6690641 https://browser.geekbench.com/v4/cpu/7110686

Noticeable improvements in memory latency:
1600X Memory Latency 5196 83.3 ns
2600 Memory Latency 4285 9.90 Moperations/sec
A difference of 17.5%. which amount to 68.7ns.

Spoiler

UPDATE:
I am wrong! Higher score relates to lower latency when compared to other Intel processors.
https://browser.geekbench.com/v4/cpu/7123767
i7-8700
single-core
Memory Latency
10134
42.7 ns
multi-core
Memory Latency
9981
43.4 ns

2600 has ~101ns latency for single core.
Also, note that the multi-core latency remained approximately the same.

 

[aldaia]

A side by side clock for clock comparison@3.4GHz. 1600X@3.4GHz 4,118 vs. 2600@3.4GHz 4,269 offering a clock for clock gain of 3.5%. 1600X@3.4GHz 18,906 vs. 2600@3.4GHz 20,102 offering a clock for clock gain of 5.9%. http://browser.geekbench.com/v4/cpu/6690641 https://browser.geekbench.com/v4/cpu/7110686

Add to that the (possible) increase on clock speed. It may be an engineering sample, but if it was a production version, the 2600 spotted is ~15% faster than 1600 in single core and ~30% faster in multicore.

 

[jdwii]

And if a new mobo is already needed to extract the maximum performance of the new Pinnacle Ridge chips, which are only a minor refresh of Summit Ridge chips. What can we expect for Zen2 and Zen3 chips?

Lol. Well, for one thing, better things than on the Intel side with forced upgrades for peanuts increases in performance?

You're really pushing the misleading boundaries there with that sentence. So, what is this "maximum performance" you're talking about vs an older platform? 10%? 20%? 100%?

I am pretty sure the improvement will not be 1% neither 4%. Engineers wouldn't waste time designing new chipsets and new mobos for getting unnoticeable performance gains.

And my question remains. If AMD and motherboard engineers are introducing a new chipset for a tiny Zen refresh, what can we expect for Zen2 and Zen3 chips?

We had silly refreshes for AM3 too. Not unheard of when it comes to Amd or Intel to be honest

 

[goldstone77]

A side by side clock for clock comparison@3.4GHz. 1600X@3.4GHz 4,118 vs. 2600@3.4GHz 4,269 offering a clock for clock gain of 3.5%. 1600X@3.4GHz 18,906 vs. 2600@3.4GHz 20,102 offering a clock for clock gain of 5.9%. http://browser.geekbench.com/v4/cpu/6690641 https://browser.geekbench.com/v4/cpu/7110686

Add to that the (possible) increase on clock speed. It may be an engineering sample, but if it was a production version, the 2600 spotted is ~15% faster than 1600 in single core and ~30% faster in multicore.

The difference between the 1600 and 1600X are just core clock speed, and comparing the 2600 to the 1600 with a 200MHz frequency deficit is kind of misleading. Similar to comparing Skylake with Kabeylake when most of the difference is higher base clock. The 1600 has a 3.2GHz base, while the 1600X has a 3.4Ghz base. The 2600 as spotted on geekbench.com is likely an engineering sample and full of anomalies. Look at the L1-L2 cache for example. I just attempt for a more accurate comparison clock for clock with the 1600X.

Edit:

2nd Edit: Z=QS sample