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

[goldstone77]

AMD preparing mobile Ryzen 5 APU with Vega 11 graphics
Published: 14th Dec 2017, 08:52 GMT

Mobile: AMD Vega 11
This is the first time Vega 11 makes an appearance in a leaked slide. A new mobile Ryzen series with higher TDP will receive more powerful integrated Vega graphics. The Ryzen 5 2400G, which is a quad-core and 8-thread APU, will feature Vega 11 Graphics. The whole package will be available in 65 and 35W variants. Another ‘confirmed’ part is Ryzen 3 2200G which is allegedly featuring Vega 8 Graphics while being a 4-thread APU.

The U-series will also receive a new SKU. The Ryzen 3 2300U will be the slowest mobile APU featuring 6 Compute Units designed in Vega architecture.

Spoiler

Desktop: AMD Raven Ridge
Meanwhile desktop series are also making an appearance. Raven Ridge APU featuring 28 CUs (1792 Stream Processors) has been spotted along alongside Fenghuang Raven platform (engineering board).

What Sisoft information is lacking, but we can already share with you, is the core count of this new APU. It’s a quad-core, 8-thread chip with a codename of “2G11SRD1P4MF6_30_N”.

 

[adamsleath]

interesting info.

 

[aldaia]

Thanks goldstone, lots of interesting info in those figures.

AMD will use the 7nm SOC. 40% higher frequency means that potentially AMD can produce chips with 5GHz base and 5.6 Ghz turbo in the same power envelope (1800X is 3.6-4GHz, 4.1 XFR). Though, probably, the extra power room will be used for extra cores and therefore base freq will be much lower, but single core turbos around 5.6 GHz are still possible (slightly higher if something like XFR is used). At the same time we know intel's 10nm achieves lower freq than intels 14nm++, and 10nm+ is on par with 14 nm++ (intels 10nm++ is not expected until late 2019 or early 2020). If both AMD and Glofo are on track, 2019 will be very interestig.

IBM will use 7nmHPC, the figure sugests 6-7 GHz is possible. At the price of using way more power, but historically that seems to be fine for IBM. For instance IBM Z14 is around 300-400W, systems built around z14 are water cooled.

That relative size comparisson gives also interesting info. We already know GloFo HD SRAM cell is 0.0269 µm² the figure sugests that High Performance cell is ~0.036 µm² (give or take 0.001 µm²). A signifficant density advantage over intel 10 nm process:
Intel HP 0.0441 µm²
Intel HD 0.0312 µm²

 

[juanrga]

Spoiler

Adding 50% moar cores and ~1GHz would increase the limit to around 350W.

This is what took you 5 minutes and is the basis for your argument for the slide being fake... .. . . . Care to explain? AM4 has a 140W+ TDP and the 1800X is a 8 core with base clock 3.6GHz and TDP of 95W.

That is only a part of my argument. The other part is where I mention the physical impossibility of 12LP reducing this wattage by a factor of ~3x to fit within the AM4 socket spec limits.

Per spec. the "AMD socket AM4 reference limit" is 128W.

140W is the rating for some coolers.

The '95W' of the 1800X is a marketing value that doesn't correspond to the real TDP. The real TDP is higher as confirmed by reviews.

The thermal design power (TDP), sometimes called thermal design point, is the maximum amount of heat generated by a computer chip or component (often the CPU or GPU) that the cooling system in a computer is designed to dissipate under any workload.

Power consumption is not TDP. Please list the reviews stating that thermal design power was not 95W. This should be fun since there is not set standard for TDP, and it depends entirely on what the manufacturer decides constitutes a 95WTDP workload.

Power consumption is TDP. The first law of thermo states all the heat dissipated by the CPU comes from the power consumed by the socket.

Virtually any review of the 1800X reported how the real TDPs aren't those that AMD marketing claims. There is a de-facto standard for the definition of TDP and AMD did broke it, inventing a false TDP formula for RyZen to mislead people. People looking at TDP marketing value would believe that 95W RyZen is more efficient than the 140W of Broadwell-E, but in reality the '95W' dissipates about same power (~130W):

In fact, despite the 95 watt TDP, the Ryzen CPU uses about the same power as the 140 watt Broadwell-E processors.

The more complete discussion is on the Hardware.fr review:

TDP and TDP ...

Indeed with a consumption measured on the ATX12 at 128.9 watts, it is obvious that the consumption of the Ryzen 7 1800X exceeds the 95 watts announced on TDP (Thermal Design Power). Indeed even if we base on a yield of 85% in the power stage of the motherboard, we arrive at almost 110 watts. An estimate confirmed by the internal monitoring of the processor which even indicates 112 watts under x264.

At Intel, the consumption limits for the Turbo and the TDP are identical, which seems the most logical since each watts consumed by the processor is discharged in the form of heat. In very rare cases, which do not correspond to a realistic load, in particular 100% AVX load, consumption can exceed this common value even at the initial frequency. For Ryzen, AMD uses another formula:

TDP (Watts) = (tCase ° C - tAmbient ° C) / (HSF ° C / W)

The three values ​​on the right are defined in this way:

tCase ° C: maximum temperature at the junction between the die and the HIS necessary to maintain the expected level of performance.
tAmbient ° C: maximum ambient temperature of the housing needed to maintain the expected level of performance.
HSF ° C / W: Minimum thermal resistance of the cooler to maintain the expected level of performance.

For the 1800X and 1700X, these values ​​are respectively 60 ° C, 42 ° C and 0.189. For the 1700, it is 72.3 ° C, 42 ° C and 0.451. The formula gives respectively 95.23W and 64.96W.

If linking all these values ​​is logical when defining thermal specifications, AMD benefits our sense of the lack of a standard for defining what a component's TDP should be. The result of this formula is not a TDP in the usual sense of the term but the number of watts of the processor that must be dissipated (and therefore it can consume) to maintain its maximum performance under certain conditions.

What are then the TDP, in the sense of the limit of consumption and thus the maximum number of watts to be dissipated, Ryzen? AMD also communicates this value, less markedly: 128 watts for the 1800X / 1700X, and 90 watts for the 1700. These are the values ​​that are the most comparable with the TDP released by Intel.

The real TDP of the 1800X/1700X is 128W. The real TDP of the 1700 is 90W.

HFR confirmed via measurement the 128W on the 1800X. Canard confirmed the 90W for the 1700:

https://mobile.twitter.com/CPCHardware/status/843109717610287105/actions

The 1700 pulls 90W in reality. AMD bullshit its TDP.

Can you give link to the discussion on Hardware.fr? Twitter link did not work. Also, by their own admission the commenter realizes their is not standard for TDP, and is making a guess based on assumptions and making it up as he goes and omits AVX workloads from Intel chips, because he just doesn't think it's realistic. So, he is picking and choosing how he wants to come up with these numbers. How can I take anything based on his opinions seriously?

If linking all these values ​​is logical when defining thermal specifications, AMD benefits our sense of the lack of a standard for defining what a component's TDP should be.

Since safety margins and the definition of what constitutes a real application vary among manufacturers, TDP values between different manufacturers cannot be accurately compared. For example, while a processor with a TDP of 100 W will almost certainly use more power at full load than a processor with a 10 W TDP from the same manufacturer, it may or may not use more power than a processor from a different manufacturer that has a 90 W TDP. Additionally, TDPs are often specified for families of processors, with the low-end models usually using significantly less power than those at the high end of the family.

The dynamic power consumed by a switching circuit is approximately proportional to the square of the voltage:[7]

where C is capacitance, f is frequency, and V is voltage.\

Edit: Have you ever used an Intel heatsink and fan? They don't come close to the TDP of the processor.

The twitter link is to CanardPC. The HFR review can be easily found by googling "hardware.fr 1800X".

The reviewer claims there is no standard in the sense of ISO standard or JEDEC standard or similar comittee-defined standard. But he mentions there is a de-facto standard in the industry, which is ignored by the formula invented by AMD for Zen: "The result of this formula is not a TDP in the usual sense of the term".

AMD was using this de-facto standard before Zen. For instance, the 125W Piledrivers are 125W CPUs.

The reviewer is not omitting AVX workloads from Intel chips. He only avoids "100% AVX loads", which never happen in real-life: "do not correspond to a realistic load,". Those unrealistic loads only happen when running certain stress tests. Recall that AVX is an extension to run over a x86-code baseline. It is not a separate ISA that runs alone for 100% of time.

In real cases Intel chips satisfy the marketing TDP values, as confirmed by the reviewer: "At Intel, the consumption limits for the Turbo and the TDP are identical, which seems the most logical since each watts consumed by the processor is discharged in the form of heat. "

This is not AMD vs Intel thing. As mentioned above AMD CPUs previous to RyZen satisfy the marketing TDP. ThreadRipper CPUs also satisfy the marketing TDP (1). Lower RyZen models also satisfy the marketing TDP. The only CPUs that violate the marketing value are the 1800X/1700X and the 1700 models.

(1) Albeit AMD forces the limit to the marketing TDP by deceiving about clocks.

From tom's hardware:

"And it's notable that the 95W Ryzen 7 1700X we're reviewing today uses less power under our stress test than Intel's 91W Core i7-7700K."

They used a power virus only for the Intel chip, not a power virus for RyZen. The AVX-256 capable stress test is able to load fully the pair of powerful 256bit SIMD units on Kabylake. Units are absent in RyZen, which has only a pair of 128bit FMACs. In fact RyZen consumes about the same power on this 'stress' test than on a ordinary workload as Luxrender. RyZen consumes 112W on both. and only 16W than on Blender. Evidenlty Luxrender is not a stress test.

Any efficiency metric using comparable workloads shows that Kabylake consumes less and is more efficient than 1800X/1700X.

 

[juanrga]

Thanks goldstone, lots of interesting info in those figures.

AMD will use the 7nm SOC. 40% higher frequency means that potentially AMD can produce chips with 5GHz base and 5.6 Ghz turbo in the same power envelope (1800X is 3.6-4GHz, 4.1 XFR).

That is at normalized frequency, and clearly in the low range. No way we will those 5.0/5.6GHz.

IBM will use 7nmHPC, the figure sugests 6-7 GHz is possible.

This is pure nonsense. The performance version of 7LP is designed for 4GHz and IBM is targeting those clocks.

 

[juanrga]

Here's another "good read" from Mr Morgan over at the next platform:

"The important thing to note here is that these are not proof of concept deployments. They are not tens or even hundreds of machines used in a proof of concept, but thousands of machines that have been deployed already in 2017 at Microsoft and Baidu, and with every prospect of this growing to tens of thousands of machines in 2018 as more workloads come into play at these companies."

Check it out here: https://www.nextplatform.com/2017/12/13/two-hyperscalers-amds-epyc-six-go/

Quoted from the article: "Microsoft could have launched a much-improved L Series instance using Skylake Xeons and added in twice as much memory and 2.5X the storage. But the point is, Microsoft did not do that, and AMD got the business instead."

There may be many reasons for that decision, not least being vendor diversification.

I've said before an I'll say it again everybody wants AMD to succeed !

An Some poeple claimed it would never happen, design wins for Epyc. An massive ones as well... I gues they couldn't of been more wrong about that.

Microsoft is offering EPYC as another option to customers. As mentioned in the article Microsoft is offering four different chips from four different companies. So being one of four is not anything special. What one has to check is what chip has selected Microsoft for its own datacenters, and which is the sales volume for each of the four options offered in Azure.

The first question is easy to answer. Microsoft doesn't chose EPYC

https://semiaccurate.com/2017/03/08/qualcomm-microsoft-team-arm-servers/
https://semiaccurate.com/2017/03/13/second-megadatacenter-goes-heavily-arm-cpus/

The second question is more difficult to answer, because AMD refuses to give detailed sales for EPYC. The more close statement was this vague remark:

John W. Pitzer - Credit Suisse Securities (USA) LLC

And then, Lisa, just as my follow-on, Kind of a similar question around EPYC, and just I know you guys have kind of talked about potentially exiting the end of next year at about a 2% share. I'm just kind of curious how you think about the ramp of EPYC. And if you can differentiate between sort of your hyperscale customers who might take it a little bit sooner versus the more traditional OEM channel, how should we be thinking about that on EPYC?

Dr. Lisa T. Su - Advanced Micro Devices, Inc.

Yeah. So, John, I'm not sure we ever said 2% by the end of next year. But what I would say is that we expect EPYC to be a sizable portion of our revenue in 2018. So the second half this year, we're doing early pilots and we're doing some early deployments. The hyperscale guys are aggressive and they are first. We will start seeing some enterprise revenue here in Q4 as the early platforms launch. But I expect more of the enterprise to fill in as we go into 2018. And I think the important point is as we look at the product, as we look at the competition and where we're positioned, the product positioning is strong. And so the customer engagements are growing. And we're seeing significant interest from enterprise customers ramp now as some of the OEM platforms are becoming available and starting their seating. So overall, EPYC will continue to ramp into 2018 and should be a sizable -- we expect it to be a sizable portion of our revenue in 2018.

So AMD cannot even promise a 2% of market share by end of 2018. This gives an idea of how little amounts of EPYCs AMD is selling.

 

[goldstone77]

Spoiler

Spoiler

Adding 50% moar cores and ~1GHz would increase the limit to around 350W.

Spoiler

This is what took you 5 minutes and is the basis for your argument for the slide being fake... .. . . . Care to explain? AM4 has a 140W+ TDP and the 1800X is a 8 core with base clock 3.6GHz and TDP of 95W.

Spoiler

That is only a part of my argument. The other part is where I mention the physical impossibility of 12LP reducing this wattage by a factor of ~3x to fit within the AM4 socket spec limits.

Per spec. the "AMD socket AM4 reference limit" is 128W.

140W is the rating for some coolers.

The '95W' of the 1800X is a marketing value that doesn't correspond to the real TDP. The real TDP is higher as confirmed by reviews.

Spoiler

The thermal design power (TDP), sometimes called thermal design point, is the maximum amount of heat generated by a computer chip or component (often the CPU or GPU) that the cooling system in a computer is designed to dissipate under any workload.

Power consumption is not TDP. Please list the reviews stating that thermal design power was not 95W. This should be fun since there is not set standard for TDP, and it depends entirely on what the manufacturer decides constitutes a 95WTDP workload.

Spoiler

Power consumption is TDP. The first law of thermo states all the heat dissipated by the CPU comes from the power consumed by the socket.

Virtually any review of the 1800X reported how the real TDPs aren't those that AMD marketing claims. There is a de-facto standard for the definition of TDP and AMD did broke it, inventing a false TDP formula for RyZen to mislead people. People looking at TDP marketing value would believe that 95W RyZen is more efficient than the 140W of Broadwell-E, but in reality the '95W' dissipates about same power (~130W):

In fact, despite the 95 watt TDP, the Ryzen CPU uses about the same power as the 140 watt Broadwell-E processors.

The more complete discussion is on the Hardware.fr review:

TDP and TDP ...

Indeed with a consumption measured on the ATX12 at 128.9 watts, it is obvious that the consumption of the Ryzen 7 1800X exceeds the 95 watts announced on TDP (Thermal Design Power). Indeed even if we base on a yield of 85% in the power stage of the motherboard, we arrive at almost 110 watts. An estimate confirmed by the internal monitoring of the processor which even indicates 112 watts under x264.

At Intel, the consumption limits for the Turbo and the TDP are identical, which seems the most logical since each watts consumed by the processor is discharged in the form of heat. In very rare cases, which do not correspond to a realistic load, in particular 100% AVX load, consumption can exceed this common value even at the initial frequency. For Ryzen, AMD uses another formula:

TDP (Watts) = (tCase ° C - tAmbient ° C) / (HSF ° C / W)

The three values ​​on the right are defined in this way:

tCase ° C: maximum temperature at the junction between the die and the HIS necessary to maintain the expected level of performance.
tAmbient ° C: maximum ambient temperature of the housing needed to maintain the expected level of performance.
HSF ° C / W: Minimum thermal resistance of the cooler to maintain the expected level of performance.

For the 1800X and 1700X, these values ​​are respectively 60 ° C, 42 ° C and 0.189. For the 1700, it is 72.3 ° C, 42 ° C and 0.451. The formula gives respectively 95.23W and 64.96W.

If linking all these values ​​is logical when defining thermal specifications, AMD benefits our sense of the lack of a standard for defining what a component's TDP should be. The result of this formula is not a TDP in the usual sense of the term but the number of watts of the processor that must be dissipated (and therefore it can consume) to maintain its maximum performance under certain conditions.

What are then the TDP, in the sense of the limit of consumption and thus the maximum number of watts to be dissipated, Ryzen? AMD also communicates this value, less markedly: 128 watts for the 1800X / 1700X, and 90 watts for the 1700. These are the values ​​that are the most comparable with the TDP released by Intel.

The real TDP of the 1800X/1700X is 128W. The real TDP of the 1700 is 90W.

HFR confirmed via measurement the 128W on the 1800X. Canard confirmed the 90W for the 1700:

https://mobile.twitter.com/CPCHardware/status/843109717610287105/actions

The 1700 pulls 90W in reality. AMD bullshit its TDP.

Can you give link to the discussion on Hardware.fr? Twitter link did not work. Also, by their own admission the commenter realizes their is not standard for TDP, and is making a guess based on assumptions and making it up as he goes and omits AVX workloads from Intel chips, because he just doesn't think it's realistic. So, he is picking and choosing how he wants to come up with these numbers. How can I take anything based on his opinions seriously?

If linking all these values ​​is logical when defining thermal specifications, AMD benefits our sense of the lack of a standard for defining what a component's TDP should be.

Since safety margins and the definition of what constitutes a real application vary among manufacturers, TDP values between different manufacturers cannot be accurately compared. For example, while a processor with a TDP of 100 W will almost certainly use more power at full load than a processor with a 10 W TDP from the same manufacturer, it may or may not use more power than a processor from a different manufacturer that has a 90 W TDP. Additionally, TDPs are often specified for families of processors, with the low-end models usually using significantly less power than those at the high end of the family.

The dynamic power consumed by a switching circuit is approximately proportional to the square of the voltage:[7]

where C is capacitance, f is frequency, and V is voltage.\

Edit: Have you ever used an Intel heatsink and fan? They don't come close to the TDP of the processor.

Spoiler

The twitter link is to CanardPC. The HFR review can be easily found by googling "hardware.fr 1800X".

The reviewer claims there is no standard in the sense of ISO standard or JEDEC standard or similar comittee-defined standard. But he mentions there is a de-facto standard in the industry, which is ignored by the formula invented by AMD for Zen: "The result of this formula is not a TDP in the usual sense of the term".

AMD was using this de-facto standard before Zen. For instance, the 125W Piledrivers are 125W CPUs.

The reviewer is not omitting AVX workloads from Intel chips. He only avoids "100% AVX loads", which never happen in real-life: "do not correspond to a realistic load,". Those unrealistic loads only happen when running certain stress tests. Recall that AVX is an extension to run over a x86-code baseline. It is not a separate ISA that runs alone for 100% of time.

In real cases Intel chips satisfy the marketing TDP values, as confirmed by the reviewer: "At Intel, the consumption limits for the Turbo and the TDP are identical, which seems the most logical since each watts consumed by the processor is discharged in the form of heat. "

This is not AMD vs Intel thing. As mentioned above AMD CPUs previous to RyZen satisfy the marketing TDP. ThreadRipper CPUs also satisfy the marketing TDP (1). Lower RyZen models also satisfy the marketing TDP. The only CPUs that violate the marketing value are the 1800X/1700X and the 1700 models.

(1) Albeit AMD forces the limit to the marketing TDP by deceiving about clocks.

From tom's hardware:

"And it's notable that the 95W Ryzen 7 1700X we're reviewing today uses less power under our stress test than Intel's 91W Core i7-7700K."

They used a power virus only for the Intel chip, not a power virus for RyZen. The AVX-256 capable stress test is able to load fully the pair of powerful 256bit SIMD units on Kabylake. Units are absent in RyZen, which has only a pair of 128bit FMACs. In fact RyZen consumes about the same power on this 'stress' test than on a ordinary workload as Luxrender. RyZen consumes 112W on both. and only 16W than on Blender. Evidenlty Luxrender is not a stress test.

Any efficiency metric using comparable workloads shows that Kabylake consumes less and is more efficient than 1800X/1700X.

Intel TDP

Spoiler

In real cases Intel chips satisfy the marketing TDP values, as confirmed by the reviewer: "At Intel, the consumption limits for the Turbo and the TDP are identical, which seems the most logical since each watts consumed by the processor is discharged in the form of heat. " Intel does not agree with this!

TDP is related to power consumption only so much as to find a CPU Cooler capable of keep the CPU within normal operating temperatures under a workload defined by AMD. You can cool the 1800X/1700X with a~$15 cooler no problem.Hyper 212 LED with PWM Fan, Four Direct Contact Heatpipes, Unique Fan Blade Design, Red LEDs, Optimized Bracket Design by Cooler Master $14.99
I found the article http://www.hardware.fr/articles/956-10/consommation-efficacite-energetique-tdp.html

Spoiler

Note: The 1800X out performing the 6900X and 6950X in single thread. Also, outperforming multithread of the 8 core 6900X, and 0.48(.02%) away from the 10 core 6950X in this test, which were 2 and 3 times more expensive at $1,100 and $1,800 CPU's.
They are using a demanding workload and max power consumption for that workload to rate TDP! This is not how TDP is measured! By that same standard the 7700K fails with 109.4W power consumption over its 95W TDP as well as the i7-5960X at 153.6W power consumption. The article also states:

At the ATX12V, on one thread, Ryzen is much better than the X99 platforms, enjoying a lower idle power consumption.

In the same way this gap is found when all hearts are stressed, the Ryzen R7 having an efficiency that approaches that of Broadwell-E (between 6800K and 6900K) and clearly above Haswell-E. We are closer to the efficiency of a 6700K in practice, which is pretty good (we note in passing that Kaby Lake, the overclocked version of Skylake, is no miracle and pay dearly its extra MHz on consumption and efficiency!).

http://www.trustedreviews.com/reviews/amd-ryzen-7-1800x-performance-gaming-and-overclocking-page-2

As for power consumption, the 1800X again comfortably beats the 6900K, with the whole system consuming just 41W at idle compared to the 71W of the 6900K. Under load the gap actually narrows, which I wasn’t expecting, but the 1800X still has a clear 10W advantage.

https://arstechnica.com/gadgets/2017/03/amd-ryzen-review/

Due to Ryzen arriving rather late in the Ars Technica offices (thanks, AMD!), I haven't fully tested Ryzen overclocking. AMD claims most 1800X chips will hit 4.2GHz at a voltage of 1.4V. That's higher than the 1.35V typically needed to get a Broadwell-E CPU to 4.4GHz, but not dramatically so, and as always with CPU overclocking some chips will fare better than others. That said, 4.2GHz seems to be the typical overclock for the 1800X, as offered by retailers like Scan, which is selling pre-overclocked systems, and it's unlikely to go much higher. Multiple publications I spoke to with earlier access to chips all report that 4.2GHz is the current limit with standard air or sealed liquid coolers.
Still, such speeds would make the £320 R7 1700 an enormous bargain, should it overclock to 1800X speeds.

Part of why Ryzen's gaming performance is so spotty, and why CPUs are more complex beasts than can be explained by a simple MHz number or IPC comparison is highlighted by the CPU-specific benchmarks.

For the most part, Ryzen is highly competitive with Intel's eight-core i7-6900K. In Cinebench R15—which is an excellent test of practical processing performance, since it splits the rendering of a complex photorealistic 3D scene across all CPU cores—Ryzen comes out ahead of Intel's chips. The chip maxed out at around 72°C with the Noctua tower air cooler, too—not bad.

Even when locked to the same 3.5GHz clock speed, Ryzen is faster. The Handbrake test works out well for Ryzen too, with the CPU again coming in faster than Intel. In real-world, practical tests, Ryzen is as an excellent workstation chip.

http://www.guru3d.com/articles-pages/amd-ryzen-7-1800x-processor-review,7.html

Next to that, we stress all CPU cores 100% and thus show peak power consumption. Unless you transcode video with the right software your average power consumption will be much lower.

Overall stress/load temperatures are very nice with temps at the ~70 C marker peaks (with a simple heat-pipe based Noctua U12S SE AM4 cooler). These, of course, are default results and not tweaked. The processor idles at roughly 45 to 50 Degrees C. Again, we used the AMD review kit supplied Noctua cooler here for cooling. We also received a liquid cooling kit (Predator 240 from EK) which we will use in the overclocking segment. That unit keeps the cores chilled at roughly 65 Degrees C under stress and being tweaked. Overall we are very happy with the temperature results here.

http://www.tomshardware.com/reviews/amd-ryzen-5-1600x-cpu-review,5014-9.html

I've said my peace on the subject of TDP and we have differences of opinion when it comes to the meaning of TDP, let's agree to disagree!

Edit: I used wrong picture for Intel TDP, I replace it.

 

[goldstone77]

Thanks goldstone, lots of interesting info in those figures.

AMD will use the 7nm SOC. 40% higher frequency means that potentially AMD can produce chips with 5GHz base and 5.6 Ghz turbo in the same power envelope (1800X is 3.6-4GHz, 4.1 XFR).

That is at normalized frequency, and clearly in the low range. No way we will those 5.0/5.6GHz.

IBM will use 7nmHPC, the figure sugests 6-7 GHz is possible.

This is pure nonsense. The performance version of 7LP is designed for 4GHz and IBM is targeting those clocks.

14nm 1.3 Normalized Frequency 7nm Soc 1.7 Normalized Frequency
1.3 to 1.7 is 30.7%@less power consumption
30.7% X 3.6GHz = 4.7052GHz
30.7% X 4.1GHz = 5.3587GHz
7nmHPC
1.7 to 2.1 is 23.5% @~2.5x power consumption
23.5% X 4.7052 = 5.810922GHz
23.5% X 5.3597GHz = 6.6179945GHz

Edit: added 7nmHPC

 

[juanrga]

I've said my peace on the subject of TDP and we have differences of opinion when it comes to the meaning of TDP, let's agree to disagree!

It is not about opinions but about measurements.

Phenom II, Bulldozer, and Piledriver FX are within the official 125W TDP.

RyZen 1600X is within the official 95W TDP.

RyZen 1500X, 1400, 1300X, and 1200 are within the official 65W TDP.

ThreadRipper 1950X and 1920X are within the official 180W TDP.

RyZen 1800X, and 1700X breach the official 95W TDP. We know now the real TDP is 128W.
RyZen 1700 breaches the official 65W TDP. We know now the real TDP is 90W.

 

[juanrga]

14LPP is optimized for about 2.5GHz. Glofo is giving the maximum possible percentage, which corresponds to the optimal regions of the node.

So 2.5GHz * 1.40 = 3.5GHz. This means fhe 7LPP node is optimized for frequencies near 4GHz.

For the top range of the process node I expect much more moders 20% improvements.

So 4GHz * 1.20 = 4.8GHz.

And Glofo in the marketing material claims the node can get up to 5GHz.

So I expect maximum clocks to be around 4.8GHz for future AMD chips.

The 6GHz and 7GHz claimed here are pure nonsense.

 

[-Fran-]

It is not about opinions but about measurements.

Right... Because when you "define" something is strictly about a measurement and not your opinion on *how* something has to be measured...

You're just wasting everyone's time with that TDP discussion.

 

[YoAndy]

14LPP is optimized for about 2.5GHz. Glofo is giving the maximum possible percentage, which corresponds to the optimal regions of the node.

So 2.5GHz * 1.40 = 3.5GHz. This means fhe 7LPP node is optimized for frequencies near 4GHz.

For the top range of the process node I expect much more moders 20% improvements.

So 4GHz * 1.20 = 4.8GHz.

And Glofo in the marketing material claims the node can get up to 5GHz.

So I expect maximum clocks to be around 4.8GHz for future AMD chips.

The 6GHz and 7GHz claimed here are pure nonsense.

I'm with you on this one, I don't think they will pass 4.8GHz (on average), Maybe 5GHz in some extreme rare cases. I personally think new Zen will be closer to the 4.5GHz.

Right now the average Zen overclocker is only getting 3.8-3.9GHz, With the 1800X getting a bit more lucky and sometimes passing the 4GHz barrier on all cores..

 

[juanrga]

It is not about opinions but about measurements.

Right... Because when you "define" something is strictly about a measurement and not your opinion on *how* something has to be measured...

You're just wasting everyone's time with that TDP discussion.

All measurements given above were performed in exactly the same way. The measurements prove that R3/R5 RyZen chips and all ThreadRipper chips are within the official TDP, whereas R7 RyZen chips violate the official TDP. AMD already admitted the real TDP for the R7s (check the quotes given above) and those real TDPs communicated by AMD agree with measurements made by both HFR and Canard.

So no, it is not about definitions, because the definition of TDP is the same for all RyZen and ThreadRipper chips... it is about some RyZen chips violating the marketing TDP values.

 

[-Fran-]

It is not about opinions but about measurements.

Right... Because when you "define" something is strictly about a measurement and not your opinion on *how* something has to be measured...

You're just wasting everyone's time with that TDP discussion.

All measurements given above were performed in exactly the same way. The measurements prove that R3/R5 RyZen chips and all ThreadRipper chips are within the official TDP, whereas R7 RyZen chips violate the official TDP. AMD already admitted the real TDP for the R7s (check the quotes given above) and those real TDPs communicated by AMD agree with measurements made by both HFR and Canard.

So no, it is not about definitions, because the definition of TDP is the same for all RyZen and ThreadRipper chips... it is about some RyZen chips violating the marketing TDP values.

So you're implying you have the official AMD definition of TDP that reviewers you have been quoting follow to the letter, right? Would you mind sharing that precise and exact definition, please? I'll see how the procedures for the sites you have quoted meet the criteria defined for measuring TDP.

 

[8350rocks]

They are also calling it Ryzen Refresh, In the slide you show you can see they had always planned on upgrading the node... 14nm to 14nm+ this is now 12nm of coarse... Zen on the new node is whats classed as Zen+. Whatever changes they make changes to the uArch remains to be seen. I suppose it would be foolish if they made none...

There is no 12nm node. "12LP" is a marketing name for 14nm+. There is no Zen+, as showed in the roadmaps given above.

Zen in 12LP continues being Zen. The microarchitecture is the same. That is why the AMD roadmap puts "Zen" in giant font.

There is no 10nm from Intel either...it is vaporsilicon

 

[adamsleath]

yes . power rises exponentially with clocks. how does that 40% project at higher clocks?
4.8GHz real turbo you reckon? +800-1000MHz ...

just imagine if intel gets the same increment at 10nm+ ? :lol:
could be a 4.8 GHz base clock by then.

..just thinking its been about 10 years to get from 4.0 Overclocked to 5.0 overclocked.
----------------

as for power usage. only have to look at the graphs to make your own assessment. no surprises there.

 

[jdwii]

14LPP is optimized for about 2.5GHz. Glofo is giving the maximum possible percentage, which corresponds to the optimal regions of the node.

So 2.5GHz * 1.40 = 3.5GHz. This means fhe 7LPP node is optimized for frequencies near 4GHz.

For the top range of the process node I expect much more moders 20% improvements.

So 4GHz * 1.20 = 4.8GHz.

And Glofo in the marketing material claims the node can get up to 5GHz.

So I expect maximum clocks to be around 4.8GHz for future AMD chips.

The 6GHz and 7GHz claimed here are pure nonsense.

Seems totally realistic and that is if the architecture can even OC that high, Currently today even liquid Nitrogen can't help Ryzen OC much.

Agree 6-7Ghz claims are laughable simply not possible on commercial processors, in a lab with the design made for high frequency's then yeah sure but Ryzen isn't made for pure speeds it was more of an balanced approach

 

[goldstone77]

14LPP is Samsung design, and 7LP is GlobalFoundries in house design.

Edit:
https://www.globalfoundries.com/sites/default/files/product-briefs/7lp-product-brief.pdf
2nd Edit:

SE: Is GlobalFoundries co-developing 7nm technology with Samsung?

Patton: We are doing it completely on our own. We had the collaboration on 14nm. We still partner with them on what I’d call pathfinding in Albany. As part of the deal with IBM, Albany used to be an all-joint-development alliance. Then it was spilt into two parts. Half of the facility continues to do the joint development work, which IBM leads. Think of that as pathfinding for 5nm and beyond. And the other half is a proprietary IBM-GlobalFoundries corridor, which is specifically focused on accelerating things into Malta. Those could be performance elements for 7nm and 7nm plus. In fact, I would envision we’ll have a bunch of performance kickers at 7nm. This will be a long node.

https://semiengineering.com/inside-fd-soi-scaling/

 

[juanrga]

..just thinking its been about 10 years to get from 4.0 Overclocked to 5.0 overclocked.

That has a name: frequency wall

 

[-Fran-]

..just thinking its been about 10 years to get from 4.0 Overclocked to 5.0 overclocked.

That has a name: frequency wall

Not quite. It's a self-imposed wall. You can get higher speeds if you're willing to consume more power to achieve it.

I can totally see IBM going for 6Ghz if they so want to if the silicon allows within a reasonable power envelope.

 

[goldstone77]

IBM claims 210-GHz SiGe transistor is world's fastest, four years ahead of competition
EE Times
6/25/2001 09:42 AM EDT


DAILY NEWS 11 April 2005
World’s fastest transistor operates at blinding speed
By
Will Knight

Feng and Hafez developed a transistor less than half a millionth of a metre long, with a maximum operating speed of 604 GHz, meaning it can carry out 604 billion operations every second.

 

[goldstone77]

Spoiler

..just thinking its been about 10 years to get from 4.0 Overclocked to 5.0 overclocked.

That has a name: frequency wall

Not quite. It's a self-imposed wall. You can get higher speeds if you're willing to consume more power to achieve it.

I can totally see IBM going for 6Ghz if they so want to if the silicon allows within a reasonable power envelope.

IBM also has water cooled servers!

 

[8350rocks]

..just thinking its been about 10 years to get from 4.0 Overclocked to 5.0 overclocked.

That has a name: frequency wall

Not quite. It's a self-imposed wall. You can get higher speeds if you're willing to consume more power to achieve it.

I can totally see IBM going for 6Ghz if they so want to if the silicon allows within a reasonable power envelope.

Yes...and even then...once things like EUV lithography, or graphene, come into play you will see that the "path of least resistance" (see what I did there?) will shift the goal posts of difficulty achieving frequency to a much higher target.

 

[juanrga]

..just thinking its been about 10 years to get from 4.0 Overclocked to 5.0 overclocked.

That has a name: frequency wall

Not quite. It's a self-imposed wall. You can get higher speeds if you're willing to consume more power to achieve it.

I can totally see IBM going for 6Ghz if they so want to if the silicon allows within a reasonable power envelope.

Wrong. The frequency wall is a well-know limit of the laws of physics. Wait sitting for those 6GHz IBM chips. LOL

 

[juanrga]

..just thinking its been about 10 years to get from 4.0 Overclocked to 5.0 overclocked.

That has a name: frequency wall

Not quite. It's a self-imposed wall. You can get higher speeds if you're willing to consume more power to achieve it.

I can totally see IBM going for 6Ghz if they so want to if the silicon allows within a reasonable power envelope.

Yes...and even then...once things like EUV lithography, or graphene, come into play you will see that the "path of least resistance" (see what I did there?) will shift the goal posts of difficulty achieving frequency to a much higher target.

EUV doesn't change the scaling laws of silicon, neither the laws of electromagnetism.