Survey: Many AMD Ryzen 3000 CPUs Don’t Hit Full Boost Clock


Overclocker Der8auer has published the results of a survey of more than 3,000 Ryzen 7nm owners who have purchased AMD’s new CPUs since they went on sale in July. Last month, reports surfaced that the Ryzen 3000 family weren’t hitting their boost clocks as well as some enthusiasts expected. Now, we have some data on exactly what those figures look like.

There are, however, two confounding variables. First, Der8auer had no way to sort out which AMD users had installed Windows 1903 and were using the most recent version of the company’s chipset drivers. AMD recommends both to ensure maximum performance and desired boost behavior. Der8auer acknowledges this but believes the onus is on AMD to communicate with end-users regarding the need to use certain Windows versions to achieve maximum performance.

Second, there’s the fact that surveys like this tend to be self-selecting. It’s possible that only the subset of end-users who aren’t seeing the performance they desire will respond in such a survey. Der8auer acknowledges this as well, calling it a very valid point, but believes that his overall viewing community is generally pro-AMD and favorably inclined towards the smaller CPU manufacturer. The full video can be seen below; we’ve excerpted some of the graphs for discussion.

Der8auer went over the data from the survey thoroughly in order to throw out results that didn’t make sense or were obviously submitted in bad faith. He compiled data on the 3600, 3600X, 3700X, 3800X, and 3900X.SEEAMAZON_ET_135 See Amazon ET commerce Clock distributions were measured at up to two deviations from the mean. Maximum boost clock was tested using Cinebench R15’s single-threaded test, as per AMD’s recommendation.

Der8auer-3600

Data and chart by Der8auer. Click to enlarge

In the case of the Ryzen 7 3600, 49.8 percent of CPUs hit their boost clock of 4.2GHz, as shown above. As clocks rise, however, the number of CPUs that can hit their boost clock drops. Just 9.8 percent of 3600X CPUs hit their 4.4GHz. The 3700X’s chart is shown below for comparison:

Data and chart by Der8auer. Click to enlarge

The majority of 3700X CPUs are capable of hitting 4.375GHz, but the 4.4GHz boost clock is a tougher leap. The 3800X does improve on these figures, with 26.7 percent of CPUs hitting boost clock. This seems to mirror what we’ve heard from other sources, which have implied that the 3800X is a better overclocker than the 3700X. The 3900X struggles more, however, with just 5.6 percent of CPUs hitting their full boost clock.

We can assume that at least some of the people who participated in this study did not have Windows 10 1903 or updated AMD drivers installed, but AMD users had the most reason to install those updates in the first place, which should help limit the impact of the confounding variable.

The Ambiguous Meaning of ‘Up To’

Following his analysis of the results, Der8auer makes it clear that he still recommends AMD’s 7nm Ryzen CPUs with comments like “I absolutely recommend buying these CPUs.” There’s no ambiguity in his statements and none in our performance review. AMD’s 7nm Ryzen CPUs are excellent. But an excellent product can still have issues that need to be discussed. So let’s talk about CPU clocks.

The entire reason that Intel (who debuted the capability) launched Turbo Boost as a product feature was to give itself leeway when it came to CPU clocks. At first, CPUs with “Turbo Boost” simply appeared to treat the higher, optional frequency as their effective target frequency even when under 100 percent load. This is no longer true, for multiple reasons. CPUs from AMD and Intel will sometimes run at lower clocks depending on the mix of AVX instructions. Top-end CPUs like the Core i9-9900K may throttle back substantially when under full load for a sustained period of time (20-30 seconds) if the motherboard is configured to use Intel default power settings.

In other realms, like smartphones, it is not necessarily unusual for a device to never run at maximum clock. Smartphone vendors don’t advertise base clocks at all and don’t provide any information about sustained SoC clock under load. Oftentimes it is left to reviewers to typify device behavior based on post-launch analysis. But CPUs from both Intel and AMD have typically been viewed as at least theoretically being willing capable of hitting boost clock in some circumstances.

The reason I say that view is “theoretical” is that we see a lot of variation in CPU behavior, even over the course of a single review cycle. It’s common for UEFI updates to arrive after our testing has already begun. Oftentimes, those updated UEFIs specifically fix issues with clocking. We correspond with various motherboard manufacturers to tell them what we’ve observed and we update platforms throughout the review to make certain power behavior is appropriate and that boards are working as intended. When checking overall performance, however, we tend to compare benchmark results against manufacturer expectations as opposed to strictly focusing on clock speed (performance, after all, is what we are attempting to measure). If performance is oddly low or high, CPU and RAM clocks are the first place to check.

It’s not unusual, however, to be plus-or-minus 2-3 percent relative to either the manufacturer or our fellow reviewers, and occasional excursions of 5-7 percent may not be extraordinary if the benchmark is known for producing a wider spread of scores. Some tests are also more sensitive than others to RAM timing, SSD speed, or a host of other factors.

Now, consider Der8auer’s data on the Ryzen 9 3900X:

Der8auer-3900X

Image and data by Der8auer. Click to enlarge

Just 5 percent of the CPUs in the batch are capable of hitting 4.6GHz. But a CPU clocked at 4.6GHz is just 2 percent faster than a CPU clocking in at 4.5GHz. A 2 percent gap between two products is close enough that we call it an effective tie. If you were to evaluate CPUs strictly on the basis of performance, with a reasonable margin of say, 3 percent, you’d wind up with an “acceptable” clock range of 4,462MHz – 4,738MHz (assuming a 1:1 relationship between CPU clock and performance). And if you allow for that variance in the graphs above, a significantly larger percentage — though no, not all — of AMD CPUs “qualify” as effectively reaching their top clock.

On the other hand, 4.5GHz or below is factually not 4.6GHz. There are at least two meaningfully different ways to interpret the meaning of “up to” in this context. Does “up to X.XGHz” mean that the CPU will hit its boost clock some of the time, under certain circumstances? Or does it mean that certain CPUs will be able to hit these boost frequencies, but that you won’t know if you have one or not? And how much does that distinction matter, if the overall performance of the part matches the expected performance that the end-user will receive?

Keep in mind that one thing these results don’t tell us is what overall performance looks like across the entire spread of Ryzen 7 CPUs. Simply knowing the highest boost clock that the CPU hits doesn’t show us how long it sustained that clock. A CPU that holds a steady clock of 4.5GHz from start to finish will outperform a CPU that bursts to 4.6GHz for one second and drops to 4.4GHz to finish the workload. Both of these behaviors are possible under an “up to” model.

Manufacturers and Consumers May See This Issue Differently

While I don’t want to rain on his parade or upcoming article, we’ve spent the last few weeks at ET troubleshooting a laptop that my colleague David Cardinal recently bought. Specifically, we’ve been trying to understand its behavior under load when both the CPU and GPU are simultaneously in-use. Without giving anything away about that upcoming story, let me say this: The process has been a journey into just how complicated thermal management is now between various components.

Manufacturers, I think, increasingly look at power consumption and clock speed as a balancing act in which performance and power are allocated to the components where they’re needed and throttled back everywhere else. Increased variability is the order of the day. What I suspect AMD has done, in this case, is set a performance standard that it expects its CPUs to deliver rather than a specific clock frequency target. If I had to guess at why the company has done this, I would guess that it’s because of the intrinsic difficulties of maintaining high clock speeds at lower process nodes. AMD likely chose to push the envelope on its clock targets because it made the CPUs compare better against their Intel equivalents as far as maximum clock speeds were concerned. Any negative response from critics would be muted by the fact that these new CPUs deliver marked benefits over both previous-generation Ryzen CPUs and their Intel equivalents at equal price points.

Was that the right call? I’m not sure. This is a situation where I genuinely see both sides of the issue. The Ryzen 3000 family delivers excellent performance. But even after allowing for variation caused by Windows version, driver updates, or UEFI issues on the part of the manufacturer, we don’t see as many AMD CPUs hitting their maximum boost clocks as we would expect, and the higher-end CPUs with higher boost clocks have more issues than lower-end chips with lower clocks. AMD’s claims of getting more frequency out of TSMC 7nm as compared with GF 12/14nm seem a bit suspect at this point. The company absolutely delivered the performance gains we wanted, and the power improvements on the X470 chipset are also very good, but the clocking situation was not detailed the way it should have been at launch.

There are rumors that AMD supposedly changed boost behavior with recent AGESA versions. Asus employee Shamino wrote:

i have not tested a newer version of AGESA that changes the current state of 1003 boost, not even 1004. if i do know of changes, i will specifically state this. They were being too aggressive with the boost previously, the current boost behavior is more in line with their confidence in long term reliability and i have not heard of any changes to this stance, tho i have heard of a ‘more customizable’ version in the future.

I have no specific knowledge of this situation, but this would surprise me. First, reliability models are typically hammered out long before production. Companies don’t make major changes post-launch save in exceptional circumstances, because there is no way to ensure that the updated firmware will reach the products that it needs to reach. When this happens, it’s major news. Remember when AMD had a TLB bug in Phenom? Second, AMD’s use of Adaptive Frequency and Voltage Scaling is specifically designed to adjust the CPU voltage internally to ensure clock targets are hit, limiting the impact of variability and keeping the CPU inside the sweet spot for clock.

I’m not saying that AMD would never make an adjustment to AGESA that impacted clocking. But the idea that the company discovered a critical reliability issue that required it to make a subtle change that reduced clock by a mere handful of MHz in order to protect long-term reliability doesn’t immediately square with my understanding of how CPUs are designed, binned and tested. We have reached out to AMD for additional information.

I’m still confident and comfortable recommending the Ryzen 3000 family because I’ve spent a significant amount of time with these chips and seen how fast they are. But AMD’s “up to” boost clocks are also more tenuous than we initially knew. It doesn’t change our expectation of the part’s overall performance, but the company appears to have decided to interpret “up to” differently this cycle than in previous product launches. That shift should have been communicated. Going forward, we will examine both Intel and AMD clock behavior more closely as a component of our review coverage.

Now Read:




10 minutes mail – Also known by names like : 10minemail, 10minutemail, 10mins email, mail 10 minutes, 10 minute e-mail, 10min mail, 10minute email or 10 minute temporary email. 10 minute email address is a disposable temporary email that self-destructed after a 10 minutes. https://tempemail.co/– is most advanced throwaway email service that helps you avoid spam and stay safe. Try tempemail and you can view content, post comments or download something

Chinese Foundry SMIC Begins 14nm Production


This site may earn affiliate commissions from the links on this page. Terms of use.

One of the longstanding trends in semiconductor manufacturing has been a steady decrease in major foundry players. Twenty years ago, when 180nm manufacturing was cutting-edge technology, there were no fewer than 28 firms deploying the node. Today, there are three companies building 7nm technology — Samsung, TSMC, and Intel. A fourth, GlobalFoundries, has since quit the cutting-edge business to focus on specialty foundry technologies like its 22nm and 12nm FDX technology.

What sometimes gets lost in this discussion, however, is the existence of a secondary group of foundry companies that do deploy new nodes — just not at the cutting-edge of technological research. China’s Semiconductor Manufacturing International Corporation (SMIC) has announced that it will begin recognizing 14nm revenue from volume production by the end of 2019, a little more than five years after Intel began shipping on this node. TSMC, Samsung, and GlobalFoundries all have extensive 14nm capability in production, as does UMC, which introduced the node in 2017.

Secondary sources for a node, like UMC and SMIC, often aren’t captured in comparative manufacturing charts like the one below because the companies in question offer these nodes after they’ve been deployed as cutting-edge products by major foundries. In many cases, they’re tapped by smaller customers with products that don’t make news headlines.

FoundryManufacturing

SMIC, however, is something of a special case. SMIC is mainland China’s largest semiconductor manufacturer and builds chips ranging from 350nm to 14nm. The company has two factories with the ability to process 300mm wafers, but while moving to 14nm is a major part of China’s long-term semiconductor initiative, SMIC isn’t expected to have much 14nm capacity any time soon. The company’s high utilization rate (~94 percent) precludes it having much additional capacity to dedicate to 14nm production. SMIC is vital to China’s long-term manufacturing goals; the country’s “Made in China 2025” plan calls for 70 percent of its domestic semiconductor demand to come from local companies by 2025. Boosting production at SMIC and bringing new product lines online is vital to that goal. That distinguishes the company from a foundry like UMC, which has generally chosen not to compete with TSMC for leading-edge process nodes. SMIC wants that business — it just can’t compete for it yet.

Dr. Zhao Haijun and Dr. Liang Mong Song, SMIC’s Co-Chief Executive Officers released a statement on the company’s 14nm ramp, saying:

FinFET research and development continues to accelerate. Our 14nm is in risk production and is expected to contribute meaningful revenue by year-end. In addition, our second-generation FinFET N+1 has already begun customer engagement. We maintain long-term and steady cooperation with customers and clutch onto the opportunities emerging from 5G, IoT, automotive and other industry trends.

Currently, only 16 percent of the semiconductors used in China are built there, but the country is adding semiconductor production capacity faster than anywhere else on Earth. The company is investing in a $10B fab that will be used for dedicated 14nm production. SMIC is already installing equipment in the completed building, so production should ramp up in that facility in 2020. Once online, the company will have significantly more 14nm capacity at its disposal (major known customers of SMIC include HiSilicon and Qualcomm). Texas Instruments has built with the company in the past (it isn’t clear if it still does), as has Broadcom. TSMC and SMIC have gone through several rounds of litigation over IP misappropriation; both cases were settled out of court with substantial payments to TSMC.

Despite this spending, analysts do not expect SMIC to immediately catch up with major foundry players from other countries; analysts told CNBC it would take a decade for the firm to close the gap with other major players. Exact dimensions on SMIC’s 14nm node are unknown. Foundry nodes are defined by the individual company not by any overarching standard organization or in reference to any specific metric. Those looking for additional information on that topic will find it here.

Now Read: 




10 minutes mail – Also known by names like : 10minemail, 10minutemail, 10mins email, mail 10 minutes, 10 minute e-mail, 10min mail, 10minute email or 10 minute temporary email. 10 minute email address is a disposable temporary email that self-destructed after a 10 minutes. https://tempemail.co/– is most advanced throwaway email service that helps you avoid spam and stay safe. Try tempemail and you can view content, post comments or download something

How Are Process Nodes Defined?


This site may earn affiliate commissions from the links on this page. Terms of use.

We talk a lot about process nodes at ExtremeTech, but we don’t often refer back to what a process node technically is. With Intel’s 10nm node moving towards production, I’ve noticed an uptick in conversations around this issue and confusion about whether TSMC and Samsung possess a manufacturing advantage over Intel (and, if they do, how large an advantage they possess).

Process nodes are typically named with a number followed by the abbreviation for nanometer: 32nm, 22nm, 14nm, etc. There is no fixed, objective relationship between any feature of the CPUSEEAMAZON_ET_135 See Amazon ET commerce and the name of the node. This was not always the case. From roughly the 1960s through the end of the 1990s, nodes were named based on their gate lengths. This chart from IEEE shows the relationship:

lithot1

For a long time, gate length (the length of the transistor gate) and half-pitch (half the distance between two identical features on a chip) matched the process node name, but the last time this was true was 1997. The half-pitch continued to match the node name for several generations but is no longer related to it in any practical sense. In fact, it’s been a very long time since our geometric scaling of processor nodes actually matched with what the curve would look like if we’d been able to continue actually shrinking feature sizes.

2010-ITRS-Summary

Well below 1nm before 2015? Pleasant fantasy.

If we’d hit the geometric scaling requirements to keep node names and actual feature sizes synchronized, we’d have plunged below 1nm manufacturing six years ago. The numbers that we use to signify each new node are just numbers that companies pick. Back in 2010, the ITRS (more on them in a moment) referred to the technology chum bucket dumped in at every node as enabling “equivalent scaling.” As we approach the end of the nanometer scale, companies may begin referring to angstroms instead of nanometers, or we may simply start using decimal points. When I started work in this industry it was much more common to see journalists refer to process nodes in microns instead of nanometers — 0.18-micron or 0.13-micron, for example, instead of 180nm or 130nm.

How the Market Fragmented

Semiconductor manufacturing involves tremendous capital expenditure and a great deal of long-term research. The average length of time between when a new technological approach is introduced in a paper and when it hits widescale commercial manufacturing is on the order of 10-15 years. Decades ago, the semiconductor industry recognized that it would be to everyone’s advantage if a general roadmap existed for node introductions and the feature sizes those nodes would target. This would allow for the broad, simultaneous development of all the pieces of the puzzle required to bring a new node to market. For many years, the ITRS — the International Technology Roadmap for Semiconductors — published a general roadmap for the industry. These roadmaps stretched over 15 years and set general targets for the semiconductor market.

SemiconductorRoadmap

Image by Wikipedia

The ITRS was published from 1998-2015. From 2013-2014, the ITRS reorganized into the ITRS 2.0, but soon recognized that the scope of its mandate — namely, to provide “the main reference into the future for university, consortia, and industry researchers to stimulate innovation in various areas of technology” required the organization to drastically expand its reach and coverage. The ITRS was retired and a new organization was formed called IRDS — International Roadmap for Devices and Systems — with a much larger mandate, covering a wider set of technologies.

This shift in scope and focus mirrors what’s been happening across the foundry industry. The reason we stopped tying gate length or half-pitch to node size is that they either stopped scaling or began scaling much more slowly. As an alternative, companies have integrated various new technologies and manufacturing approaches to allow for continued node scaling. At 40/45nm, companies like GF and TSMC introduced immersion lithography. Double-patterning was introduced at 32nm. Gate-last manufacturing was a feature of 28nm. FinFETs were introduced by Intel at 22nm and the rest of the industry at the 14/16nm node.

Companies sometimes introduce features and capabilities at different times. AMD and TSMC introduced immersion lithography at 40/45nm, but Intel waited until 32nm to use that technique, opting to roll out double-patterning first. GlobalFoundries and TSMC began using double-patterning more at 32/28nm. TSMC used gate-last construction at 28nm, while Samsung and GF used gate-first technology. But as progress has gotten slower, we’ve seen companies lean more heavily on marketing, with a greater array of defined “nodes.” Instead of waterfalling over a fairly large numerical space (90, 65, 45) companies like Samsung are launching nodes that are right on top of each other, numerically speaking:

I think you can argue that this product strategy isn’t very clear, because there’s no way to tell which process nodes are evolved variants of earlier nodes unless you have the chart handy. But a lot of the explosion in node names is basically marketing.

Why Do People Claim Intel 7nm and TSMC/Samsung 10nm Are Equivalent?

While node names are not tied to any specific feature size, and some features have stopped scaling, semiconductor manufacturers are still finding ways to improve on key metrics. The chart below is drawn from WikiChip, but it combines the known feature sizes for Intel’s 10nm node with the known feature sizes for TSMC’s and Samsung’s 7nm node. As you can see, they’re very similar:

Intel-10-Foundry-7

Image by ET, compiled from data at WikiChip

The delta 14nm / delta 10nm column shows how much each company scaled a particular feature down from its previous node. Intel and Samsung have a tighter minimum metal pitch than TSMC does, but TSMC’s high-density SRAM cells are smaller than Intel’s, likely reflecting the needs of different customers at the Taiwanese foundry. Samsung’s cells, meanwhile, are even smaller than TSMC’s. Overall, however, Intel’s 10nm process hits many of the key metrics as what both TSMC and Samsung are calling 7nm.

Individual chips may still have features that depart from these sizes due to particular design goals. The information manufacturers provide on these numbers are for a typical expected implementation on a given node, not necessarily an exact match for any specific chip.

There have been questions about how closely Intel’s 10nm+ process (used for Ice Lake) reflects these figures (which I believe were published for Cannon Lake). It’s true that the expect specifications for Intel’s 10nm node may have changed slightly, but 14nm+ was an adjustment from 14nm as well. Intel has stated that it is still targeting a 2.7x scaling factor for 10nm relative to 14nm, so we’ll hold off on any speculation about how 10nm+ may be slightly different.

Pulling It All Together

The best way to understand the meaning of a new process node is to think of it as an umbrella term. When a foundry talks about rolling out a new process node, what they are saying boils down to this:

“We have created a new manufacturing process with smaller features and tighter tolerances. In order to achieve this goal, we have integrated new manufacturing technologies. We refer to this set of new manufacturing technologies as a process node because we want an umbrella term that allows us to capture the idea of progress and improved capability.”

Any additional questions on the topic? Drop them below and I’ll answer them.

Now Read: 




10 minutes mail – Also known by names like : 10minemail, 10minutemail, 10mins email, mail 10 minutes, 10 minute e-mail, 10min mail, 10minute email or 10 minute temporary email. 10 minute email address is a disposable temporary email that self-destructed after a 10 minutes. https://tempemail.co/– is most advanced throwaway email service that helps you avoid spam and stay safe. Try tempemail and you can view content, post comments or download something