Too Hot to Last? Investigating Intel's Claims About Ryzen Reliability
AMD's Ryzen 3000-Series processors landed two months ago, bringing with them an incredible increase in real-world performance and upsetting the pricing paradigm with an impressive increase in performance-per-dollar, but the launch has been marred by reports that many users aren't receiving the rated boost speeds. AMD announced this week that it had identified an issue with its firmware that reduces performance in some situations and that it would update the community on an incoming fix on September 10.
As we often see in marketing, Intel has chosen to attack during AMD's perceived time of weakness. At the IFA tradeshow this week, Intel presented a slide deck to members of the press that includes information from a recent survey conducted by YouTuber Der8auer in which a surprising number of respondents reported they have been unable to reach the rated boost frequencies with their Ryzen 3000 processors.
Interestingly, Intel then drove further on the issue, citing a report that claims reliability is behind AMD's apparent, but not proven, reasons for reducing its chips' frequencies.
We were already investigating the claims Intel cited in regards to the relationship between Ryzen's clock frequencies and longevity, and we had secured comment from AMD before its admission that there was an issue with its firmware. Today we'll present some of the testing we conducted to investigate those claims.
Also at CRN.
Previously: Survey Says Many Ryzen 3900X CPUs Can't Hit Rated Boost Clock Speeds, BIOS Fix on Sept. 10th
Related Stories
Survey Says Many Ryzen 3900X CPUs Can't Hit Rated Boost Clock Speeds
Survey: Only 5.6 Percent of Ryzen 9 3900X Hit Advertised Speeds, Most Other Models Suffer, Too
Overclocker and hardware reviewer De8auer, widely known for his Intel delidding tools and overclocking videos, has released the results of a survey he conducted late last month concerning Ryzen 3000's ability to reach its advertised boost clocks. Only 5.6% of respondents reported that their Ryzen 9 3900X is reaching its rated boost speed. The results are somewhat better with other SKUs, but still indicate that the majority of Ryzen 3000 series processors are not hitting their rated boost speeds.
Users and reviewers alike have been questioning whether or not AMD's new CPUs are always able to boost to the advertised clock speeds. We recently published an analysis on the 3600X detailing Ryzen 3000's new boosting behavior, and AMD confirmed that only one core on any given CPU is guaranteed to hit the rated boost clock. However, according to the survey, more users aren't even reaching the advertised frequency on any core.
(Score: 2, Informative) by Anonymous Coward on Sunday September 08 2019, @08:28AM (1 child)
Tom's Hardware was always a biased Intel's bitch. For several decades.
(Score: 3, Interesting) by jon3k on Sunday September 08 2019, @01:42PM
This was acknowledged by AMD on twitter [twitter.com] and there is a fix coming. But I agree, there is a tremendous amount of FUD in this article, implying that your AMD CPU is going to fail over time.
But, to be fair, from the conclusion:
and
(Score: 5, Insightful) by Rosco P. Coltrane on Sunday September 08 2019, @10:32AM (9 children)
Back in the Pentium days, I remember the only difference between 75, 90, 133 and 200Mhz CPUs is how much they heated up under load at the final testing stage. They were the same chip, with different ratings printed on them and different price tags tacked to them according to the final test.
I know cuz back then, I made a killing selling overclocked PCs with el-cheapo Pentium chips and beefed up heatsinks. None of my customers ever complained.
So I'd say AMD probably oversold lesser quality chips, is all. But said lesser quality chips are normal and should be expected - just not at premium quality chip prices.
(Score: 3, Touché) by Anonymous Coward on Sunday September 08 2019, @11:42AM
Not unlike the chips Intel sells that are still vulnerable to Meltdown and Specter...
But this is where the similarities end.
(Score: 4, Insightful) by takyon on Sunday September 08 2019, @12:18PM (4 children)
Manual overclocking is starting to wane as CPUs can automatically overclock themselves based on "the silicon lottery", workloads, and the available cooling solution. There is apparently not much overclocking potential for something like Ryzen 3000, unless you start talking about absurd and expensive cooling. It has already been optimized to reach the clock speeds it reaches.
When it comes to the turbo/boost clock speeds, definitions matter. The company can choose to either highlight the all-core boost, single-core boost, or even the boosts for 1, 2, 3, 4, 8, etc. cores at the same time. For Ryzen 3000, the publicized boost clock is only guaranteed for one core. But apparently many users couldn't even reach that. That will be addressed by a BIOS fix on Sept. 10.
Intel harping on longevity is a little bizarre. Has anyone here actually had any CPU, overclocked or not, die after 5-10 years? By the time it dies, replacement systems should run circles around it. CPU death also isn't the tragedy that a hard drive failure with data loss could be.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 1, Interesting) by Anonymous Coward on Sunday September 08 2019, @01:38PM (3 children)
I think the proper stat should be area under the boost curve. Ie, for an 8core cpu the curve could be:
ncore = 1:8
boost = c(4.5, 4.4, 4.3, 4, 3.8, 3.6, 3.2, 3)
auc = sum(sma(boost, 2))
In this case auc = 27.05. Perhaps then we could get some measure of energy used or heat generated and get the ratio. Say at max boost the cpu is using 100 w. Then the stat would be 270.5 GHz/mW.
(Score: 3, Informative) by takyon on Sunday September 08 2019, @02:38PM (2 children)
0.28 MHz/mW, surely?
See my naive calculation (I just picked 3.5 GHz as an average, didn't try to work backward):
3.5 GHz * 8 cores = 28 GHz
28 GHz / 100 Watts = 0.28 GHz / Watt
0.28 GHz / Watt = 280 MHz / Watt = 0.28 MHz / mW
Green500 top 10 systems [top500.org] have 10-15 gigaflops per Watt.
Zen 2 has 16-32 FLOPS per cycle [wikichip.org]. Multiply 16 by 0.28 GHz / Watt and you get about 4.48 gigaflops per Watt for double precision. Which is probably not a real world number but may be in the ballpark.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 0) by Anonymous Coward on Sunday September 08 2019, @02:46PM (1 child)
I meant GHz/kW.
(Score: 2) by takyon on Sunday September 08 2019, @02:58PM
Looks good.
Seeing the "aggregate GHz" go up over time as core counts increase is nice. For the Ryzen 9 3950X, 3.5 GHz base clock * 16 cores. It doesn't take into account IPC improvements though.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 1) by jon3k on Sunday September 08 2019, @01:32PM
Are you talking about chip binning [wikipedia.org]? That's still very much a thing.
(Score: 0, Troll) by Ethanol-fueled on Sunday September 08 2019, @01:59PM (1 child)
Your customers didn't hear the near-ultrasonic squealing of their chips being worked to a premature death like field niggers on a plantation?
Man, am I glad that even back in those days I saw overclocking for what it really was: A big gimmicky swindle.
(Score: 0) by Anonymous Coward on Sunday September 08 2019, @04:01PM
Are you the king of stupid around here, or what?
(Score: 2) by Bot on Sunday September 08 2019, @12:52PM (1 child)
You are telling me Intel has to resort to a third party to judge a board? They cannot buy 10 of them and check out for themselves? Especially since they have likely put some under very close scrutiny anyway.
Account abandoned.
(Score: 4, Interesting) by takyon on Sunday September 08 2019, @01:30PM
AMD already admitted they have a problem, with a fix coming out on Tuesday.
The real puzzler here is that Intel seems to have determined that AMD's CPUs will have a reliability/longevity problem.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 2) by Azuma Hazuki on Sunday September 08 2019, @08:32PM (1 child)
Don't think I've forgotten what they did back in the Athlon days. Intel, now as then, got sloppy and can't compete, so they're attempting to poison the well. I've been wondering when they'd try this, and looking forward to seeing them do something blatantly illegal again...
I am "that girl" your mother warned you about...
(Score: 4, Interesting) by takyon on Sunday September 08 2019, @08:54PM
I think I might give in to Intel if they put something out with 3x or more performance/$ than AMD.
So, basically never. Although there is a chance to do something like that by being first to bring a 3DSoC design to market. AMD is working toward the interim design [pcgamesn.com] for 2021 or later.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]