AMD's Ryzen 9 3900X wrestles wPrime world record from Core i9-7920X

[onetheycallEric]

What just happened? Yet another feather in AMD's cap is the recent wPrime world record that the Ryzen 9 3900X managed to wrestle from Intel's i9-7920X. The record was set by Australian overclocker "jordan.hyde99" as verified by HWbot.

As one might imagine, copious amounts of LN2 were required for the record-setting overclock. However, what you may not have imagined is that the Ryzen 9 3900X took the record at a considerably lower frequency than that of the previous record holder.

The Ryzen 9 3900X and i9-7920X are both 12-core, 24 thread parts. The former is based on AMD's most recent Zen 2 microarchitecture and comes equipped with a 3.8 GHz base clock and a 4.6 GHz boost clock, while the latter is a Skylake derivative with a 2.9 GHz and 4.4 GHz base and boost clock, respectively.

Think: Extreme Overclocking

The Ryzen 9 3900X was overclocked to 5,625 MHz to complete a Wprime 1024M benchmark run in 35 seconds and 517 milliseconds. Comparatively, the i9-7920X was overclocked to 5,955 MHz to complete the same benchmark pass in 35 seconds and 693 milliseconds.

While the win may seem marginal, winning only by a matter of milliseconds, the fact that the Ryzen 9 3900X could do so at a lower frequency is further testament to the IPC improvements of Zen 2.

Also read: 4GHz CPU Battle: Ryzen 3900X vs. 3700X vs. Core i9-9900K

Permalink to story.

https://www.techspot.com/news/83202-amd-ryzen-9-3900x-wrestles-wprime-world-record.html

 

[wiyosaya]

We have officially reverted to was it the Core days when sIntel had CPUs running at a massive frequency advantage over AMD CPUs but, nonetheless, had no performance advantage. sIntels marketing shtick in those days was frequency, frequency, frequency. It was just as lame as "glue".

 

[polord]

I just use wprime too just too check, but my temp went to 100 degrees in an instance, pc crash, I thought my cpu was dead pff. me no use it anymore... lol

 

[lipe123]

We have officially reverted to was it the Core days when sIntel had CPUs running at a massive frequency advantage over AMD CPUs but, nonetheless, had no performance advantage. sIntels marketing shtick in those days was frequency, frequency, frequency. It was just as lame as "glue".

That's an oversimplification and half truth.
Back in the "core days" the core 2 duo chips destroyed amd's athlons in benchmarks. It was not just marketing BS.
What we are seeing now is finally a reverse where AMD actually is beating intel with real performance and not just some hype.

The problem is that this isn't a level playing field, Intel tried to milk 14nm too long and AMD jumped to 7nm giving them a HUGE lead. When Intel eventually does catch up is going to be tough again for amd.
I mean right now intel is still within 90% of AMD's performance with dies that's double the size.

 

[Evernessince]

That's an oversimplification and half truth.
Back in the "core days" the core 2 duo chips destroyed amd's athlons in benchmarks. It was not just marketing BS.
What we are seeing now is finally a reverse where AMD actually is beating intel with real performance and not just some hype.

The problem is that this isn't a level playing field, Intel tried to milk 14nm too long and AMD jumped to 7nm giving them a HUGE lead. When Intel eventually does catch up is going to be tough again for amd.
I mean right now intel is still within 90% of AMD's performance with dies that's double the size.

Zen 2's die size is 74mm2. The 9900K's die size is 177mm2. Each zen 2 die houses 8 cores, equal to the number of cores the 9900K has. It looks to me like you have your math reversed. Intel is the one with die sizes more then double AMD's for the same core count. For it to be close you'd have to include AMD's IO die but that would simply be misleading as the IO die was purposefully made on a larger node (and thus larger) as the elements included on it do not benefit as much from scaling. That said, you stated "die size", not "combined die size" so I'm inclined to compare die to die. It should also be noted that Intel uses a monolithic design while AMD uses a chiplet based design. A chiplet based design is not as space efficient as monolithic but it is far easier to manufacture, cheaper, allows for higher core counts, allows for better binning, and various other benefits.

You are sorely mistaken if you think it's the node that's the main source of AMD's advantage. Their chiplet architecture and IO gains have been far more important. Intel simply moving to a smaller node won't be a magical fix for them. Intel needs a completely new architecture. Hopefully one that doesn't have a new major security hole every month.

 

[amstech]

So core for core, their about even, but hats off to the 3900X, quite impressive.
Still though, 5600MHz for the Ryzen?
Might burn a hold in the ozone!
Hoping Comet Lake will hit 5.5GHz 24/7 reliably, even if it needs water.

 

[lipe123]

Zen 2's die size is 74mm2. The 9900K's die size is 177mm2. Each zen 2 die houses 8 cores, equal to the number of cores the 9900K has. It looks to me like you have your math reversed. Intel is the one with die sizes more then double AMD's for the same core count. For it to be close you'd have to include AMD's IO die but that would simply be misleading as the IO die was purposefully made on a larger node (and thus larger) as the elements included on it do not benefit as much from scaling. That said, you stated "die size", not "combined die size" so I'm inclined to compare die to die. It should also be noted that Intel uses a monolithic design while AMD uses a chiplet based design. A chiplet based design is not as space efficient as monolithic but it is far easier to manufacture, cheaper, allows for higher core counts, allows for better binning, and various other benefits.

You are sorely mistaken if you think it's the node that's the main source of AMD's advantage. Their chiplet architecture and IO gains have been far more important. Intel simply moving to a smaller node won't be a magical fix for them. Intel needs a completely new architecture. Hopefully one that doesn't have a new major security hole every month.

No.
I'm talking about this: https://www.techcenturion.com/7nm-10nm-14nm-fabrication

 

[Gezzer]

AMD's current advantage is the Zen architecture is much less mature than Intel's Core architecture. I expect it to at least catch up with Core if not surpass it. Intel has tweaked Core for what, 10 years now? Who knows how much more they can squeeze out of it. So if everything's equal (which is not guaranteed by any means) AMD cpus should be kicking Intel's over the next 3-10 years.

What does that mean for the average end user? Not a lot. Performance leaps have been diminishing every year for CPUs as it is. Moore's law is breaking down as we try for ever smaller nodes. More importantly, except for a few edge cases no one really needs all the power a bleeding edge top tier CPU has to offer. Even gamers.

It really doesn't matter which camp you go with IMHO. Personally I think the top tier Ryzen 5 cpus are the best bang for buck and anything higher tier is just overkill where CPUs are concerned. This coming from someone that always bought the top tier i7's (pre Ryzen) when building a new system. But I just can't justify that any more.

 

[JimboJoneson]

So core for core, their about even,

Yup, completely even core for core - very accurate analysis -- even in frequency, like the article clearly states. If AMD keeps this up, they might be even able to catch up in mutli-threaded loads one day!

Might burn a hold in the ozone!

For sure!

Hoping Comet Lake will hit 5.5GHz 24/7 reliably, even if it needs water.

I heard comet lake will do 5.6GHz while just breathing softly on an air cooler. No fan needed. And that was under 10 core load of wPrime, while using just 65 watts! Its going to be amazing!

 

[wizardB]

That's an oversimplification and half truth.
Back in the "core days" the core 2 duo chips destroyed amd's athlons in benchmarks. It was not just marketing BS.
What we are seeing now is finally a reverse where AMD actually is beating intel with real performance and not just some hype.

The problem is that this isn't a level playing field, Intel tried to milk 14nm too long and AMD jumped to 7nm giving them a HUGE lead. When Intel eventually does catch up is going to be tough again for amd.
I mean right now intel is still within 90% of AMD's performance with dies that's double the size.

I think he meant P4 days and not core days, the P4s had lots of speed but couldn't keep up with AMDs processors in the workload end.

 

[Boilerhog146]

We have officially reverted to was it the Core days when sIntel had CPUs running at a massive frequency advantage over AMD CPUs but, nonetheless, had no performance advantage. sIntels marketing shtick in those days was frequency, frequency, frequency. It was just as lame as "glue".

I think your referring to the P4 /Athlon Days when the athlons w,ere named as a comparison to what intel speeds were .a 3500 + for example ran faster than a 3.5 gig P4 ahh netburst .yeah the AMD chips were crazy expensive .and the P4 was reasonable.capitalism at its finest!

 

[Irata]

That's an oversimplification and half truth.
Back in the "core days" the core 2 duo chips destroyed amd's athlons in benchmarks. It was not just marketing BS.
What we are seeing now is finally a reverse where AMD actually is beating intel with real performance and not just some hype.

The problem is that this isn't a level playing field, Intel tried to milk 14nm too long and AMD jumped to 7nm giving them a HUGE lead. When Intel eventually does catch up is going to be tough again for amd.
I mean right now intel is still within 90% of AMD's performance with dies that's double the size.

You are forgetting the "pre Core" days.

It's also worth mentioning that this is probably the first time AMD has had a process lead over Intel - sometimes the latter were two nodes ahead. This must be a totally new situation for them.

 

[Irata]

I think your referring to the P4 /Athlon Days when the athlons w,ere named as a comparison to what intel speeds were .a 3500 + for example ran faster than a 3.5 gig P4 ahh netburst .yeah the AMD chips were crazy expensive .and the P4 was reasonable.capitalism at its finest!

I think your referring to the P4 /Athlon Days when the athlons w,ere named as a comparison to what intel speeds were .a 3500 + for example ran faster than a 3.5 gig P4 ahh netburst .yeah the AMD chips were crazy expensive .and the P4 was reasonable.capitalism at its finest!

I think your referring to the P4 /Athlon Days when the athlons w,ere named as a comparison to what intel speeds were .a 3500 + for example ran faster than a 3.5 gig P4 ahh netburst .yeah the AMD chips were crazy expensive .and the P4 was reasonable.capitalism at its finest!

Found an old review from 2003 quoting the following prices (end of August):

Processor Price (US$)
AMD Athlon XP 2500+ 87
AMD Athlon XP 2600+ 93
AMD Athlon XP 2800+ 180
AMD Athlon XP 3000+ 265
AMD Athlon XP 3200+ 464

Intel Pentium 4 2.4C 178
Intel Pentium 4 2.6C 218
Intel Pentium 4 2.8C 278
Intel Pentium 4 3.0 417
Intel Pentium 4 3.2 637

Earlier on, a Pentium 4 also needed very expensive RDRAM in order to get good performance. DDR RAM did help bring platform cost down.

Here's another price chart from January 2002:

Processor Price per 1000*
Intel Pentium 4/2200A US$ 562
Intel Pentium 4/2000A US$ 364
Intel Pentium 4/2000 US$ 342
AMD Athlon XP 2000+ US$ 340
AMD Athlon XP 1900+ US$ 234
AMD Athlon XP 1800+ US$ 174

 

[amstech]

Intel has tweaked Core for what, 10 years now? Who knows how much more they can squeeze out of it. So if everything's equal (which is not guaranteed by any means) AMD cpus should be kicking Intel's over the next 3-10 years.

I'd be highly surprised if AMD held this edge for more then 12-18 months. Ryzen's new, absolute best CPU is barely edging out Intel's architecture from, lets face it, many years ago.
When you compare the 8/16 3700X/3800X to the 8/16 9900K, they are about equal across the board, with the Intel winning some benchmarks, the AMD winning most, overall all of them real close and then the Intel absolutely destroying them in gaming.
From Bloomfield, then Sandy Bridge, all the way to Coffee Lake its really been just improvements to the same architecture, or similar architecture, with big leaps here and there, like 3D transistors in Ivy Bridge. Not saying performance hasn't increased or that there hasn't been architectural changes/improvements, but nothing completely new. A 4770K still compares well to a 7700K in most real world scenarios.
This is the first time AMD's CPU's have actually challenged Intel's stuff, and beat it in tests. It's impressive but it was either this or AMD going out of business, because their discrete GPU's only take up 20% of the market and about 5% of Steam gamers, nobody uses that junk.
It's no secret how AMD is getting all this performance with their design and IPC, but its only a matter of time before Intel comes out with something better. Not trying to rain on AMD's parade, Ryzen is awesome, but core for core, its only marginally better or equal to Intel's stuff, and lets face it, they've been sitting on their hands. They aren't sitting on their hands anymore.

 

[JimboJoneson]

Cool. Let's do an IPC analysis on this article's results. I'll break it down for anyone still learning about IPC and how its calculated.

IPC is "instructions per clock" - basically, how many instructions can be done in a single clock cycle. So we defer clock speeds until the final calculation, if there are differences there.

We also need to know the differences in cores and threads and the amount of work done in a given time. In this case the score was for all intents and purposes, identical - so the amount work done (instructions), and time completed was exactly the same between the two. This is great for easy math.

The cores and thread count is also identical - this will also help make the math easy.

So considering the cores, threads and performance, was all exactly the same, the only thing left to do is look at clocks and the CPU that ran with lower clocks clearly has the IPC advantage. You can't argue straight up math.

So the clocks we have is 5.955 for the i9 HEDT part and 5.625 for the mainstream desktop part. When we calculate that out to get the percentage difference we land at 5.7% difference.

Due to everything else being equal, the Ryzen desktop part calculates to a 5.7% advantage in IPC over the HEDT part. That said, AMD's SMT is quite a bit better than Intel's hyperthreading, this will come into play on the results depending on the workload.

There are other factors that go into IPC calculations, like cache latency during a specific workload, or as I mentioned, the efficiency of SMT, so each workload will vary a bit in IPC calculations.


If we use Cinebench R20 as our "workload" for example, when we match clock for clock, cores for core, and thread for thread, you will see 3700x Ryzen have over a 12% advantage in IPC over 9900k (quite different from the mere 5.7% calculated from the above article's results). Here the only differing factor is amount of time taken to do an identical workload, so we can simple calculate the % differences like we did with the clock speeds above.

So it really does depend on the specific workload being calculated on. (actual math was done on results from a Techspot article comparing 9900k and 3700x both clocked to 4ghz, image below)

You can also calculate from the below, that in this workload, AMD has increased IPC by over 16% in a single refresh on the same platform. Makes you wonder if they can do that again with Zen3 in 2020? Sounds like a stretch but it will be exciting to watch how that unfolds.

Image source: https://www.techspot.com/article/1876-4ghz-ryzen-3rd-gen-vs-core-i9/

 

[wiyosaya]

That's an oversimplification and half truth.
Back in the "core days" the core 2 duo chips destroyed amd's athlons in benchmarks. It was not just marketing BS.
What we are seeing now is finally a reverse where AMD actually is beating intel with real performance and not just some hype.

The problem is that this isn't a level playing field, Intel tried to milk 14nm too long and AMD jumped to 7nm giving them a HUGE lead. When Intel eventually does catch up is going to be tough again for amd.
I mean right now intel is still within 90% of AMD's performance with dies that's double the size.

I said nothing about Core 2. That's your take, not mine.

 

[wiyosaya]

You are forgetting the "pre Core" days.

It's also worth mentioning that this is probably the first time AMD has had a process lead over Intel - sometimes the latter were two nodes ahead. This must be a totally new situation for them.

AMD destroyed sIntel in the "Core" days - the "Core" generation before "Core 2".

EDIT: While this article is about Core 2 performance vs Athlon 64 FX, note that it states that AMD had the performance lead - https://www.tomshardware.com/reviews/benchmarks-conroe-fx-62,1263.html

 

[wiyosaya]

I'd be highly surprised if AMD held this edge for more then 12-18 months. Ryzen's new, absolute best CPU is barely edging out Intel's architecture from, lets face it, many years ago.

One factor that, IMO, will weigh heavily in this is that Intel's current CEO is a business/finance guy much like AMD's previous CEO Rory Reed. AMD's current CEO is firmly grounded in technology.

 

[JimboJoneson]

One factor that, IMO, will weigh heavily in this is that Intel's current CEO is a business/finance guy much like AMD's previous CEO Rory Reed. AMD's current CEO is firmly grounded in technology.

I find financially oriented people are really more about smoke and mirrors and deflection than anything. Its because they aren't firmly grounded in anything except financials - and lets face it - high level financials is all about making things look like they are not in order to gain competitive advantage.

I think Intel's myriad of recent marketing faux pas reflect this mindset exactly. All these stunts and headlines that seem to be upsetting and ticking off enthusiasts who know better, is just collateral damage of Intel's desire to persuade investors who generally know little about things on the ground.

For example, Intel claiming their list of "real world" programs that are used on desktop, was actually their own data collected from their laptop and tablet users exclusively. Not one desktop was included in the source for that list - and Intel marketed this at presentations and for their shareholders. Its not until some enthusiast gets a hold of the slides and reads the fine print that the lies get exposed. Intel doesn't care because investors don't read that technical fine print, and generally don't read enthusiast tech sites / videos.

Throwing enthusiasts under the bus has had a large play in why in a few short years Gamers Nexus affiliate sales have gone from 95/5 percenatage of Intel to AMD sales to 5/95 for AMD. Again, Intel could care less because in their eyes, the investors are the customer, and their actual customers are just seen as the milk cow from which to squeeze money.

The latest Intel roadmap slide showing they will be on 1.5nm within 9 years is again, just smoke and mirrors to keep investors happy. Maybe they can hit that, but maybe they should just try to get 10nm out after half a decade of trying, first, before trying to convince us of 1.5nm? Again its a slide for investors "benefits" not ours.


Lisa Su on the other hand is an electrical engineer by trade and worked as an actual engineer for Texas Instruments, Freescale and IBM. At IBM she worked her way through management and their new CEO at the time, who had no semiconductor experience, picked her to be his "right hand man (er ... woman)" to help bridge the gap between him and the company - she was appointed technical assistant to the CEO. She learned about how a CEO should operate and how to decision make, from him.

So the two CEOs (Bob and Lisa) are pretty much polar opposites in how they got to where they are - Bob was always a finance guy, was CFO at eBay before Intel, Lisa worked her way up through businesses from engineer to CEO. This obviously is a huge factor in her success with AMD.

Part of the reason that AMDs SMT is magically so much better than Intel's HT is that she used her connections at IBM to help design the SMT needs for Ryzen. IBM is no stranger to SMT offering up to 4 or more threads per core on some of their CPUs. Its little things like this that give AMD their current advantages here and there.

Intel admitted that they wouldn't be able to to properly respond to AMD in server and desktop markets until 2021 or 2022. So no one needs to speculate that much, Intel's own leaked guidance back near the beginning of the year stated this. Granted, things can change ...

In the meantime, continue to expect them to produce some great looking slides that impress people who don't know better.

Comet lake won't be clocking to 5.5 on air with a 10 core load no matter how much hoping one does. 10nm won't be hitting 5ghz either, and will never be appropriate for high end desktop use. Intel needs to get to 7nm EUV and they need an entirely new architecture ASAP.

Also, I think they need to start abandoning the 5.0ghz marketing ploy -- its quite likely that no smaller node than 14nm will hit 5.0 and still be able to greatly improve on IPC - there's a tradeoff. Intel only reached 5.0 because 10nm was a failure, and that gave them the time needed to continually improve 14nm production, minor refresh after minor refresh after minor refresh for that specific node. Moving to a new node, like 10nm, starts that process all over again ... If none of their smaller nodes than 14nm can hit 5.0ghz, that will be seen as a failure in the eyes of those who have been convinced (by Intel's own marketing) that Intel's superiority lies only in that magical 5.0 number. They are starting to get caught in their own "catch-22".

As IBM found out in the 80s or 90s, when your ship is that big, if you find yourself going off course, getting back on course takes a lot of time.

I think sometime in 2021 their socks will be pulled up again, but the other variable is how fast will AMD be able to move within that time frame? Its not like they're taking their time with advancements - they'll be on 5nm EUV by 2021 according to their timeline, which they have been keeping extremely well.

We'll have to wait and see this play out in the next couple years.


Edit: just a note to add ...

Some of you might be aware that Intel CEO recently stated to investors that "the company will no longer waste resources, nor have interest in trying to keep a 90% market share in CPUs, and that they will be diversifying into other TAMs as that's where the "real" money is" (paraphrased).

This is "CFO" speak to imply this: "While we acknowledge that AMD will be eating hard into our market share in the coming years, please note that this is not because AMD is getting better or building better products or that the industry will want to shift towards their products, but rather you can be confident that as this happens, its because we chose to let it happen so we could diversify into other markets and make even more money, so don't worry about your stocks - they are safe with us!"

-- its a smoke and mirrors tactic to deflect from the reality of losses in market share caused from a strong reemergence of AMD, again, for investors benefits. Avoiding any direct comparison is the goal here ... like the 10th gen HEDT 6 hour pull back of NDA lift so it wouldn't be compared vs Threadripper ... and the lack of a 16 core HEDT part that if they produced it, would ultimately have to be compared directly to the R9 3950 desktop part. (That strategy failed, as the 3950x is generally a better performer than the 18 core in HEDT workloads even, which ended up being the comparison many reviewers went with)

They can't use these illusional tactics forever, and its already starting to hurt them as intelligent people start to notice it.

 

[Gezzer]

I'd be highly surprised if AMD held this edge for more then 12-18 months. Ryzen's new, absolute best CPU is barely edging out Intel's architecture from, lets face it, many years ago.
When you compare the 8/16 3700X/3800X to the 8/16 9900K, they are about equal across the board, with the Intel winning some benchmarks, the AMD winning most, overall all of them real close and then the Intel absolutely destroying them in gaming.
From Bloomfield, then Sandy Bridge, all the way to Coffee Lake its really been just improvements to the same architecture, or similar architecture, with big leaps here and there, like 3D transistors in Ivy Bridge. Not saying performance hasn't increased or that there hasn't been architectural changes/improvements, but nothing completely new. A 4770K still compares well to a 7700K in most real world scenarios.
This is the first time AMD's CPU's have actually challenged Intel's stuff, and beat it in tests. It's impressive but it was either this or AMD going out of business, because their discrete GPU's only take up 20% of the market and about 5% of Steam gamers, nobody uses that junk.
It's no secret how AMD is getting all this performance with their design and IPC, but its only a matter of time before Intel comes out with something better. Not trying to rain on AMD's parade, Ryzen is awesome, but core for core, its only marginally better or equal to Intel's stuff, and lets face it, they've been sitting on their hands. They aren't sitting on their hands anymore.

Sure, but the only way I see Intel coming out with something better is with a brand new artitecture, and I don't think they really can at this point. Like I said Intel has been tweaking the Core architecture for over 10 years now, how much more can they squeeze out of it? And if they did have a better one in the wings wouldn't they be stating as much?
I personally feel that the next performance leap will be quantum, whenever they get that working. Until then Moore's law will keep becoming more and more elusive due to node shrinks becoming every increasingly hard to pull off. So the only other area to improve? Tweaking the architecture.
I'm not saying that Zen will dominate Intel, hard to say on that one. But the very least I expect to see is Zen pull up to be neck and neck with Core, if for no other reason then the fact that they should have more room to tweak the design, in theory at least.