News AMD's Ryzen 9 5950X Shatters PassMark Records

[CerianK]

For those confused about the single-thread performance improvement in PassMark, keep in mind that AMD had worked directly with PassMark to identify a performance deficit with Zen 2 which was related to a Windows 10 update and the new version of PassMark released early this year. Thus AMD had access to source code to identify the issue. The solution was for PassMark to change the PRNG (pseudo random number generator) in order to divorce undocumented Windows 10 code from creating a problem. Subsequently a new version of PassMark was released that was more favorable to AMD. I made a point to David over at PassMark that both the old and new PRNGs used actually are not very good in terms of randomness and performance.

Therefore, it is not inconceivable that AMD used knowledge of PassMark to improve Zen 3, but modifications would have had to occur within the last 6 months. However, decent improvements are trickling out in other benchmarks, which would make the case that Zen 3 is simply better under all circumstances. Whether or not PassMark is truly an exaggerated case, based on the limited data so far, remains to be seen.

 

[helper800]

Me: 30% better IPC than Intel seems unlikely
You: No, that's where they should based on these unverified/unsourced numbers I'm using
Me: No, those numbers don't look accurate, I know it is slower in games for example
You: You don't know what you're talking about, you're wrong
Me: No, here are actual test results
You: Those numbers don't count because I don't want them to, no counter source provided

The obfuscation and denial of real world results just doesn't stop. The world is flat everyone. If all you got is personal attacks against me, then you got nothing. Barring something completely unexpected, the 5000 series is going to sweep pretty much, if not, all IPC tests here. Those results will count to me, just like the current ones do, and I bet they'll suddenly count for you too. We'll see how many of you put asterisks next to AMD's winning results.

I did not resort to ad hominem. I am denying nothing... I am excusing your lack of knowledge on how games are not unbiased metrics regarding CPU performance. Either way, please check out the Gamer's Nexus video on the 3000 series review for reference, as well as, the anandtech article jeremyj_83 linked to above our conversation. I am being completely candor, please do not read into my posts as derogatory, as they are not.

 

[Endymio]

I did not resort to ad hominem. I am denying nothing... I am excusing your lack of knowledge on how games are not unbiased metrics regarding CPU performance.

When calculating IPC, there is no such thing as an unbiased metric, game or no. The particular instruction mix affects the results more than a bribed FIFA referee at the World Cup.

 

[helper800]

When calculating IPC, there is no such thing as an unbiased metric, game or no. The particular instruction mix affects the results more than a bribed FIFA referee at the World Cup.

I wouldn't go that far. Most synthetic's are unbiased, they just hit the cores in different ways. Games, however, are very biased and introduce even more variables to consider.

 

[GenericUser]

Hey, what's a few possessive "it's" errors between friends?

Side rant, but that's a tradeoff I would take if it means I would never have to see the word "lose" constantly misspelled as "loose". I want to drive nails into my skull with how ubiquitous that seems to have become.

 

[Endymio]

I wouldn't go that far. Most synthetic's are unbiased, they just hit the cores in different ways. Games, however, are very biased and introduce even more variables to consider.

When I use the word bias, I'm not implying intentional bias. All benchmarks, including synthetics, are biased simply by virtue of the fact that there is no one "right" mix of instructions. Two different synthetics can yield wildly different results: which one is correct? One is biased one way, the other another.

Side rant, but that's a tradeoff I would take if it means I would never have to see the word "lose" constantly misspelled as "loose". I want to drive nails into my skull with how ubiquitous that seems to have become.

From one left-brain word fetishist to another:

 

[Pitbull Tyson]

How important is passmark. Real world and AI and threaded optimizations. Every app uses cpu usage which in turn uses a bunch of your cores however a single thread app will just raise the usage percentage on your CPU usage until it reach 100 percent then what ? That is it? Obviously if your using the right app it supports SMT or HT and you know single core is not that important what is important how the cores work together to produce the 3D or ArtWork or PS or Premiere etc. Ill throw a dual socket xeon 26 processors with 1TB DDR4 RAM. What will perform better in photoshop a so called single threaded app which is not true as all Adobe apps are HT apps. Highly optimized for Intel but AMD will do as well but what AMD doesn't tell you all these companies go to Intel because they know the business and pros will all use Intel Xeons for servers and HEDT processors for workstations.🙈

 

[JBB_123]

What are you doing? You're not allowed to say anything that even resembles negative about AMD here unless you want to be labeled a paid Intel shill.

Funny comment. I was interested in building a Ryzen 3000 series system when I saw the first reports come in about single threaded performance. But as time went by I continued to watch them slip down the table as more and more samples came in, so I decided to wait. I just wanted to make sure people are fully informed before making a purchase. I will build a Ryzen system at some point, but just waiting to see what the true score is after more samples come in.

 

[Juan_Bijero]

From the reviews that I have seen today on Hardware Unboxed, Hardware Canucks, etc., AMD really knocked it out of the park with the 5000 series processors. That said, I'm not spending $800 on a new CPU. But good for them - even if the 5900X actually beats 5950X in some benchmarks.