Can AMD RIP the Core i9? - Ryzen Threadripper

[Emily Young]

AMD's finally back in the HEDT space with Threadripper - But can they go toe to toe with Team Blue?

 

There is a table showing the Threadripper 1900X as having a 4-core boost clock of 3.7 GHz despite having a base clock of 3.8 GHz. This is an error, and should be 4.0 GHz. Its price should also be $549. Further details are unknown at this time. Sorry about that.

 

 

Buy the Threadripper 1950X:
Amazon: http://geni.us/WuK1aQ
Newegg: http://geni.us/NYAPoe8

 

Buy the Threadripper 1920X:
Amazon: http://geni.us/LIO7Va
Newegg: http://geni.us/tN78M

 

Buy the ASUS ROG Zenith Extreme:
Amazon: http://geni.us/LOSaF
Newegg: http://geni.us/rb4bo6

[Nicnac]

yay I guess!

[Guest]

Apparently it can .

 

But not in CS:GO xp.

[kerradeph]

Cool. AMD's putting pressure across the board. Good to see it really heating up now. 

[Max_Settings]

TBH I’m not impressed. Yes threadripper is faster than the 7900K in most applications, and yes I’m still going to recommend it. But the the gains that it had in rendering applications were not like an massive crushing defeat where it should have been considering it has double the cores. Once again AMD’s poorer IPC compared to Intel is showing.

[Guest]

What about the EPYC cpu's, they have 32 cores and 64 threads. I understand that they are server chips, but they would outperform intels core i9 lineup when it comes to multi core workloads.

[Alpha Shadow]

Now what would be really amazing for team blue is if they develop a processor that can prosses data faster and energy efficient.

[Mr.Dingle]

So , Hardware Unboxed with the same CPUs tested shows the 1950x using 20-30W less power than the 7900X , Gamer Nexus shows roughly the same difference , yet LTT shows the 1950X/1920X using more power ? Can someone clarify this 

[Hunter259]

Are you guys serious? You tested CPU's at 4K. Something that is almost 100% GPU bottlenecked and can get the same performance on low end NAHALEM from 2009. Give me a break.

[Lathlaer]

16 minutes ago, Hunter259 said:

Are you guys serious? You tested CPU's at 4K. Something that is almost 100% GPU bottlenecked and can get the same performance on low end NAHALEM from 2009. Give me a break.

 

Agreed. RIP methodology. Slap another 1080ti there for SLI at least so you can be less bottlenecked in 4k and as a bonus you'd be able to see that PCI Lanes actually do matter and x16/x16 is better than x8/x8.

[Hunter259]

1 minute ago, Lathlaer said:

 

Agreed. RIP methodology. Slap another 1080ti there for SLI at least so you can be less bottlenecked in 4k and as a bonus you'd be able to see that PCI Lanes actually do matter and x16/x16 is better than x8/x8.

It doesn't. 

 

http://www.gamersnexus.net/guides/2488-pci-e-3-x8-vs-x16-performance-impact-on-gpus

[Lathlaer]

1. It literally says so in the article "we are not sure how this scales with SLI"

 

2. Watch this:

 

 

 

and this:

 

 

Ofc. I am not the maker of those vids but I have seen them and I have talked to the author. I am willing to learn so if you could point out errors in methodology there, I would be grateful.

 

 

[Hunter259]

Just now, Lathlaer said:

-snip-

I'll trust  someone with a much larger track record than some russian youtuber.

 

SLI : https://www.pugetsystems.com/labs/articles/Titan-X-Performance-PCI-E-3-0-x8-vs-x16-851/

 

Again. It doesn't matter. They had to go to 4k Surround AND run the lowest settings possible to make it have any serviceable difference. 

[Lathlaer]

1. Polish, not Russian. One of the people behind Extreme Hardware is a guy whose nick is Xtreme Addict and if you'd known a bit about the overclocking community you'd known that he is top tier overclocker (or used to be, I think he is retired now or smth). This is as far as credentials go. Also, they buy their own stuff, so no way is anything skewed towards any company.

 

2. That second link doesn't show any game performance.

 

I will even quote from the article:

 

"Whether you will see lower performance with x8 versus x16 is going to highly depend on the application. Some may see a difference, others won't."

 

So I am waiting for those SLI in games performance results that refute what I posted.

 

If you could show actual game performance in more than 1 game like I did that shows different results, I'd be grateful. 

[Hunter259]

2 minutes ago, Lathlaer said:

-snip-

Uh. There were two games tested with single and dual GPU's tested....

[tom_w141]

1 hour ago, Max_Settings said:

it has double the cores.

2 x 10=16?

 

I think you failed math

[tom_w141]

46 minutes ago, Lathlaer said:

 

Agreed. RIP methodology. Slap another 1080ti there for SLI at least so you can be less bottlenecked in 4k and as a bonus you'd be able to see that PCI Lanes actually do matter and x16/x16 is better than x8/x8.

 

44 minutes ago, Hunter259 said:

It doesn't. 

 

http://www.gamersnexus.net/guides/2488-pci-e-3-x8-vs-x16-performance-impact-on-gpus

GN has a recent video showing x16x16 makes fuck all difference. Can't link from this PC

[tom_w141]

1 hour ago, ColonelThunder said:

So , Hardware Unboxed with the same CPUs tested shows the 1950x using 20-30W less power than the 7900X , Gamer Nexus shows roughly the same difference , yet LTT shows the 1950X/1920X using more power ? Can someone clarify this 

Their reviews are 25-30 mins and have paper articles to accompany them. Now compare that to the 8 min LTT flash in the pan that contains 4k testing and less depth that the kiddie pool. Who do you think is right?

[Mr.Dingle]

23 minutes ago, tom_w141 said:

Their reviews are 25-30 mins and have paper articles to accompany them. Now compare that to the 8 min LTT flash in the pan that contains 4k testing and less depth that the kiddie pool. Who do you think is right?

Well , the answer is already obvious , im asking because im curious what could cause 20W +/- results , or more . Cant be the CPU alone ? 

[tom_w141]

1 minute ago, ColonelThunder said:

Well , the answer is already obvious , im asking because im curious what could cause 20W +/- results , or more . Cant be the CPU alone ? 

Weak methodology, low quality

[Alpha297]

Can someone explain why the value chart is curved for Intel but straight for AMD?

In the vid at 6:29:

 

https://youtu.be/9voQqU73-Mg?t=6m29s

 

Edit:  Or just tell me how to read that graph in general?  Both lines are Intel with one curving and one flat?  

[Emily Young]

2 hours ago, ColonelThunder said:

So , Hardware Unboxed with the same CPUs tested shows the 1950x using 20-30W less power than the 7900X , Gamer Nexus shows roughly the same difference , yet LTT shows the 1950X/1920X using more power ? Can someone clarify this 

Gamers Nexus recorded the power draw at the 12V line to the CPU, while ours was measured at the wall. That means they have a reading that shows the CPU's power draw alone, while ours shows the whole system. Considering otherwise identical hardware, the difference in power draw from idle and CPU-only synthetic load should be accurate.

1 hour ago, Hunter259 said:

Are you guys serious? You tested CPU's at 4K. Something that is almost 100% GPU bottlenecked and can get the same performance on low end NAHALEM from 2009. Give me a break.

RotTR and DXMD testing these days is to compare DX11 and DX12 performance deltas, which is something that's tough to convey, but pay attention to the 97th percentiles, because they tell the bigger story, even at 4K.

 

Normally, we would also have included For Honor and Ghost Recon: Wildlands at 1080p and 4K as well, but issues with our UPlay accounts stymied us. To compensate, we ran more productivity benches.

 

In scenarios like this, we SHOULD have went with 1080p, but we'd already completed 4K testing at that time. Time constraints dictated that we couldn't go back and redo them, so here we are.

8 minutes ago, Alpha297 said:

Can someone explain why the value chart is curved for Intel but straight for AMD?

In the vid at 6:29:

 

https://youtu.be/9voQqU73-Mg?t=6m29s

 

Edit:  Or just tell me how to read that graph in general?  Both lines are Intel with one curving and one flat?  

The curved line is 1920X vs 7900X, while the flat line is the 1950X vs 7900X. Basically, since they're the same price, the value proposition never changes, making the 1950X always a better value overall than the 7900X in our testing. As for gaming, it's basically never a better value from our testing, but as mentioned before.

[Streetguru]

So what was your actual RAM speed and timings? Did you use Samsung B-die? assuming TR loves it as much as Ryzen. Did you test using UMA or NUMA memory settings?

I get that yall maybe just want to put out a quick under 10 minute on everything, but for a product like this I think you should strive for more info and much deeper dives.

Did you go into Game mode for your gaming tests? Gamers Nexus showed a decent gap using it at times especially in 1% and .1% frame times since windows seems to be jumping around all of the threads not knowing what to do, not that it would matter much at 4k.

[Alpha297]

4 minutes ago, GabenJr said:

The curved line is 1920X vs 7900X, while the flat line is the 1950X vs 7900X. Basically, since they're the same price, the value proposition never changes, making the 1950X always a better value overall than the 7900X in our testing. As for gaming, it's basically never a better value from our testing, but as mentioned before.

I appreciate the reply, but I'm still totally confused by the Overall Value graph and the Gaming Value graph a little bit later in the video.  I don't see how the "value" is being calculated in these charts.  The Gaming chart even says Threadripper 1920x is technically a better value at a total system cost of $100 USD, but considering it costs more than that on it's own, I don't see how a value at that amount can be technically obtained.  

 

Did I miss the specific methodology for these value calculations, or are these summations based on the previous charts that are mathemagically compiled into one line for each product?   

[Emily Young]

7 minutes ago, Alpha297 said:

I appreciate the reply, but I'm still totally confused by the Overall Value graph and the Gaming Value graph a little bit later in the video.  I don't see how the "value" is being calculated in these charts.  The Gaming chart even says Threadripper 1920x is technically a better value at a total system cost of $100 USD, but considering it costs more than that on it's own, I don't see how a value at that amount can be technically obtained.  

 

Did I miss the specific methodology for these value calculations, or are these summations based on the previous charts that are mathemagically compiled into one line for each product?   

It's being calculated based on the weighted results across all testing (gaming specifically average FPS and 3DMark+Superposition). It's based on the methods @Enderman and @patthehat pointed out in the Core i9 correction thread. The technically comment is just calling out the oddity there at <$100.