AMD Ryzen Quad-Core 2+2 versus 4+0 Core Setups Analyzed

Agent-A01

2017-03-27 16:34

So the poor gaming performance is due to another issue. Maybe a game engine with 'proper' optimization will bring some improvements, if not poor game performance will continue on until Ryzen 2.

#5412962

Toss3

2017-03-27 16:46

Would be nice to know what is causing the bad performance in games if it isn't the CCX interconnect. AMD should get on this asap (or maybe they know it cannot be solved and are thus staying quite).

#5412971

rl66

2017-03-27 16:55

So the poor gaming performance is due to another issue. Maybe a game engine with 'proper' optimization will bring some improvements, if not poor game performance will continue on until Ryzen 2.

So exept in bench ryzen is not an Intel killer (with on top of that, price not so low as anounced).and with Intel mainstream with over 4 core (yay at last ) with new socket in a few. so it's a 50% success (i am 50% happy for them but was expecting more).

#5412973

rl66

2017-03-27 17:01

7700K is 4.8GHz and ryzen 4.0GHz so add another 20% for ryzen and compare clock to clock and situation changes in favor of ryzen so i think amd has done good job with ryzen.

clock to clock is stupid as they haven't the same reference architecture. (it's like comparing a pentium D 955 VS a core 2 duo clock to clock). even with difference of clock they should be compared stock clock.

#5412977

Dazz

2017-03-27 17:08

Well memory bandwidth seems to increase performance quite a bit, kinda reminds me of the P4 netburst where high latency cache prefers high bandwidth memory, same for the AMD FX chips. However not sure if this is memory bandwidth or the fact the bus speed has increased from 100MHz to 123MHz increasing PCI-e throughput significantly which much like the AMD FX overclocking the HT yielded very good performance gains. Since all traffic is going through this 4x PCIe 3.0 connection 4GB/sec each direction graphics cards start taking a noise drive at this speed and have other stuff connected to it i think that extra 23MHz = 23% more data traffic will help more than the memory it's self. Good luck getting to 3600MHz memory need Samsung chips to do that. https://www.youtube.com/watch?v=RZS2XHcQdqA#t=0.151099

#5412979

isidore

2017-03-27 17:16

Since when is 7700k 4.8ghz?

Since it's saying in the screens. He is referring to the speed they run the benches.

#5412984

Camaxide

2017-03-27 17:22

Since when is 7700k 4.8ghz?

Did you even look at the test..?

#5412996

Kaarme

2017-03-27 17:52

7700K is 4.8GHz and ryzen 4.0GHz so add another 20% for ryzen and compare clock to clock and situation changes in favor of ryzen so i think amd has done good job with ryzen.

Let's talk about clock to clock when AMD and GloFo manage to get the clocks up. Until that time, no Kaby Lake owner is going to downclock their expensive CPU.

so it's a 50% success (i am 50% happy for them but was expecting more).

100% successful at forcing Intel to finally consider 6-core mainstream.

#5413003

edilsonj

2017-03-27 18:07

Core i3 @ 4.8GHz Core i5 @ 4.8GHz Core i7 @ 4.8GHz Overclocked Intel vs a stock 4GHz Ryzen (1800?). I really hate this.

#5413009

H83

2017-03-27 18:19

I wasn´t expecting this one. I thought having all the cores in one cluster would be much better than having the cores divided between 2 clusters but i was completly wrong.

#5413015

rl66

2017-03-27 18:42

L100% successful at forcing Intel to finally consider 6-core mainstream.

sadly it was on plan before Zen (now called Ryzen)... i'm still on my 50% let's see if time make it better

#5413020

Silva

2017-03-27 18:48

Or maybe the problem is ryzen overclocks like a dog? Let's just oc both to the max: 7700k @ 5Ghz vs Ryzen @ 4.1Ghz

Let's blame a new 8 core chip for things it can't do while we praise an overpriced 4 core chip because it clocks slightly better costing an arm and a leg doing so. Why don't you compare OC with Intel own 8 core and add price into consideration? Oh right, Intel Fanboys don't care about the price factor cuz AMD is trash.

#5413026

Amx85

2017-03-27 18:59

Well memory bandwidth seems to increase performance quite a bit, kinda reminds me of the P4 netburst where high latency cache prefers high bandwidth memory, same for the AMD FX chips. However not sure if this is memory bandwidth or the fact the bus speed has increased from 100MHz to 123MHz increasing PCI-e throughput significantly which much like the AMD FX overclocking the HT yielded very good performance gains. Since all traffic is going through this 4x PCIe 3.0 connection 4GB/sec each direction graphics cards start taking a noise drive at this speed and have other stuff connected to it i think that extra 23MHz = 23% more data traffic will help more than the memory it's self. Good luck getting to 3600MHz memory need Samsung chips to do that. https://www.youtube.com/watch?v=RZS2XHcQdqA#t=0.151099

The hexacores looks the better option :infinity: offtopic: i checked that video, RyZen @3600MHz RAM is getting equal/bigger performance @4GHz vs Kabylake @5GHz, looks great no? :pc1:

#5413027

Amx85

2017-03-27 19:04

I wasn´t expecting this one. I thought having all the cores in one cluster would be much better than having the cores divided between 2 clusters but i was completly wrong.

i think higher lv3 cache amount make a difference, 2+2 gets 16MB lv3 and 4+0 gets 8MB

#5413029

Amx85

2017-03-27 19:07

Funny comments everywhere. When Bulldozer came, Everyone was all around IPC metric. Now IPC is dirty word. Yes, Ryzen does not play as good as Kaby in many games. But it still plays damn well. And it is something to keep for quite a few years. + All those spare threads... Some people here know no shame.

Hahaha, Bulldozer problem was core utilization amount, even today most of software are limited under 4 threads and old instruction sets; under SSE4.1 ot SSE4A(if supported), just some games with highly AI trends over 4c but i see still limited between 6 and 8 threads

#5413038

Kaarme

2017-03-27 19:25

sadly it was on plan before Zen (now called Ryzen)... i'm still on my 50% let's see if time make it better

After a decade of Intel 4-core max mainstreams, I'm not going to believe that. AMD's new architecture was talked for a long time, and it's not like anybody sane would have expected it to max at 4 cores.

Funny comments everywhere. When Bulldozer came, Everyone was all around IPC metric. Now IPC is dirty word. Yes, Ryzen does not play as good as Kaby in many games.

Kaby Lake is nothing but a Skylake overclock (plus lolptane), so it's meaningless to simply match it IPC wise if you can't get anywhere near the clocks. Now, I'm not saying Ryzen would be bad. It's what I'd be buying if my Intel system wasn't only a year old. I'd have supported anything but Intel's bottomless greed if I had had a choice.

#5413059

The Reeferman

2017-03-27 20:44

Last time I looked in the Ryzen OC forum thread I saw one eating an OCed I7 7700k for breakfast. It did need very fast ram speed to achieve that but it won, with wide margin, in GTA5, BF1 and others. Somewhere in this thread around page 23, can find it this quick. http://forums.guru3d.com/showthread.php?t=412876&page=22

#5413065

-Tj-

2017-03-27 21:04

Seems 4+0 is just the same as 2+2 which both are worse than 4+4 or 3+3 if there even is a difference. Would mean the interconnect isn't the issue after all. To me that conclusion is actually positive.

Yeah, I think most games that have a large gap is mostly because it favors intel HT or intel's way in general. Anyone remember those intel logos in the past? Fallout4, newer FC series, Hitman, F1, Dirt Rally, RE5, LP2, Crysis1, 2, etc. are all intel branded and optimized for it, so something like FC primal or Fallout4, or even F1 2016 won't get any fixes anymore.. Those are last season now, it was like so in the past.. Unless AMD pays them extra so they add Ryzen optimization code in it.

#5413081

schmidtbag

2017-03-27 21:43

I'm a little confused - how do we know the "4+0" isn't just 4 cores without SMT, and how do we know "2+2" isn't just 2 cores with SMT? To me, the only way to definitively prove the performance of a single CCX is by testing only 1 core without SMT. Regardless, those benchmarks do imply something good: either the CCXs don't hinder performance, or, SMT scales VERY nicely.

Dunno what they changed in Dota2, but atleast there was quite big uprade atleast with game update (+25% roughly).

Yes, I have heard about this on the Phoronix forums. It seems when utilized properly, Ryzen is quite powerful.

#5413087

PrMinisterGR

2017-03-27 21:57

Does anyone have a benchmark on Dota 2 after the Ryzen update? The only thing I can find is a guy on the Chinese forum claiming he went from ~130 to ~160 fps. Also all this conversation about quads being a better choice for getting a CPU in 2017, reminds me of some equally surreal conversations in the graphics subforum, where people were insisting that the GTX 960 2GB was a better choice to the R9 380 4GB, because it was a bit faster on some titles, and "you can't use the extra memory on such a low-end GPU anyway". This whole argument kind of has this vibe. After seeing Jayz playing a bit more with his CPU after a BIOS update, this whole thing became an even greater no-brainer for me.