AMD Ryzen Memory Analysis: 20 Apps & 17 Games, up to 4K

[medi01]

nonsense Youtuber

Oh dear, give me a break.

He's grasping at straws

He is stating facts, comparing effect of AMD GPU vs nVidia GPU in CPU performance context.

Maybe it won't use a PLX bridge chip this time?

That's a mainboard manufacturer's choice.

 

[Vayra86]

Oh dear, give me a break.


He is stating facts, comparing effect of AMD GPU vs nVidia GPU in CPU performance context.


That's a mainboard manufacturer's choice.

'Facts' that change with every passing day as things get adapted towards Ryzen, which does not exclude Nvidia at all. Meanwhile, CPU tax in one game is different from another and he tested it while being GPU-limited in every single title, reducing the value of these different CPU loads to 'oh look my CPU is doing less' while it has no bearing on ingame FPS at all. Don't get me wrong, I like it that Ryzen gaming benches are starting to look better, but its very easy to see that the Ghz bottleneck isn't gone, this reality has not changed at all.

Thing is, for 60hz gaming, Ryzen is MORE than fine and it was more than fine at launch too. For 120hz/fps however, and on cpu-limited games, the reality has NOT changed, and the 7700k is still the go-to CPU. The net value of all these wonderful conclusions is quite precisely zero.

 

[uuuaaaaaa]

That's a mainboard manufacturer's choice.

Not really, I was talking about the PLX chips that connects both gpu's in a dua gpu card like the Pro Duo/ R9 295X2, HD 7990/6990/5970/4870X2. Since vega will support infinity fabric, they might get rid of the PLX and directly bridge both gpus.

 

[nemesis.ie]

We should really call them PCIe switches, PLX are not the only makers of themm I think Broadcom for one have some.

Maybe "PCIe-Sw" ot something shoter would be good though.

 

[Super XP]

Welp, what issues?
Perf difference (in games) between 4 + 0 and 2 + 2 is lower than 5%. How far lower should it be not to be considered an issue? In my humble opinion that's dayum good.

Zen 2 will show more consistent performance, according to anand. When tuning all that "branch prediction" logic, Intel has enough resources to do it using much wider range of apps than AMD. That's what Zen 2 (using AM4 socket, f*ck you intel) will do.


Makes me wonder about implication about dual chip "Vega Pro", connected using IF.

I wish I can locate that Link with AMD discussing about ZEN2 enhancements over ZEN1. And how they explain ZEN1 issues won't be present in ZEN2. Paraphrasing of course, let me look.

 

[nemesis.ie]

As an aside, the UEFI with the new AGESA/R5 support just arrived for my Asrock.

Installed but I've not played with it yet.

 

[Kanan]

I wish I can locate that Link with AMD discussing about ZEN2 enhancements over ZEN1. And how they explain ZEN1 issues won't be present in ZEN2. Paraphrasing of course, let me look.

I confirm this, Lisa Su said in an interview that they are *currently* working at fixing the biggest flaws of Ryzen for Ryzen II (or "Ryzen 7 2xxx"), and then going from there to the smaller kinks.

 

[anubis44]

I confirm this, Lisa Su said in an interview that they are *currently* working at fixing the biggest flaws of Ryzen for Ryzen II (or "Ryzen 7 2xxx"), and then going from there to the smaller kinks.

Just a semantic point, but I would find it hard to believe Lisa Su used the word 'flaw' in describing improvements in Ryzen II over Ryzen.

 

[medi01]

For 120hz/fps however, and on cpu-limited games, the reality has NOT changed, and the 7700k is still the go-to CPU.

You make it sound as if 7700k wins vs 8 cores (including Intel's) all the time. But that's not the case.

Paying 300$+ for a 4 core CPU in 2017 is outright wrong, in my humble opinion, with major consoles going 8 core (first 6 and now 7 are usable in games) and even Blizzard optimizing games for 6 cores.

 

[TheLaughingMan]

From what I saw in games for memory speed it was only two outcomes, it either made no difference or going from 2133 to 3200 gained about 10 FPS. That can be a big deal. For example in Fallout 4 at 1440p that pushed it form 57.4 to 66.6 avg. That is a 13.8% improvement. The same for Hitman and Civ IV. And the price of RAM is not going down any time soon. I am going to go for the higher clocks first and lower timing if I can afford it here soon.

 

[Vayra86]

You make it sound as if 7700k wins vs 8 cores (including Intel's) all the time. But that's not the case.

Paying 300$+ for a 4 core CPU in 2017 is outright wrong, in my humble opinion, with major consoles going 8 core (first 6 and now 7 are usable in games) and even Blizzard optimizing games for 6 cores.

Oh but I agree on that second half! The 4c/8t at that price is retarded. But still, to hit the highest min fps and especially above 60, you really do want the Ghz not the cores. The reason Ryzen is bottlenecking at high refresh is exactly because there is always that ONE game thread you can't divide across multiple cores, and that is exactly where the min. fps takes a hit. The higher averages are the result of multiple cores taking the other threads while the main thread hits a GPU wall before it hits the CPU wall. But the main thread is still limited by the 1-core perf.

Also keep in mind that the comparison to consoles is broken by design because the consoles aim for 30 or 60 fps targets, as HDTV's are generally built for 50/60hz. And because the consoles still use a Phenom II derivative core - an entirely different beast from Ryzen or Intel CPUs.

 

[Kanan]

Just a semantic point, but I would find it hard to believe Lisa Su used the word 'flaw' in describing improvements in Ryzen II over Ryzen.

Right, you got me there.

 

[Super XP]

I confirm this, Lisa Su said in an interview that they are *currently* working at fixing the biggest flaws of Ryzen for Ryzen II (or "Ryzen 7 2xxx"), and then going from there to the smaller kinks.

I don't recall anybody at AMD mentioning the word "Flaw". Though Lisa did mention they are hard at work ironing out minor kinks and stuff in ZEN1, and will include enhancements over ZEN1 into ZEN2.

I highly doubt she say Flaw in any relation to ZEN1. These chips are great. They just need optimizations, modifications and tweaking.

 

[TheLaughingMan]

I don't recall anybody at AMD mentioning the word "Flaw". Though Lisa did mention they are hard at work ironing out minor kinks and stuff in ZEN1, and will include enhancements over ZEN1 into ZEN2.

I highly doubt she say Flaw in any relation to ZEN1. These chips are great. They just need optimizations, modifications and tweaking.

The only "flaw" I can think of is I think the Infinity fabric needs to be DDR. It appears to be running at the Bclk rate of the memory controller. I don't see any reason it can't move data on the rising and fall edge like DDR memory. Everything else will come with time.

 

[Kanan]

I don't recall anybody at AMD mentioning the word "Flaw". Though Lisa did mention they are hard at work ironing out minor kinks and stuff in ZEN1, and will include enhancements over ZEN1 into ZEN2.

I highly doubt she say Flaw in any relation to ZEN1. These chips are great. They just need optimizations, modifications and tweaking.

Man I don't wanted to trigger your fanboyed-ness, calm down. Yes she didnt' use the word flaw, as someone else right before you already noted, but it's a flaw nonetheless, she just used a typical "political correct" word for it, rather than flaw. She essentially said it will get more optimized, doesn't change a thing.

@TheLaughingMan That and its low clocks! Two big flaws. Right now, Intel processors simply win in games because of higher clocks. Ryzen @ high clocked Ram isn't bottlenecked in games, but 4000 MHz simply isn't enough to win vs. 7700K overclocked (and a lot of reviewers are comparing like that).

 

[EarthDog]

Makes sense to compare stock for stock, same clocks (for ipc and h2h), as well as overclocked. Since ryzen 1700/1700x/1800x doesn't really overclock at all past xfr (except to bring all cores there), it loses out by 25% clockspeed.

 

[Super XP]

Man I don't wanted to trigger your fanboyed-ness, calm down. Yes she didnt' use the word flaw, as someone else right before you already noted, but it's a flaw nonetheless, she just used a typical "political correct" word for it, rather than flaw. She essentially said it will get more optimized, doesn't change a thing.

@TheLaughingMan That and its low clocks! Two big flaws. Right now, Intel processors simply win in games because of higher clocks. Ryzen @ high clocked Ram isn't bottlenecked in games, but 4000 MHz simply isn't enough to win vs. 7700K overclocked (and a lot of reviewers are comparing like that).

Lol, you can remove that fanboy comment, I ain't a fanboy of either company.

Have you seen the recent Gaming Benchmarks from various reviewers. Ryzen does well enough in Gaming today versus when they were 1st released.

 

[TheLaughingMan]

@TheLaughingMan That and its low clocks! Two big flaws. Right now, Intel processors simply win in games because of higher clocks. Ryzen @ high clocked Ram isn't bottlenecked in games, but 4000 MHz simply isn't enough to win vs. 7700K overclocked (and a lot of reviewers are comparing like that).

That chip was not the Ryzen 7 target. Nor was pure gaming. It is also the first generation of a brand new architecture filled with brand new tech. Clock speed will take a generation or two. That is not a flaw, but more or less the reality of any highly complex product.

 

[techtard]

Nice writeup. Thinking about jumping on a Ryzen setup, maybe a 6 core with slow ram for now and if they iron out all the bugs (and when prices drop) get some faster ram later.
I don't need a new setup, but I'm kinda bored with my 2500k and the skylake stuff isn't that appealing. I was hoping Intel would have released a mainstream 6 core i5 and a 6c/12t i7 by now.

 

[Aquinus]

Infinity Fabric Is in its infancy. I just find it quite innovative. What if that ran at the CPU speed instead of the IMC?

What if it was just HyperTransport and was able to run in its own clock domain with its own multiplier? I can see there being benefits to clocking different parts of the CPU at different speeds and having control of that. Considering communication between each complex and how important it can be. Being able to clock it higher under certain conditions or lower when it's not needed could offer power saving options that are a little more granular than they are now.

I personally find what AMD has produced to be fascinating and there are classes of applications (such as web servers utilizing non-blocking I/O,) that can realize the power of a multi-core system. My exploration has involved using between 75% and 95% of the 8 threads on my 3820. I need more cores to test how far this scales.

 

[Kanan]

That chip was not the Ryzen 7 target. Nor was pure gaming. It is also the first generation of a brand new architecture filled with brand new tech. Clock speed will take a generation or two. That is not a flaw, but more or less the reality of any highly complex product.

I know, I know. Still it's a flaw when a interconnect between CPU's is so narrow that you have to overclock the Ram to get better performance ("better" not saying "full"). The low clock speed is another flaw, low clocks isn't something acceptable these days, every other architecture (including FX) has high clocks out of the box or is overclockable to 4.5 - 5 GHz (eg. Core architecture, entire line). I'm sure both flaws are well known at AMD and are right now worked on.

 

[techtard]

~4.0 ghz is not bad, it can still keep up pretty good from the benches around the 'net. And smokes my FX that's running @ 5.0ghz. Maybe the clockspeed wars are finally over and we're solidly moving to the core wars.

Could be that this is AMDs version of the original i7, maybe we'll get a 'sandybridge' type r7 that hits 5.0 with Ryzen 2 when they get all the kinks worked out and the process matures.

 

[medi01]

But still, to hit the highest min fps and especially above 60, you really do want the Ghz not the cores.

Actually...
There is not that clearly explained "better min fps" effect of Ryzen CPUs, (mostly) vs Intel's 8 cores.
I'd thought it's, perhaps, that XFR thing, but "consensus is" (among some anonymous dudes on the internet, lolz) that likely the larger cache plays a role.

 

[Vayra86]

Actually...
There is not that clearly explained "better min fps" effect of Ryzen CPUs, (mostly) vs Intel's 8 cores.
I'd thought it's, perhaps, that XFR thing, but "consensus is" (among some anonymous dudes on the internet, lolz) that likely the larger cache plays a role.

Ryzen is a bit different but still hasn't got the performance to apply that statement as a general one (cache makes Ghz irrelevant) because its too much of a black box as of yet and performance is going all over the place (better or worse, engine specific etc etc), while high Ghz does apply as a general guarantee for better min. fps.

The eternal problem of only being able to choose one CPU is at play here

 

[TheLaughingMan]

The low clock speed is another flaw, low clocks isn't something acceptable these days, every other architecture (including FX) has high clocks out of the box or is overclockable to 4.5 - 5 GHz (eg. Core architecture, entire line).

You do realize that out of the box, Intel only has 3 chips with a clock speed higher than 4.0 GHz right? That would be the infamous 7700K and the recent i3 7350K and i3 7320. So yes it is not only 100% acceptable, its 99.9% of the market normal. 4.1 GHz being a brick wall for even the best binned chips is an issue (short of LN2 cooling) that needs to be addressed. Flaw it is not. They built a brand new chip, on a brand new architecture, with brand new tech, and old tech they have never used before. We are lucky then got to the 4.0 GHz. You clearly just want to call something a flaw so have at it man, but you are off base.