GeForce GTX 1080 Ti: Core i7 7700K vs. Ryzen 7 1800X Linux Gaming Performance

[torsionbar28]

Price-wise, the 1800x is comparible to the 7700k. IT is an entirely appropriate article if you are looking to build a Linux gaming machine and aren't sure whether to splash the cash on intel or AMD

Nonsense, AMD has never released competing CPU's that were priced the same as their intel counterparts. You must be fairly new to this. AMD has always been priced significantly lower than the comparable-performance intel part. It was true in 1996 when I built my first PC using an AMD K5, and it's true today. The 1800x does not compete with 7700k in the market. It competes with 6900k. Don't take my word for it - compare the 3 using a 16-thread benchmark, and see for yourself. AMD competes with intel by offering same or better performance at a lower price, and always has.

[torsionbar28]

However there's still plenty of bugs and quick-and-dirty implementations to be worked out of BIOS:es so performance could improve quite a lot without application developers having to do anything.

IMO a larger performance gain will come with the widespread availability of DDR4-3200 later this year. Micron is ramping up production as we speak.

[L_A_G]

IMO a larger performance gain will come with the widespread availability of DDR4-3200 later this year. Micron is ramping up production as we speak.

Seeing how most motherboards don't support it (some don't support anything beyond 2166 MHz) and those who do suffer from performance limiting bugs I'm still going to say BIOS updates are going to be the bigger boon for performance.

[pixo]

Keep in mind that ryzens inter core connect as well as cache bandwidth is tied to the memory frequency! All the reviewer's got like the fastest possible RAM which still runs fine with Zen's current memory controller!
in the end, CPU performance might be limited by cache bandwidth, as the execution pipelines are starved of work... Memory clock frequency has a much bigger impact on the Zen platform than on the Intel platform..
Normaly you won't run into memory bandwidth saturation in games (Arma 3 is a edgecase) but L3 Cache BW is saturated pretty quickly if core scheduling is bad...

It is not cache, that is tied to core and L3 is tied to highest clocked core.
What is tied to memory speed is Infinity Fabric, memory controller and I/O Hub.
Biggest problem is IF as that becomes bottleneck. CCX, I/O and Memory content for it.
Its 32B/cycle can be as low as 22GB/s when using 1600 DDR4 modules. And the latency is bad at 800Mhz on which it run with such DDR4.

[dungeon]

So GamersNexus nailed it with title like "An i5 in gaming, i7 in production" isn't it ... in production is like $1000 Intel, while in gaming sounds more like $250 something

Some guys inspected AdoredTV's claim how BD is now faster than SB and found that opposite is actually true...

On average nothing happened, same true like 5 years ago

[Shevchen]

Yes @dungeon, but do you try to disregard AdoredTVs findings or could it be that both have done a good job and both are right? GN said, that he has to write a review for "now" and not a review for "something in half a year" and he is correct with that, but also mentioned, that quite some horsepower is not used yet. So nothing new here. Gaming performance lacks because of missing optimizations (Both GN and AdoredTV say this) and AdoredTV goes a little deeper into the "why", while GN showed some parts that "its not" (like higher frequencies on the RAM, which doesn't have any immediate effect on gaming-performance RIGHT NOW)

This means: That once all primary bottlenecks are solved and the gaming performance starts to ramp up into expected levels, we will find another stack of "minor performance improvements" and this may as well include DRAM-frequencies, as the throughput bottleneck is right now a lesser problem and is not yet hit, because another bottleneck is so grave, that it prevents the throughput to be even saturated - in this case scheduler/core-hopping.

Its not as simple as "AMD sucks".

edit: What I have learned from those benchmarks, is that software-development in games is heavily dependent on optimizations, while other software is not, and just distributes the load to all cores and "thats it". It also shows me, that the gaming-industry is in need of coders that understand and implement those optimizations in a meaningful way and that a year-long 0815-approach may lead to a bias, that kills their own performance and thus potential of their games. The compute-benchmarks should also suffer from the "core-hopping" and bad BIOS-versions and from being unoptimized, but they don't. You can extract information from this fact also.

[dungeon]

Yes @dungeon, but do you try to disregard AdoredTVs findings or could it be that both have done a good job and both are right?

Well, AdoredTV findings are different than here HU get ... they get virtually the same results on average like 5 years ago, so i don't want to give people false hope on average

I am fine if someone buy Ryzen while knowing that currently it is somewhat just like as GN said "an i5 in gaming, i7 in production" but it might (hope not, as hope never dies but still might not happen to degree someone might hope up to) also stay mostly like that in next couple years too, as speaking about future who knows that other than to guess it... that future is not still and not yet, up to the facts

[Shevchen]

I hear ya and thats GN stance on it - they want to present facts and stand for them. They are accurate in their testing-setup and go deeper than some other reviewers (like bigger outlets, that have a standard approach without checking much on "behind the scenes") So GN is correct for the Ryzen-performance right now and if you take this as an argument to buy/not buy a Ryzen, thats fine by me.

But as I buy hardware that shall last for several years, I also take the possible future into consideration and there, the only thing I can make is an educated best guess. Educated by researching the current problems, evaluating if they are solvable and will be solved (here your history comes into play, when you say "Bulldozer was not optimized for 5 years, why should Ryzen?) and also try to take some guesses around technology that may not be available but is in planning, like Infinite Fabric.

Infinite Fabric for example is something I can only see hard facts on, when Vega is out. So I can either speculate what IF might do, or wait for Vega and a combined benchmark, that takes advantage of IF.

So to summarize, I make a buying decision on that educated guess as well as actual performance. If my guess is wrong, my bad - I tried my best and failed, meaning that I need to either educate myself more or was "tricked by false promises and/or lazy devs". In general, my assumptions were never far off my experiences when I buyed hardware, so if I would buy a Ryzen now, I wouldn't have a bad feeling in my guts about it. But again: Thats my personal point of view and I would not force those views on other people - still, I can share them.

[efikkan]

Basically Ryzen still has Bios Issues, missing optimizations and so on.

Optimizations of what?

Ryzen is simply an inferior architecture in terms of gaming performance.

Look at the single thread results here: http://www.anandtech.com/show/11170/...0x-and-1700/22
Ryzen's IPC is comparable to Intel's. But given the clock speed difference, single core performance isn't. It's in the same ballpark, but lagging. Though, not by the margin Michael has found in many games...

It really depends on what you're measuring, consider the following:
- Intel: 3 execution ports, each with ALU and FPU. Very advanced prefetcher.
- AMD: 4 ALU ports + 2 FPU ports. More simple prefetcher.
So it really depends on what kind of workload you are measuring. AMD is arguably almost "twice as superscalar", so under specific workloads they can scale much better.

Not all workloads will not scale with a more superscalar CPU, and the prefetcher also matters in terms of being able to feed the execution ports. The efficiency of branch prediction and precaching can also hurt the scaling.

Gaming is a kind of workload that's not superscalar friendly at all, suffering greatly from branch mispredictions and cache misses. The efficiency of the prefetcher matters a lot here, which is the reason why Ryzen is crushed in gaming. There is no lack of "optimization" for AMD here, this is a simple fact that unoptimized code runs better on Intel's architecture. And it can't be easily fixed either, games would have to be completely rewritten with less bloat.

Right. 20% perf difference could be explained by the clock difference (i.e. 7700k max turbo is >20% higher than 1800x's), but the rest seems like result of nvidia driver differences between this and Michaels last Zen tests.

First of all, gains in gaming performance is almost flat over 4.0 GHz with Intel, since the games are usually not bottlenecked any more. Secondly, 1800X with XFR boost way beyond 4.0 GHz if cooled properly.

Obviously there's a lot of optimizations to be had with the nvidia driver.

What?

[dungeon]

Zen means balance - "here i am faster than you expected and there i am slower than you expected"