OcUK Ryzen 2000 series review thread

[chrcoluk]

Both arma3 and farcry 5 can scale past 4 cores easily, this myth that they got stuck at 4 threads, is because people are using them on 4c/8t cpus and they wont utilise logical cores only physical cores.

On a XEON I seen arm3 run on it easily utilises all 20 cores.

 

[chrcoluk]

Wait,wut?? ARMA III has a crap ancient engine which shows very poor thread scaling but is very popular. FarCry5 shows crap scaling over 4 threads due to its use of the Dunia engine,and is very popular too and loves clockspeed. Look on Steam. Both top 20 games.

Creation uses one primary thread,followed by one which is less used,and then shows crappy scaling to a further 4 threads,because Bethesda somehow forced the engine to do so,which is an improvement over Skyrim. Its based on the Gamebryo engine used in Morrowind in 2002.

Saw how your Core i5 8600K and a Core i7 7700K matched the Core i7 8700K?? That tells you how well threaded it is.
.

You cannot use logical cores to determine how well something is threaded. Sadly a lot of people have been conned by HTT, thinking games will use it, most games will not.

So e.g.

if Game A seems to hit a performance bottleneck so say a 6600k matches a 7700k so 4c/4t matching 4c/8t people have assumed its limited by 4 threads right?
Then suddenly you pop in a 6 core chip and its faster, whats going on? The answer is there is now 2 extra real cores so there is more actual threads to utilise. Which is why we see performance jumps from 6600k/7600k to 8600k/8700k..

Its been claimed as another example arma3 and fc5 are capped to 4 threads, this assumption is probably because people are running 4c/8t chips, but both games can utilise way more than 4 cores.

You need "real" not "logical" cores.

Now in some data we see chips like a 7700k outperform a 7600k or a 8700k outperform a 8600k, what is happening here then?

Sadly these graphs often dont show the clock speeds, but i7's have higher stock clocks than i5's they also also have a larger cache. It may be even if the chips are overclocked that the tester managed to get a higher overclock on their i7 vs i5, or it could simply be that logical threads are helping, logical threads can help in some games but if it does its only a small amount.

Of all the games I have analysed, I have never seen a game that only utilises max 4 threads, usually its 1 thread for everything, or 1 thread for rendering and a 2nd thread for everything else, or a really high thread limit. I have tested games on 20 core chips.

 

[humbug]

Both arma3 and farcry 5 can scale past 4 cores easily, this myth that they got stuck at 4 threads, is because people are using them on 4c/8t cpus and they wont utilise logical cores only physical cores.

On a XEON I seen arm3 run on it easily utilises all 20 cores.

When people say one core / thread that's not actually the CPU, the game engine has one call thread. you don't necessarily see that manifest its self as one core usage on the CPU, the game engine could be doing lots of thing across lots of threads but older less sophisticated game engines have less of that parallelization so more of it may still depend on a single thread that can become saturated.

 

[chrcoluk]

And what's the ratio of those overclocked PCs to stock of the shelf HPs and Dells? It's low single digit percentages. For every Alienware PC Dell sells, Dell and HP sell 50 Inspirons/Pavilions. Even HPs own Omen gaming PC's aren't overclocked, and they sell plenty of those.

Off all the 8700k's out there, less than 5% of them (from checking Steam surveys, ORB results etc) are overclocked. I don't know why you expect main stream reviewers to cater to the 5% versus the 95%.

You need to understand (and maybe accept) that overclocking is still very much a niche in PC gaming.

I accept there will be people who might get say a K series chip for their birthday, and just run it how it is out the box.

But that sort of person wont be watching these reviews either.

I got a friend who doesnt manually overclock hasnt a clue how to do it, he brought a gaming PC. That PC was tuned by the seller, voltages etc. like s*** and OCUK do on their gaming prebuilt rigs.
Most prebuilt rigs I see for sale are not K series parts, when I do find a K series rig, most I find are overclocked by the seller.

It is a "niche" but most people also dont run K series chips.

What we seeing here is a deliberate withold of data on the graphs, thats it really, its very simple. The same happened on the AMD vega reviews. By all means put stock performance on there, so people like yourself who think thats the way to go have that data, but the OC data should also be there. I dont see why you have an issue with this, what reason do you personally would want to see the data witheld? that you havent explained.

 

[chrcoluk]

When people say one core / thread that's not actually the CPU, the game engine has one call thread. you don't necessarily see that manifest its self as one core usage on the CPU, the game engine could be doing lots of thing across lots of threads but older less sophisticated game engines have less of that parallelization so more of it may still depend on a single thread that can become saturated.

Possible yes, but what is been claimed here is that games like FC5 and arma3 do not scale their performance above 4 core chips, that is definitely not true.

What you are talking about is OS cpu scheduling, so e.g. Microsoft Windows can make a single processing thread seem spread cross all cpu cores. Because it rapidly moves the thread from one core to the next. This can give an illusion something is more threaded than it actually is. There is ways to debug this tho to see the actual amount of threads.

There is also the problem if the CPU is not bottlenecking in any way so you will see a performance ceiling, so then a weak chip gives the same result as a powerful one.

 

[humbug]

Possible yes, but what is been claimed here is that games like FC5 and arma3 do not scale their performance above 4 core chips, that is definitely not true.

What you are talking about is OS cpu scheduling, so e.g. Microsoft Windows can make a single processing thread seem spread cross all cpu cores. Because it rapidly moves the thread from one core to the next. This can give an illusion something is more threaded than it actually is. There is ways to debug this tho to see the actual amount of threads.

It is true, they are both build on very old engines, Duna Engine (Farcry) is a split from Cryengine 2 (Crysis 1 - Warhead)

 

[B1gbeard]

Testing games has some 2700x vs 1800x.

 

[chrcoluk]

It is true, they are both build on very old engines, Duna Engine (Farcry) is a split from Cryengine 2 (Crysis 1 - Warhead)

I have software from 2005 that can utilise 16 threads.

 

[humbug]

I have software from 2005 that can utilise 16 threads.

You mean game engines? i use Cryengine 5, it too can use 16 threads, for all sorts of things like Physics and post processing lighting, NPC's ecte..... add a lot of Draw Calls and one thread gets saturated pretty quickly, the performance is then bottlnecked by that saturated thread.

 

[Scougar]

Testing games has some 2700x vs 1800x.

Nice incremental gains. Nothing earth shattering.

Some of those benches seem very inconsistent however (GTA for one). In certain scenes, the 2700X is registering as SLOWER than the 1800X. This seems to be more down to the dynamic nature of some of those runs I assume (rather than anything they stripped out or reduced from the cpu).

 

[chrcoluk]

You mean game engines? i use Cryengine 5, it too can use 16 threads, for all sorts of things like Physics and post processing lighting, NPC's ecte..... add a lot of Draw Calls and one thread gets saturated pretty quickly, the performance is then bottlnecked by that saturated thread.

When I tested arma3, there was about 80% performance improvement from 4 to 8 cores, which is excellent considering context switching overheads, up to 16 cores it was more difficult as we couldnt keep the cpu as a bottleneck easily, but we applied load to the cores so basically the cores were not fully idle outside of the game and achieved similar levels of performance scaling.

FC5 has had a lot less testing, but the chip is only running at 1.6ghz and is outperforming a 6 core 4.9ghz chip which means its scaling to at least 18 cores fairly well.

 

[CAT-THE-FIFTH]

All the games in that list will load up 8 cores. A 8 core loading game = thread heavy. GTA5, project cars, FC5 and BF1 have actually been tested on 24 core parts and loaded up all 24 cores.

There is not a single JRPG, not a single indie game.

You cannot use logical cores to determine how well something is threaded. Sadly a lot of people have been conned by HTT, thinking games will use it, most games will not.

DF has made comparisons:

https://www.eurogamer.net/articles/digitalfoundry-2017-intel-kaby-lake-core-i7-7700k-review

Some games where the Core i7 and Core i5 are at the same clockspeeds,and the Core i7 is slightly ahead might be down to the cache,but in the other situations where its a much larger difference it isn't.

I doubt its always down to cache only as IIRC,in the past its been shown to not always really have as much of an impact in all games:

https://wccftech.com/intel-amd-l3-cache-gaming-benchmarks/

Look at the difference between 6MB and 8MB L3 cache. If you look at the thread utilisation of the games where the Core i7 is pushing ahead a great deal,ie,you see all 8 threads being used.

Another example is the Haswell based 2C/2T Pentium G3258 against the equivalent 2C/4T Haswell Core i3 CPUs running at lower clockspeed:

https://pclab.pl/art57691-3.html

The Core i3 4100 series had the same amount of L3 cache.Look at how a 4.7GHZ G3258 does against a 3.5GHZ Core i3 4150.

Remember,review systems are most likely clean OS installs,not like a normal gaming rig,which probably has a whole lot of other stuff like AV,etc running too.

The reason why those games are gaining with HT is probably more down to those games being probably developed partly with consoles in mind too,and those have 8 weak cores,so its probably more a balancing issue.

So HT does seem to help out in some games - sure actual cores will be more useful than HT in those games,where it does help,but I do think a CPU with 6 to 12 reasonably fast threads is what I would be targeting in a new system build if you want the CPU to last for the next few years,especially if you are getting a reasonably fast graphics card.

Sure,a 2C/4T or 4C/4T CPUs still have their place,but really for more budget systems IMHO. Considering 6C/6T and 6C/12T CPUs are now going as low as £140 to £150,its not really even the case you are saving much money going for a 4C/4T CPU now.

The main issue why less people bought Core i7 CPUs was because the Intel HT tax was ridiculous,and they even blocked the cheapo Xeon E3 CPUs on consumer motherboards from Skylake onwards where you could get a locked 4C/8T CPU for as little as £170ish.

However,thats the thing the 4C/4T Xeon E3 CPUs also had 8MB cache like their 4C/8T counterparts. I used a few.

So e.g.

if Game A seems to hit a performance bottleneck so say a 6600k matches a 7700k so 4c/4t matching 4c/8t people have assumed its limited by 4 threads right?
Then suddenly you pop in a 6 core chip and its faster, whats going on? The answer is there is now 2 extra real cores so there is more actual threads to utilise. Which is why we see performance jumps from 6600k/7600k to 8600k/8700k..

Its been claimed as another example arma3 and fc5 are capped to 4 threads, this assumption is probably because people are running 4c/8t chips, but both games can utilise way more than 4 cores.

You need "real" not "logical" cores.

Now in some data we see chips like a 7700k outperform a 7600k or a 8700k outperform a 8600k, what is happening here then?

Sadly these graphs often dont show the clock speeds, but i7's have higher stock clocks than i5's they also also have a larger cache. It may be even if the chips are overclocked that the tester managed to get a higher overclock on their i7 vs i5, or it could simply be that logical threads are helping, logical threads can help in some games but if it does its only a small amount.

Of all the games I have analysed, I have never seen a game that only utilises max 4 threads, usually its 1 thread for everything, or 1 thread for rendering and a 2nd thread for everything else, or a really high thread limit. I have tested games on 20 core chips.

Wut?? Gamebryo engine is developed in the 1990s,and Creation is a branch of it - its never been well threaded,and it shows your lack of knowledge of Bethesda game,saying its well threaded.

I actually have played 1000+ hours of Fallout 4,Skyrim,etc on multiple types of processors,have loads of friends who played it on loads of CPUs,and anyone who has played it would now how badly it scales with threads. Its primarily limited by one to two threads,and always has,and more over yes ARMA III does not scale well with more cores.

These are all based on old engines.

Here,lets look at ARMA III:

https://pclab.pl/zdjecia/artykuly/mbrzostek/2017/cfl_s/wykresy/nv_arma3.png
https://pclab.pl/zdjecia/artykuly/radek/2014/pentium_g3258/charts/arma3_1920n.png
https://www.hardware.fr/getgraphimg.php?id=223&n=14
https://i.embed.ly/1/image?url=https://i.imgur.com/MSi4Imj.jpg?fb&key=522baf40bd3911e08d854040d3dc5c07

None of that supports what you are saying. None of the 6C and 8C HEDT CPUs show any scaling.

Look at FO4:

https://i.imgur.com/6KXlHIV.png
https://cdn.sweclockers.com/artikel/diagram/14241?key=349d9720f637f7bc86523c9db8088f78
https://cdn.sweclockers.com/artikel/diagram/13584?key=f5badbfb7e350a00a625439c44b8eab8

Diamond City is an NPC rich area. It scales to 8 cores - no it doesn't.

Look at the HEDT CPUs.

Go and play the game and start building settlements,and going into NPC rich areas,and look how it works.

Its primarily limited by two cores,and has more limited scaling upto 4 more threads.

Most of the FO4 game testing is not intensive,and any one who actually has played the game,would know settlements,etc where there are large NPC concentrations,etc are what bogs down performance.

This can be easily tested by using the console to spawn in more and more NPCs,and then you see what your CPU usage profile is like.

Don't go onto any forum where people play the game,mod,etc and start trying to say it scales to 8 cores - that includes modders who know the engine much better than most people.

 

[chrcoluk]

cat the fifth engines can be modified and updated, and also a game doesnt necessarily just use the engine alone in its raw format without any additional coding on top.

Skyrim hasnt been tested on any of these sites for ryzen 2.
Fallout 4 likewise

 

[chrcoluk]

arrgh this is annoying lost my typed post, soyou lost detailed reply sorry

in short fallout 4 and skyrim conveniently omitted from ryzen reviews so wont comment on those

the arma3 results are interesting, but I have done my own testing as well and it is offering better performance with more real cores. The difference with my tests I am doing it all on the same cpu, ram, motherboard, same components, nothing been swapped out. If Ireduce cores, I simply disable the cores in software.

Bear in mind I have never disputed that with spectre+meltdown patched at similar clocks a ryzen 2 can come close to a coffee lake cpu. I have not disputed that at all on here in any threads.

 

[humbug]

When I tested arma3, there was about 80% performance improvement from 4 to 8 cores, which is excellent considering context switching overheads, up to 16 cores it was more difficult as we couldnt keep the cpu as a bottleneck easily, but we applied load to the cores so basically the cores were not fully idle outside of the game and achieved similar levels of performance scaling.

FC5 has had a lot less testing, but the chip is only running at 1.6ghz and is outperforming a 6 core 4.9ghz chip which means its scaling to at least 18 cores fairly well.

An old Cryengine 3.8 mess-about of mine...

Just as an example, i have turned most of everything that might use Parallel workloads off and cranked the draw calls right up into the troposphere.

Notice the performance is very bad, core utilisation is ignoring the last 4 threads but no one thread is under excessive load, the GPU is completely strangled?

This is an example of where an engine can be both very highly parallelized but also still be hampered by other workloads that are not.

 

[CAT-THE-FIFTH]

cat the fifth engines can be modified and updated, and also a game doesnt necessarily just use the engine alone in its raw format without any additional coding on top.

Skyrim hasnt been tested on any of these sites for ryzen 2.
Fallout 4 likewise

The problem as I said repeatedly a lot of testing is BS in the game,it needs to be tested in settlements and areas of the city with large NPC concentrations. The testing games one,was better than average,but there are more taxing areas like near court 35,etc,but is still better than some of the meh testing I have seen elsewhere. I would say that Ryzen 5 2600X,with HT disabled and similar clockspeeds to the Ryzen 7 2700X will probably show a similar result.

It seems most of the people testing FO4 are clueless about the game,and its bottlenecks. Only HardOCP tested the game in settlements.

I have had a few years to explore performance myself,and in mates rigs,and read as much as I can as possible as I could. I know where it bogs down,when its unmodded,modded,etc as its easy to do experiments using the console. Bethesda Game Studios(not Bethesda the publisher)has a history of not bothering updating games post launch in any meaningful way when it comes to hardware.

The only time I can remember in recent history was with the original Skyrim in 2011. It used X87 based instructions(in 2011),which meant worse performance than the SSE type which Intel/AMD said people should use and had meh CPU performance. The community had to step in and sort it out themselves,and eventually like a year or so later,I think out of shear embarrassment fixed it.

The games are fun,but the community really is what saves the day for them.

I don't want people to think an 8C CPU is useful for it,since it really does not scale. Like I said it really only scales to six threads,and needs two strongish threads,and decent memory bandwidth(its one of the few games which seems to scale well with silly speed RAM).

Those two threads can be whacked the most especially with large amount of NPC spawns,etc,such they become the limiting factor,and if you mod it,things can get worse.

Starfield apparently will use the last hurrah of the engine and if that scales well and evenly across 8 cores I will do a victory dance and probably hug anyone in the vicinity!

 

[chrcoluk]

Yeah, by the way I know I appear really passionate, but I kind of need to confess I find these AMD discussions golden for getting the posts in on the forum

I am curious now if amd have got microcode updates out for ryzen 2 so I have requested in the spectre/meltdown thread if any of our ryzen 2 users can check on inspectre if its mitigated already on windows 10, that will give an indication if ryzen 2 performance figures include spectre kernel mitigation or not.

 

[humbug]

Yeah, by the way I know I appear really passionate, but I kind of need to confess I find these AMD discussions golden for getting the posts in on the forum

I am curious now if amd have got microcode updates out for ryzen 2 so I have requested in the spectre/meltdown thread if any of our ryzen 2 users can check on inspectre if its mitigated already on windows 10, that will give an indication if ryzen 2 performance figures include spectre kernel mitigation or not.

They are, check my post count

 

[DragonQ]

Has Computerbase's review been discussed yet? They paired the R7 2700X with 3466 MHz DDR4 with pretty tight timings and it had a nice response in their gaming tests. Ryzen 2 clearly loves RAM speed and tight timings as much as its predecessor.

 

[humbug]

Interesting, TechSpot made a clock for clock comparison, it seems Ryzen 2 has about a 7% gain in IPC in games like BF1, about 3% in Cinebench.




https://www.techspot.com/review/1613-amd-ryzen-2700x-2600x/page5.html