Ryzen-A Fail for Gamers?

[krumme]

BF1

Non of the situations displayed gives any significant load on the cpu. All outright light load.

I played with people today that complained their 6700k went under 60fps.
Operations 64 scar sector 4.
Happens often on 4c8t. In those operations maps.
Just doesnt fit the narrative.
I know their problem. What could i say. Get a ryzen or a 6900 or play conquest or some other maps.

 

[krumme]

I mean serious. Look at those video before you post them. Its a ghost town in a ghost world.
The most boring gaming ever.
Hello hello. Where can i find some friends?
Lol. I didnt even knew you could play bf1 like that. Boring.

Horserace. Wow. Zzzz.
At least someone should try to shoot him with an aa or use explosives to get a flying horse. That would be fun.

Nothing like people flying in the map after beeing blown up.

 

[Pookums]

Are you Trolling or what ??

The comparison was

1. Clock to Clock
2. Core/Threads to Core/Threads
3. Same RAM speed

And you calling it not fair ?? read again, it wasnt comparing the Core i7 7700K but the architecture (Kabylake).
It doesnt getting more fair than this.

I mentioned this awhile back when I asked stilt about memory controller and its half speeds, but the option was never tested.

Could someone who has 6800k, 6850k, 6900k, 6700k or 7700k and also has a ryzen please Compare across the board where the Uncore is equivalent(or as close to approximation) in both Zen and kabylake. Particular of interest, in Latency measurements, and in games?

For those who want to understand what I'm asking to be done. Lets assume the same 4ghz, and 2400 mhz CL15. If this is the case then on the Skylake/broadwell system you ALSO have to reduce Uncore down to 1200mhz(Yes its abhorrent I know) to match the ryzen. The problem im seeing with Ryzen tests is they use the same clocks and ram, but the broadwell/skylake have substantially faster memory controller speeds. It would also help if you could then do Majincries drawcall bench on both systems to compare.

This I suspect is the true bottleneck. NOT SMT. NOT scheduling, etc. I also suspect Ryzen will outperform skylake/broadwell IPC when this happens. Now, there is no solution to this issue if The Stilt is correct and a software alteration to these speeds is impossible. However, it would at least give some damn peace of mind rather than chasing in different directions all of which keep giving non-repeatable performance numbers across platforms. It would also explain the variation. Hardware Locking those speeds to a 1/2 ratio of MEMCLK I suspect will be the worst decision AMD ever made. I would just like to see it tested as a datapoint.

 

[Markfw]

BF1

There are some setting not the same In the tank scene, there are planes in the air that I don;t see in the other two. There are other more subtle differences also.

 

[Arachnotronic]

All CPUs are at 4Ghz.
btw 6700k stock never boost more than 4Ghz when uses more than 1 core.So 6700k is at stock and 3570k is i think also oc.
I have 6700k and it NEVER boost more than 4Ghz in games.So it is OC ryzen and oc 3570k vs stock 6700k

6700K runs at 4GHz when 2-4 cores are loaded, 4.2GHz only for a single core load which is actually pretty rare.

7700K does 4.4GHz with 2-4 cores loaded, 4.5GHz with a single core loaded.

 

[Head1985]

There are some setting not the same In the tank scene, there are planes in the air that I don;t see in the other two. There are other more subtle differences also.

Its from Multiplayer

 

[Markfw]

Its from Multiplayer

Wouldn;t that invalidate it as benchmark ?

 

[Head1985]

6700K runs at 4GHz when 2-4 cores are loaded, 4.2GHz only for a single core load which is actually pretty rare.

7700K does 4.4GHz with 2-4 cores loaded, 4.5GHz with a single core loaded.

Yeah 7700k is like haswell 4790k with 2-4cores but 100mhz more in 1core.

 

[Head1985]

Wouldn;t that invalidate it as benchmark ?

Nope its better because in MP there is cpu bottleneck as you can see 3570k is bellow 60fps is some scenes and 6700k is same scene have 110+fps.
Single player is not cpu heavy no point testing it.

 

[Markfw]

Nope its better because in MP there is cpu bottleneck as you can see 3570k is bellow 60fps is some scenes and 6700k is same scene have 110+fps.
Single player is not cpu heavy no point testing it.

OK, so the point is, the Ryzen does just fine ?

 

[Head1985]

OK, so the point is, the Ryzen does just fine ?

Yes with GTX1070.6700k sits on GPU bottleneck most of the time.Ryzen uses it around 90-99% and 3570k is just too slow there its huge bottleneck with 1070.
With 1080TI ryzen and 3570k will have +- same performance like with 1070.6700k better because its still gpu bottleneck with 1070.

 

[PotatoWithEarsOnSide]

So the best way to review the CPUs is to use essentially the next best GPU, and check to see whether the CPUs are strong enough to force the GPU to be the bottleneck...?
For that to make sense, why not just max out everything and hand the award out to the CPU that has the lowest utilisation figure?

 

[rsndetre]

OK, so the point is, the Ryzen does just fine ?

More like an overclocked to the limit Ryzen (not all chips can do 4GHz and very few 4.1GHz) can barely keep up with some Intel chips who still have lot of room to OC.
If i5-3570K was pushed toward 4.5GHz, it would have trade blows with Ryzen (as it can be already seen doing it at 4GHz in the other videos linked).


Seems to me, these Ryzen CPUs are good upgrades, for gaming, over older AMD cpus or non K Intel. But then ... are quite expensive and have cheaper alternatives from Intel.


I'll be waiting for r5s.

 

[Markfw]

More like an overclocked to the limit Ryzen (not all chips can do 4GHz and very few 4.1GHz) can barely keep up with some Intel chips who still have lot of room to OC.
If i5-3570K was pushed toward 4.5GHz, it would have trade blows with Ryzen (as it can be already seen doing it at 4GHz in the other videos linked).


Seems to me, these Ryzen CPUs are good upgrades, for gaming, over older AMD cpus or non K Intel. But then ... are quite expensive and have cheaper alternatives from Intel.


I'll be waiting for r5s.

Virtually all of the Ryzen 7 have gome to 4.0. As for the cheaper Intel, I think the point I would like to make is that The Ryzen is great at all but gaming (like the 8-10 core Intel) at way cheaper, BUT its FINE for gaming, as in not always the best.

So you would buy one not primarily as a gaming CPU, but if you do other things AND gaming, its probably the best buy,.

 

[unseenmorbidity]

I am buying it primarily for gaming, because it will be a better long term investment than a 7700k. I suspect this cpu will last 5 years or more, and then I will upgrade to the last am4 cpu.

 

[krumme]

The 3D mark firestrike physics results perfectly mirrors my experience for 60fps 0.1% min in bf1. As of today.

If anyone wants to know how your cpu performs in 4 years in a wide range of games look at the 3dmark score.

 

[rsndetre]

Virtually all of the Ryzen 7 have gome to 4.0.

Not what I have read until now. According to this site, that tested a batch of 1700s, only 4 out of ten could reach 4GHz. In fact, they don't recommend going for 3.8-3.9 GHz either, without a good MB.

As for the cheaper Intel, I think the point I would like to make is that The Ryzen is great at all but gaming (like the 8-10 core Intel) at way cheaper, BUT its FINE for gaming, as in not always the best.
So you would buy one not primarily as a gaming CPU, but if you do other things AND gaming, its probably the best buy,.

Ok, but we don't compare with that. Almost no one, outside few enthusiasts, will spend few thousand dollars/euros to game at 4k with 8/10c intel cpus. Since AMD insist on marketing Ryzen as gaming worthy, their competition are the Intel K chips. If 1700 was in the price range of i5-7600K, or even lower, yeah, then it was an argument.
Second, people are not doing the kind of work Ryzen is suited for, at home. Even encoding family videos, that takes like 0.01% of their usage time. Again, marketing enterprise cpus for home usage makes little sense.

I am buying it primarily for gaming, because it will be a better long term investment than a 7700k. I suspect this cpu will last 5 years or more, and then I will upgrade to the last am4 cpu.

A highly used argument but not so valid. In 4-5 years the current generation cpus should be outdated anyway. Yeah, maybe Ryzen will go from being 20% behind 7700K to being 10% in front of it, but both should be far from top dogs. If I'll buy Ryzen now, chances are that it will never be top in gaming while if I get the latest i7 K, I'll enjoy top gaming performance for an year.

Anyway, 7700K or 1700, personally I find hard to justify a buy like this.

 

[PotatoWithEarsOnSide]

Low resolution testing is apparently a drawcall test, and Ryzen has a cross-CCX penalty of 16-22% in drawcalls. Until the scheduling issue is rectified, these low resolution tests are unreliable/worthless.

 

[krumme]

I have a MSI log file now for a ryzen 3.8 2667 in bf1 60hz. Its pretty difficult to analyze, and i cant identity the mins, and i dont think its correct. I read somewhere bf1 is supposed to issue 10 threads, but as i can tell there is at the same time activity in 14 threads. So the data seems wrong.
The utility of the threads besides the main thread always stays below 70% or more like in the 40-50%.
A typical picture on operations, amiens with high load on mainthread is is
9, 95, 53,14, 45, 6, 30,16, 47, 0, 19, 42,33, 2,13,2
(total 426)
5,100, 55, 9, 52, 2, 41, 41, 3, 47, 0, 44, 3, 25, 0, 23
(total 450)
Mind you i didnt have dips below 60fps and its cpu capped at 60fps. (same picture if cpu capped at 60fps or not), the main thread seems to stay a good deal lower for most of the time, this is corner cases.
My guess would be that as BF issues threads now, a ryzen will be single thread limited above aprox 100fps 1%min, so naturally on 144 7700k is a good deal faster here.
In general it is crazy how much cpu load bf1 have for a game in mode operation - amiens 64. And so many threads its using for a dx11 game from 2016. No wonder i5 have trouble lifting this kind of load on 4 cores even for 60fps.
Will have to try dx12 again and surely the scheduling is very important. Its just incredible to see how many threads bf1 can use, and also to that degree - if the data is correct.
I dont have the skill to analyze the situation in this game, but it seems we can have either complicated netcode or drawcall situations, pure cpu physics limitations, single thread and for the upper end of fps memory subsystem. They might play different roles depending on the 0.1% min fps demand or 5% min demand.
MSI file:
https://sabercathost.com/8V57/HardwareMonitoring.hml#.WMVNyHXmVEk
(select 2 (amiens) and 3 (scar plus amiens))

 

[frozentundra123456]

Low resolution testing is apparently a drawcall test, and Ryzen has a cross-CCX penalty of 16-22% in drawcalls. Until the scheduling issue is rectified, these low resolution tests are unreliable/worthless.

nvm, wrong post

 

[frozentundra123456]

Not what I have read until now. According to this site, that tested a batch of 1700s, only 4 out of ten could reach 4GHz. In fact, they don't recommend going for 3.8-3.9 GHz either, without a good MB.



Ok, but we don't compare with that. Almost no one, outside few enthusiasts, will spend few thousand dollars/euros to game at 4k with 8/10c intel cpus. Since AMD insist on marketing Ryzen as gaming worthy, their competition are the Intel K chips. If 1700 was in the price range of i5-7600K, or even lower, yeah, then it was an argument.
Second, people are not doing the kind of work Ryzen is suited for, at home. Even encoding family videos, that takes like 0.01% of their usage time. Again, marketing enterprise cpus for home usage makes little sense.




A highly used argument but not so valid. In 4-5 years the current generation cpus should be outdated anyway. Yeah, maybe Ryzen will go from being 20% behind 7700K to being 10% in front of it, but both should be far from top dogs. If I'll buy Ryzen now, chances are that it will never be top in gaming while if I get the latest i7 K, I'll enjoy top gaming performance for an year.

Anyway, 7700K or 1700, personally I find hard to justify a buy like this.

Yea, that has always been the go-to agrument of AMD fans since Bulldozer: just wait, we will be better in x years, even if not right now. I still remember the "wait for windows 8" things will be much better then. When it actually came, the gains were only a few percent, and intel gained as much as AMD.

Truth is, nobody really knows, especially with the poor overclocking of Ryzen. Now there is some merit to that argument, but to make a blanket statement that it will be "better" is certainly not a proven fact. More accurate is that "indications are that it will gain performance relative to a quad core in at least some games", but nobody knows for sure or how much.

 

[Markfw]

Not what I have read until now. According to this site, that tested a batch of 1700s, only 4 out of ten could reach 4GHz. In fact, they don't recommend going for 3.8-3.9 GHz either, without a good MB.



Ok, but we don't compare with that. Almost no one, outside few enthusiasts, will spend few thousand dollars/euros to game at 4k with 8/10c intel cpus. Since AMD insist on marketing Ryzen as gaming worthy, their competition are the Intel K chips. If 1700 was in the price range of i5-7600K, or even lower, yeah, then it was an argument.
Second, people are not doing the kind of work Ryzen is suited for, at home. Even encoding family videos, that takes like 0.01% of their usage time. Again, marketing enterprise cpus for home usage makes little sense.




A highly used argument but not so valid. In 4-5 years the current generation cpus should be outdated anyway. Yeah, maybe Ryzen will go from being 20% behind 7700K to being 10% in front of it, but both should be far from top dogs. If I'll buy Ryzen now, chances are that it will never be top in gaming while if I get the latest i7 K, I'll enjoy top gaming performance for an year.

Anyway, 7700K or 1700, personally I find hard to justify a buy like this.

Please read what I said. Ryzen is great at all BUT gaming (meaning productivity) LIKE the Intel 8/10 core chips, but cheaper. And at gaming it does just fine.

 

[Wiidesire]

R7 1700 vs i5 3570K in Battlefield 1 Multiplayer

Thanks for sharing my video. This was really hard to benchmark, took me around 10 hours. You need to find a server (with a free slot) where one team doesn't dominate so you can stay between D-B-E and also you need to balance not staying too far away from the action which would invalidate the results all the while staying alive for two minutes to make a valid run (I did each configuration 5x 2 minutes).

Good video.
I would just note that
1. playing conquest 64 on st scar is typically a far lighter load than operations 64.
2. Secondly 1% min is imo way to much. You cant play bf1 and your framerate sucks 1% of the time.
Find your own limit. I think its very personal.
10 sec of 1 hr is for me the limit. So more like 0.1% is imo the right number to use.

1) Completely valid criticism. Benchmarking the 4th Stage of Operations 64 Players on St. Quentin Scar would have resulted in a higher CPU load but it would take even longer to benchmark correctly. That's why I chose the compromise with Conquest.

2) As you said it's a personal preference, in my opinion 1% Low is the perfect metric to show how smooth the game feels like.

 

[krumme]

Thanks for sharing my video. This was really hard to benchmark, took me around 10 hours. You need to find a server (with a free slot) where one team doesn't dominate so you can stay between D-B-E and also you need to balance not staying too far away from the action which would invalidate the results all the while staying alive for two minutes to make a valid run (I did each configuration 5x 2 minutes).

1) Completely valid criticism. Benchmarking the 4th Stage of Operations 64 Players on St. Quentin Scar would have resulted in a higher CPU load but it would take even longer to benchmark correctly. That's why I chose the compromise with Conquest.

2) As you said it's a personal preference, in my opinion 1% Low is the perfect metric to show how smooth the game feels like.

Its very nice you take the time to do some real world testing.
Its insane how much we need that.
Much appriciated.

For the reference for others: What happens in operations vs conquest is that more people cluster together because you have 64 man for 2 or 3 flags. Rushing.

What is interesting and gives the extreme loads for operation is that there is typically fighting going on between b and a in the village while perhaps 4 persons looking out for c. It means eg aprox 55 man fighting on a narrow street. If you start to move 2 tanks down the alley things get nasty for the cpu - and damn fun for us assault folks btw

In amiens the same often happens around a and b in sector 4. I think amiens is the worst of the 2 maps. Around the train things can also get cpu heavy.

There is a lot of people playing bf and they complain in game and in forums about performance. Their experience doesnt fit most reviews so they get frustrated.

So again thanx for the work. Very important.

 

[dfk7677]

Thanks for sharing my video. This was really hard to benchmark, took me around 10 hours. You need to find a server (with a free slot) where one team doesn't dominate so you can stay between D-B-E and also you need to balance not staying too far away from the action which would invalidate the results all the while staying alive for two minutes to make a valid run (I did each configuration 5x 2 minutes).

1) Completely valid criticism. Benchmarking the 4th Stage of Operations 64 Players on St. Quentin Scar would have resulted in a higher CPU load but it would take even longer to benchmark correctly. That's why I chose the compromise with Conquest.

2) As you said it's a personal preference, in my opinion 1% Low is the perfect metric to show how smooth the game feels like.

May I recommend some safer ways to benchmark?
1) Table top view in spectator mode (preferably Amiens map, it is the heaviest) . Yes I know that it seems to have nothing to do with the game. Still it is more taxing to the CPU than most 1P cam scenarios.
2) A specific cam without moving it from the initial position (1st Cam in Amiens is the heaviest, afaik).

Of course these benchmarks are not good for a video, but they are more indicative of a processors performance in the specific game.

P.S: Of course in full 64p servers.