The guy that did those "totally biased 1080p benchmarks

File: lol.jpg (20KB, 1280x720px) Image search: [Google]

20KB, 1280x720px

>watchdogs 2 utilizing all 8 cores / 16 threads
>still slower than 7700k

>>59224155
All benchmarks are shit until the mainboards get better and stable BIOS and Microsoft updates the scheduler of Windows.

>>59224245
That is probably due to SMT reducing performance due to software bugs. There's confirmed bugs where SMT is making the two threads fight for resources inefficiently in gaming even though it works great in Cinebench and so on.

Ideally he should have done a test with both SMT and HT disabled.

>>59224334
this, I remember the p67 fiasco with sandy/ivy

t. p67 owner

File: 1432263277632.jpg (7KB, 207x200px) Image search: [Google]

7KB, 207x200px

>>59224346
>making the two threads fight for resources inefficiently
Inherent to AMDs SMT or is this a per application thing?

>>59224155
USE A FUCKING HISTOGRAM YOU RETARD

>>59224334
This. There's alot more involved here than just the processor. Everything in these test rigs is brand new for fucks sake.

File: getgraphimg.png (63KB, 600x748px) Image search: [Google]

63KB, 600x748px

>>59224401
Its a game engine thing, even intel chips see significant performance regressions in some titles.
Though as a bonus intel also has a buggy as fuck single core turbo.

It comes down to the fact that SMT only provides a 30% uplift in throughput. If you have two critical software threads you should be running them on two separate physical cores. Putting them both on one core will hurt your overall performance.

>>59224401
It seems confirmed that it's that many game engines have their own scheduling code specifically for HT that is not working correctly for SMT.

It appears to be a problem with the games never accounting for a different type of simultaneous multi threading ever being in use besides HT, while SMT and HT work differently.

While if you look in Cinebench R15, SMT gets like 9.5x the performance with 8 cores over 1 core when the 6900k gets like 8.75x the performance in multithreading.
So SMT seems to actually work better than HT, but the game engines are fucking it up thinking it's HT when it's not.

>>59224473
>even intel chips see significant performance regressions in some titles.
Intel sees a 50% performance increase in Watch Dogs 2 with HT enabled. Many other games see at least 20%.
The performance regression from HT initially I think was a Windows scheduler bug seeing it as an entirely different core when it's not. Now days it's incredibly rare or non existent to have HT-on regress performance.

Since this benchmark here is with HT on and SMT on, I have a feeling that with both off the performance delta would be very close. And once SMT gets working right, you'd see the delta remain similar just with Ryzen performing better.

>>59224245
you literally cannot fully multithread most individual tasks. modern game engines already have pipelined architectures which add concurrency at the cost of more latency. certain tasks like video rendering are different because you can process different pixels in parallel (and this is what GPUs take advantage of to get their performance advantage in certain tasks like video rendering).

>In most cases, performance is a constant point of concern, and a parallel pipelined architecture is adopted to allow multiple processors to work in parallel instead of sequentially. Large command buffers on GPUs can buffer an entire frame of drawing commands, which allows them to overlap the work on the CPU, which generally gives a significant frame rate boost at the expense of added latency.
https://web.archive.org/web/20140719053303/http://www.altdev.co/2013/02/22/latency-mitigation-strategies/

>>59224588
HyperThreading and SMT are the same thing, guy. HyperThreading is just intel's marketing name for their SMT implementation.
Take a good hard look at that chart.

I remember a bunch of Intel chips had the same problems with HT and would run better with it off. Did that ever get fixed or was the performance different just to minor to care?

File: vivaldi_2017-03-03_18-14-18.png (195KB, 1001x859px) Image search: [Google]

195KB, 1001x859px

>>59224656
Lmao. They do not work exactly the same.

Holy fuck you can't read. All you can do is regurgitate.

>>59224722
It was fixed. The other guy is an idiot citing old as fuck tests.

>>59224727
Anon, they are the same thing. Simultaneous MultiThreading is the theoretical concept, HyperThreading is intel's marketing name for their specific implementation in their architecture.
You're trying to argue that wheels and tires on a care are separate things.

Don't post about a topic you're completely clueless on.

>>59224155
>he should have done an SMT disabled
Isn't it disabled already? If not these actually are some pretty impressive results.

>>59224843
Nope. He shows the cores/threads in BF1 and it has 16 threads being utilized.

>>59224727
Gamers Nexus ? Into the trash it goes.

>>59224588
>Windows scheduler bug seeing it as an entirely different core when it's not
Maybe it's the same problem with Ryzen? I remember seeing a Linux commit that fixed an issue like that with Ryzen CPUs.

>>59224781
>for their specific implementation
So it's different then.

>>59224888
Oh I should've checked the video first. Seems like ryzen is actually pretty good for gaming after all despite all the hysteria from the intel brigade. And it will only get better with time.

>>59224947
How are you this fucking stupid?
Every implementation of SMT is different. They're all still SMT.
You very clearly thought that SMT and HT were different features both present on an intel processor, which is laughable.

Stay in /v/, kid.

>>59224656
he's referring to the different implementations by intel and amd

>>59224155
>Intel has more frames in gaming

Everyone knew this already dumb AMDshill

>>59225069
No, he isn't.
>Since this benchmark here is with HT on and SMT on
Referring to the bench I uploaded here: >>59224473

The little /v/ retard is just using terms he doesn't understand.

File: 1487463821177.png (2MB, 1440x900px) Image search: [Google]

2MB, 1440x900px

>>59224986
>Every implementation of SMT is different
Said it himself and still doesn't get it. wew.
What a kid.

>>59224935
AMD seemed to say that it's a bug in the game engines themselves not recognizing any SMT method other than HT, and not Windows.
Optimizing for AMD SMT itself will be more work, but they could a least patch to make sure their optimizations for HT are only targeting Intel and to otherwise use cores only.

Developers have to test whether they should use HT or not with an Intel tool, VTune. It isn't exactly intuitive or easy. They just have to run their program and tool, and see if spawning the thread on a logical thread makes performance better. If not, they need to set it to only spawn on a physical thread.

Doing such optimizations just with VTune was not a problem when only Intel's implementation of SMT, Hyperthreading, was in use for consumer CPUs.
So the problem is the game engine thinks some things will be improved using SMT when they won't purely based on tests with Intel machines.

Though there could be some bugs in Windows as well that are also contributing. It seems weird that it's ALWAYS worse in games when they are both 2-way SMT that work a bit similarly, but in workstation applications SMT is giving big benefits.

>>59225269
You're pathetically desperately shifting goal posts after you got called out for talking out of your ass.
Give it up.

>>59225123
I hope you can read that graph correctly and that you're just trolling.

It was a pretty good one, if so. Expected it coming.

>>59225069
I'm using the abbreviation "HT" to differentiate between AMD's implementation because AMD simply calls their SMT SMT. This is what most people do.
AMD has no special name for theirs. You can call it "AMD SMT" if you want to get specific but there has been no need. The other guy is just being an idiot.

>>59225293
>And not a single thing he said was refuted by you

>>59225302
You can't save face after embarrassing yourself this hard.

File: 1484109919337.jpg (77KB, 640x638px) Image search: [Google]

77KB, 640x638px

>>59225293
>he's posting things I don't understand quick something something about shifting goal posts whatever that means when there were never any goal posts to start with and say that all this stuff I don't understand is just him talking out of his ass!!!! xd

>>59225316
>>59225293
>if I repeat myself I win

>>59225249
intel SMT is almost always referred to as HT especially when talking about gaming. if he had just said that SMT was on it wouldn't be clear that it was enabled for intel as well

>>59225315
There is nothing to refute.
See: >>59225249

This /v/tard tried to use terms he know nothing about, now hes trying to do damage control.
The little retard tried to claim the bench I posted showed both SMT and HT, as if they were both distinct features, and both present on the single intel CPU being tested there.

>>59225323
Even more damage control.

>>59225338
Try again.
The bench in question is testing one CPU, an i7 6950X.
You can't explain away what he said. He very clearly was clearly the i7 6950X had both SMT and HT, as if they were separate features.
He flat out doesn't know what they are.

>>59224155
I'm sure /v/ kiddies are too young to remember, but this is the same story as it was with AMD Athlon when it came out first time. Comments were like that: good performance, but since it was derived from server chip (Opteron) is good for scientific calculations but not for games. Intel is better overclocker and gaming chip. Once Athlon matured it showed its power and Intel didn’t recover till Core.

>>59225123
>5GHz processor with similar IPC but 4 less cores has fairly better performance in games than a 8C/16T processor clocked 1.1GHz lower

Intel shills are fucking retarded for going ape shit over something everyone else was already expecting.

File: SMT v HT comparisons.jpg (84KB, 595x767px) Image search: [Google]

84KB, 595x767px

>>59225387
I don't think I've ever gotten to witness someone getting completely BTFO as hard as you're getting BTFO right now.

I'd recommend you not just never visit 4chan again, but no other site that lets you write anything to other other.
What do you even do on sites when you can't hide your stupidity behind anonymity?
There's no way you're going to be able to keep yourself from being made a complete fool of elsewhere unless you just give up on ever posting anywhere ever again.

File: 1488059530739.jpg (217KB, 800x1000px) Image search: [Google]

217KB, 800x1000px

>>59225356
>>59225387
>>59225700

>>59225891
>make an absolute fool out of yourself by using terms you're absolutely unfamiliar with
>instantly get hostile when corrected
>back track
>move goal posts
>desperately google terms to try and save face
>post a completely different chart from the one in question, which is totally irrelevant
>and at the end of it all you still quote the wrong post

Its like watching pottery in motion.
You insisted that an i7 6950X had *both* SMT and HT. It is abundantly clear, you cannot deny it.
Plain as day: "Since this benchmark here is with HT on and SMT on"
The chart I posted was looking at intel's single core turbo and hyper threading on the i7 6950X. Its only looking at that single Broadwell-E chip. It is not an AMD vs intel chart, so your backtracking defense holds no water.
I simply corrected you, and you instantly started shitposting like the little /v/ cretin you are.

This entire exchange has been you trying to save face for being a loud mouthed gaymer retard who got called out for talking out of your ass, and your autistic ego couldn't handle it.
Good job. Really productive.

File: 1484111197324.jpg (263KB, 800x601px) Image search: [Google]

263KB, 800x601px

>>59226077
>actually bothering to respond instead of hiding forever
>more being unable to refute anything
>more doubling down on retarded claims
>doesn't understand that "HT on and SMT on" clearly means "HT on for Intel while AMD's SMT is also on" while if someone just said "with SMT on" they might only mean AMD's SMT on since Intel's is almost always referred to as HT except in the case with Xeon Phi that uses a different implementation.

wew

>>59226249
Again, that pathetic childish defense holds no water. The chart is question does not feature any Ryzen chip. It is for the i7 6950X alone.
You stated "Since this benchmark here is with HT on and SMT on" because you legitimately thought they were totally distinct features on intel's chip. You didn't understand what the terms were, and you started shitposting when you were called out for it.
You're still just shitposting now.
You haven't argued anything that warrants refuting, you've just been trying to save face after you embarrassed yourself.

You can take a /v/irgin out of /v/, but you can't take the /v/ out of the /v/irgin.

>>59226077
"Since this benchmark here is with HT on and SMT on" was clearly referring to >>59224155 having SMT and HT on, you imbecile, not the 6950X one that is HT on/off with turbo on/off.

How the fuck could "Since this benchmark here is with HT on and SMT on" refer to >>59224473 when it has 4 different bars and clearly has "off" when that line was only referring to a benchmark that had both on and was talking about the performance delta if they were both off?

Holy shit you're being BTFO so hard. Just end your life.

File: 1486329456451.png (60KB, 400x400px) Image search: [Google]

60KB, 400x400px

>>59226327
>>59226338

File: ▄█▀ █▬█ █ █ █ █ █ █ ▀█▀.jpg (38KB, 450x495px) Image search: [Google]

38KB, 450x495px

>I don't think I've ever gotten to witness someone getting completely BTFO as hard as you're getting BTFO right now.

>I'd recommend you not just never visit 4chan again, but no other site that lets you write anything to other other.
>What do you even do on sites when you can't hide your stupidity behind anonymity?
>There's no way you're going to be able to keep yourself from being made a complete fool of elsewhere unless you just give up on ever posting anywhere ever again.

He was warned but he posts more dumb shit anyway. The absolute mad man. >>59226327 >>59226077

>>59226338
I presented a simple fact, and posted a chart along with it to demonstrate the point.
Even intel's chips can see significant performance regressions with SMT, Broadwell-E in particular has some major performance regressions with its buggy single core turbo as well.
That is what the /v/ cretin responded to.

A dumb little /v/irgin using terms he doesn't understand got butthurt for being corrected.
Stop trying to defend him. All he does is shitpost, and defending him puts you right on the same level.

>>59226282
>stock 1800X only 10fps behind the twice as expensive 6900k despite having performance issues related to the mobo and OS/game optimization
Looks pretty good to me.

>>59226472
The bigger thing in this graph is to see how HT improves performance in that game, but SMT drastically tanks it.

Compare the 7600k, even OC to 4.7ghz, and it has 50.7 0.1% low FPS.
While the 7700k stock, which is basically a 7600k with HT, gets 111 0.1% low FPS and almost 15% higher average FPS on much lower clocks.

Most games now days benefit from HT by about 20-50%. Yet those same games are making Ryzen with SMT enabled perform 8-12% less than turning it off.

HT had the same exact problems in the past like: >>59224473

So imagine when SMT gets fixed. All these benchmarks are just going to be void.

>>59226524
> The bigger thing in this graph is to see how HT improves performance in that game, but SMT drastically tanks it.
Will, probable, be much better in future, when software/uefi will get better.

https://forums.anandtech.com/threads/official-amd-ryzen-benchmarks-reviews-prices-and-discussion.2499879/page-107#post-38771400

>>59226759
Great link, anon.

very autistic semantic argument itt but as far as Ryzen UEFI performance gains go, don't expect much of a difference after the big feature bullet points like SMT are fixed without optimisation work from game developers themselves. Joker's Gigabyte motherboard might be nearly the best case scenario for a while yet. And frankly I don't ever see people like Firaxis ever bothering to fix Civ VI.

>>59224155
Cant you just disable extra cores/threads in Ryzen to make it faster?

>>59224155
>buy i7-7700k
>microstutter like fuck

WTF senpai

>>59227021
>disable cores
>make it faster
AMDrones, ladies and gentlemen.

>>59227021
Probably. Since one of the problems is the windows scheduler moving threads from one CCX to another.

If you can disable an entire CCX, that should actually improve performance in the games it does very bad in.

>>59227021
Yes but it still trails the 7700k from what I've seen in most gayms. There's more to the story, whether its bugs in the bios and microcode or whether there's systematic inherent biases in the software tested. No one's bothered to validate with HT-0 Broadwell-E chips as far as I know.

>>59227107
>>59226759

>>59227145
The point about the CCX's is interesting, I just don't know how much I buy it. Does this just happen in games? Why or why not?

>>59224781
Have you ever coded with HT in mind you stupid fuck?

File: Untitled-1-640x353.jpg (29KB, 640x353px) Image search: [Google]

29KB, 640x353px

>>59227204
Things like Cinebench and stuff have more consistently loaded threads that Window's scheduler won't keep trying to move around.

It happens in games because their threads have way more inconsistent performance so keep getting moved to try to balance the load across threads.

Applications can also decide how their threads are scheduled and deleted manually. This can be done in games in ways that's very suboptimal for AMD.

But even then.. I'm not sure how much of cache is really being utilized in games and how much losing that shared cached and having to hit RAM again is really affecting performance. But it's definitely got to be SOMETHING.
The biggest issue is definitely SMT making performance worse while HT often makes it better. That's why I really wish we could get these same tests by the guy that shows them disabled.

>>59227272
He clearly hasn't. Don't worry, he was BTFO enough.

>>59224155
>Personally, I think he should have done an SMT disabled and HT disabled test. HT is working properly in software and gets Intel 20-50% benefits in some games. SMT is currently dropping Ryzen performance 8-12% varying by game.
He's actually going to do it.

>>59227021
>spend $750 on 1800x+mobo because muh cores
>disable half the cores because muh performance
>now you have less performance than 7700k in all regards

>>59227469
>I'm not sure how much of cache is really being utilized in games and how much losing that shared cached and having to hit RAM again is really affecting performance.
it's huge

>>59227707
Well completely DISABLING the cache would be much more than a 10% difference in performance.
But this isn't disabling. It's forcing it to uncommit and re-fetch by swapping threads around when they're swapped around.

So it's definitely something. But I can't say how much exactly.

>>59227674
Should be much better than the 7600k still when SMT is working, though.

The 7700k heavily, heavily outperforms the 7600k in many games despite them being virtually identical except that the 7700k has HT and the 7600k does not.
I'd be happy with performance just halfway between the 7600k and 7700k in those games.

>>59228027
you want as few problems with the cache as possible. most current game engine architectures have data access patterns as one of the top priorities.

https://www.youtube.com/watch?v=rX0ItVEVjHc

>>59228120
Yeah this might actually be the biggest reason that SMT disabled Ryzens still underperform so much.

Like I mentioned, the scheduler is going to shift threads around to different cores way more in a game than consistent tasks.

Have to wait to truly see, though...

I'm going to go with this being a issue:
https://forums.anandtech.com/threads/official-amd-ryzen-benchmarks-reviews-prices-and-discussion.2499879/page-128#post-38774366

>each zen thread is being registered as an individual core with its own L2 and L3 cache. i.e. totaling 136 MB cache!!. this is using Windows Sysinternals. This explains the SMT troubles in the event that a thread bounced to a HT thinking its the real deal.

Windows thinking it has 136MB of L2 and L3 to share across 16 threads and not 20MB is bound to be an issue. It will also help when Windows understands the CCX layout and power states too.

>>59224334
>It's not AMD's fault they have launched another shit chip, itttt's Microsoft, and asus and and MSIm and games makers.

>>59224727


The ladyboy on gamers nexus even came out and admitted to fucking up his test...

Then has the balls to say something along the lines of "I stand behind it because this is how they sent it" "the 5 things I did wrong only caused a 5% difference in performance each, it's fine!"

>>59229801
Yeah someone else posted that.
That seems to definitely be the biggest culprit besides just general bad settings, too slow of memory, and basics like that.

>>59230067
Yeah, what an asshole. The guy has been rubbing me the wrong way with how stubborn and sure of himself is when there were clear issues with their write-up. It's obvious his test methodology was flawed but "we wrote 10,000+ words and spent 1000 hours total!!!"

At first I was giving him the benefit of the doubt, because it's easy to just be ignorant to a small detail, overlook something, and get caught up on what you believe to be true when it's not. Shit happens.
But then you have it coming out that it's very obvious something was wrong that he should have noted in their review, and that they were pretty obviously making their review bias on top of that (doing 1440p with SMT enabled only when their other tests showed it was obviously bugged, for example) and not pointing out some clear positives.

>>59228120

it's really only a concern right now for consoles, which have super weak CPUs. shitty pc exclusives built on decades old object oriented shitcode (i.e arma) don't use data oriented design at all.

File: chrome_2017-03-04_02-52-41.png (87KB, 968x618px) Image search: [Google]

87KB, 968x618px

I hope this is just reporting wrong

Coreinfo output.

If it's right, it's showed the cache isn't shared at all. Or rather, Windows doesn't see it as shared.

>>59230410

doesn't matter if it's reported incorrectly or not. the OS has no control over what data gets put into cpu caches.

>>59230769
It matters because the OS doesn't know to prefer to move threads around to a logical thread that shares the same unified cache if possible if it doesn't see it.

File: intelonsuicidewatch.png (408KB, 882x1418px) Image search: [Google]

408KB, 882x1418px

>>59230025

>>59231002
They are fairly cherry picked, but that guy's configuration was definitely outperforming most others.

It's the same guy that benchmarked in this video at 720p that did those to show there was mostly not a GPU bottleneck except in Sniper Elite 4 and Overwatch, pretty much. And sometimes in Tomb Raider, but sometimes not.

File: 17103289_10155035829263454_5901110875835269422_n.jpg (65KB, 865x519px) Image search: [Google]

65KB, 865x519px

What did they mean by this

File: 2.jpg (118KB, 601x830px) Image search: [Google]

118KB, 601x830px

>>59231002

>>59231747
5% difference with bugged SMT disabled while the Intel chips benefit from working HT.

Nothing to see here.

>>59231747
why is the 6950K performing worse than the 7700K, it costs like 4x as much

>>59231757
>bugged SMT
You mean shitty implementation of SMT.

We saw the exact same thing with first generation HT cpu's of Intel. It took them a couple of generations to get it straight (and games to optimise for it), AMD isn't fixing this one with a patch

>>59231783
Lower clock speeds and IPC.

It's not a gaming cpu

>>59231183
How are they cherry picked at all? There's 10 games being benchmarked there

>>59231803
Didn't the 1700x beat the 6950K on a lot of non-gaming benchmarks?

>>59231818
Not sure how that is relevant, we are discussing gaming performance here

>>59231788
>shitty implementation of SMT

No, it's just that games are only accounting for Intel's HT since AMD's SMT never existed when they launched their games. >>59227469

>>59231827
Which is exactly what I said, and that's a problem, because that is not fixable with a patch

ITT anon completely forgets that windows doesn't have any zen specific patches right now. Anon also forgets how windows effectively didn't support bulldozer upon release either. Anon ALSO forgets windows has been patching in support for Intel's new (at the time) chips as each generation is released.

The upcoming creators edition hopefully has some zen specific tweaks that will improve performance . Right now most reviwers don't run their test system in high performance mode which had a negative impact on inter core zen performance.

>>59231803
Is that what the R5 series will be geared more towards? Fewer cores and higher clock speeds?

>>59231876
Fewer cores yes, higher clock speeds, probably, but I'm not expecting Kabylake clockspeeds.

the rx 1800x does 4.1 ghz on watercooling, so the 6c/12 might do 4.2-3 ghz on air, the quad cores might do 4.5 ghz, but I still think it's unlikely

>>59231847
Huh? What you mean by "SMT implementation" is confusing, then.
It makes it sound like you think AMD's SMT hardware is inferior, when it's actually superior.

It's just that games aren't using it properly like they are HT.

>>59231876
No, the clock speeds likely won't be higher. It might do 4.0 or 4.1 on more cores than 2 with XFR, but they aren't going to go to 4.2 or higher on a core, I'm almost completely certain.
The chip is far worse perf/watt past 3.9 and its sweet spot is at 3.3 and below for an average binning.

Most of the performance problems seem to be from Windows scheduler moving threads from one core cluster to another, making them lose their L3 cache and have to uncommit, fetch from memory, and recache. And from SMT being enabled when it currently gives a performance penalty of 8-12% in almost every game.
These are fairly simple to patch issues so we should see better benchmarks before R5 launches.

>>59231934
>when it's actually superior.
How is it superior?

>>59231876
8 threads for half the price of a 7700k.

The 7700k is the cheapest 8 thread CPU that Intel jews offer.
The 7400, 7500, etc, are the exact same CPU that they arbitrarily disable hyperthreading on just to try to justify the 7700k costing 50% more than a 7600k.

>>59231941 >>59227469

File: AMD vs intel vs nvidia.gif (1MB, 690x200px) Image search: [Google]

1MB, 690x200px

>>59225387
in case anyone thinks he's trolling, it's actually true
t. oldfag

>>59231947
It's not superior for gaming though, in fact it's clearly inferior. And we are talking about gaming performance here.

It's like you're saying a pro tennis player is better at playing soccer than a pro soccer player because the tennis player is better at tennis

>>59231975
t. can't read

>>59231983
You're BTFO

>>59231975
The R5 lineup will offer a much better gaming performance than the R7 series though.

It's like comparing a 6900K to a 7700K.

>>59232011
Really? What will the R5 series change about the lower clockspeeds compared to Kabylake and the fact that SMT produces worse results in games while HT produces better results?

The clock speeds will be marginaly higher, but with the 1800x on 4.1 ghz on watercooling I'm not expecting much

>>59229801
This is probably the real reason there's so much difference in benchmarks and games
Benchmarks use small data sizes, games use big ones.

>>59231986
No. You're a fucking moron.

At $175, the 1300 with 4cores and 4 threads would destroy the 7600k if SMT is working as well in games as HT at that point.

Pin the 7700k down to only 3.3ghz and it still beats the 3.8/4.2ghz 7600k in most recent games since they pretty much all get a significant performance gain from HT. That's a simple fact. Most games the past few years are running much worse with only 4 threads. That's what you're looking at with the Ryzen 1300.

AMD launched these first so nerds would find the issues for them to fix by the time of the R5 and R3 launches. There's no architectural reason Ryzen is doing bad. If you could read for shit and weren't such a dumbass then you'd understand that.

Anyone who thinks it's still 2011 and hyperthreading doesn't help gaming is a fucking moron.

>>59232018
No it's likely that workstation applications keep consistent loads on the threads so the scheduler doesn't move them around.
Games get wildly different load along the threads from one moment to the next so they keep getting moved around.

>>59232014
Clockspeeds will be higher though, more headroom due to fewer cores being run.

They are way, way cheaper than Intel's competition too.

>>59232042
>if SMT is working as well in games as HT at that point.

Can't you fucking read? We had the exact same thing with the earliest cpu's of Intel with HT.

It took developers several cpu generations top optimise for it. AMD can't fix this with a patch.

This means SMT will continue to neck performance in games instead of improving it in the foreseeable future.

Stop fucking delusional and wake up to reality

>>59232063
Yes they will be higher, but not as high as their Kabylake counterparts.

Competing on pricepoint is possible, but that wasn't the discussion

>>59232075
The thing with Hyperthreading, when intel introduced it, it was one of the first, if not the first implementation of IBM's SMT technology on the consumer market, developers had to learn how to work with it.

That was more than 10 years ago now, so for them to adapt to how to work with another implementation of SMT will not be difficult.

File: 1449164100166.jpg (122KB, 602x655px) Image search: [Google]

122KB, 602x655px

>>59224155
>run top of the line brand new AMD processor at 1080p
>performs worse than something older
>NO U MUST RUN TE BENCHMARK AT 720P
who the fuck is buying a top shelf ryzen to play games at 720p? are people really this desperate for a good benchmark?

>>59224401
same thing was happening when intel made first HT chip

>>59232101
It's about testing the CPU.
If the GPU can pump out 900 frames per second at 720p that means the framerate is solely based on the CPU and so you can see which is the best performing one, relative to the others. Which you then factor into a purchase decision by weighing costs.

>>59224155
I don't know, I'm pretty sold on that smooth line. It means drivers run very efficiently, graphics pipeline performs ideally while game itself just doesn't know what to do with new arch.

>>59232083
You all need to stop assuming the lower core count Ryzens will clock higher.

The facts are against your baseless deductions.

Ryzen starts hitting a wall past 3.9ghz on average silicon where it starts requiring far more power to remain stable.

The all-core record is currently 5.3ghz on LN2 while the single core record is only 5.8ghz, also on LN2. And that's at like 1.97v.

No Ryzen CPU is going to go over 4.0 stock. 4.1 with XFR. What they MIGHT do is have 4 cores go to 4.1 with XFR while Ryzen 7 will only go to 4.1 on 2 cores max, but that's it. I guarantee it.

>>59232099
He is saying that HT took years for games to start optimizing for.
I don't want to bother to argue it and guess how long it will take this time around.
But personally I think SMT optimizations will come faster due to experience with HT and it being a similar concept. AMD just needs to release a tool to analyze for how to optimize for SMT like Intel has with VTune. In the short term, developers can release a patch that checks for an Intel CPU before attempting HT optimizations that are negatively affecting Ryzen.

>>59232123
but then if one performs better at a high resolution doesnt it mean that's the better processor overall?

>>59232177
they argue that in time those small difference will become huge ones at 4k with newer GPU

but since we can test older up to 5-7 eyars E series CPUs and see that it's not true at all I don't know where their logic is coming from
you see that ryzen has no jumps? it means GPu performs ideally, executing driver commands without delay, game stuck at unrelated to graphics parts because it doesn't have even slightest idea what it's dealing with and treats it like intel cpu

>>59232149
That also assumes that developers will start optimising for SMT straight away, we don't know if that will happen.

>>59232272
at least they will start adding new arch into code, I think it will gain performance just from "understanding" that it's not 6900K

>>59232272
We know that AMD and bethesda have entered a long term partnership. The results of which will likely be proper multicore and SMT/HT support in future titles.
SMT could possibly be patched into existing titles that have some degree of HT support, such as the Battlefield series, anything Frostbite of Cryengine as both of those support up to octa cores, and HT.

>>59232301
>SMT could possibly be patched into existing titles that have some degree of HT support

That is based on nothing, a baseless assumption

>>59232301
will see with Prey in couple months, but it's cryengine it should be fine

>>59232227
Kaby lake actually produces far more stutter and uneven frametime at 4K than Broadwell-E and Ryzen due to it's ridiculously unstable clockspeed out of the box.

However, if you're deadset against 1440p or 4K for the next 3 years and don't want an upgrade path, the intel CPU might be the better buy

>>59232323
>Kaby lake actually produces far more stutter and uneven frametime at 4K than Broadwell-E and Ryzen due to it's ridiculously unstable clockspeed out of the box.

Source or full of shit

>>59232314
the only thing that changes it from intel HT is shared L3
considering AMD is really open about how it works for developers, i'd say it can be done

>>59232344
If that is true, we will see improvements soon, if not, you're probably full of shit

>>59232323
On 1440p and 2600 for a year now, I'd like to get a 1700 after seeing all rebenchmarks and frametimes.

>>59232350
I think biggest concern is MS fucking it up as usual with year long patches.

>>59232350


All you really have to do is look at 4K benchmarks.

If there was nothing else other than a GPU bottleneck holding back all of these CPU's, then sites like Gamers Nexus would be right, but as it turns out Ryzen ends up ahead by 3-4FPS almost every time with more than half of games tested

File: 8.jpg (115KB, 1200x704px) Image search: [Google]

115KB, 1200x704px

>>59232401
>but as it turns out Ryzen ends up ahead by 3-4FPS almost every time
Except you're wrong, and I don't know what graph you are basing this on.

Show me a broad average across lets say 20 4k games or you can't make a statement like that

File: 7.jpg (120KB, 1200x675px) Image search: [Google]

120KB, 1200x675px

>>59232421
And another

>>59232421

>posts one of only 3 benched games where it's not the case
laughing my ass off, thanks kid

>>59232451
Actualy I posted two, while you haven't posted anything

File: 11.jpg (312KB, 1204x597px) Image search: [Google]

312KB, 1204x597px

>>59232451
Here, have some more, that's 4-0

>>59232483

>1fps
>well within margin of error

Also note that performance regression for the 700k going from 4.5ghz to 4.8ghz in gears of war.

>>59232483
>i7 2600k
>56min/69avg

>Still above 1800x which has 59/69
Objectively shit benchmark.

Not the dude you're arguing with about by the way. He may be able to cherry pick 1-2 games where his claim is true, but otherwise the difference at 4k is so fine there's no point arguing about it.

>>59232505
Your statement was:
>but as it turns out Ryzen ends up ahead by 3-4FPS almost every time

So even if they end up being equal, it proves your statement wrong. I'm still waiting for actual data to backup your argument, or are you just full of shit?

>>59224401
Windows scheduling

>>59232296
And that AMD SMT is not HT

File: Screenshot from 03-03-2017 16:45:41.png (294KB, 915x813px) Image search: [Google]

294KB, 915x813px

>>59232421
Here's a compilation of 4k where the 7700k doesn't fare all that well.

Wow it's almost like people have different setups and unpatched BIOSes or something.

>,Running in the balanced mode prevents Ryzen from switching clockspeeds and voltages any faster than 30ms. High Performance allows it to go into full-speed 1ms switching.

Great going MS

>>59233464
AMD apparently submitted fixes to MS long before launch but Windows won't release it.

>>59233502
Thanks Intel

>>59233508

You joke but that actually could be the case. Wintel is still strong.

>>59233553
There's a GTA case where Ryzen uses 4GB higher VRAM than a 7700k in the same scene, the CPU doesn't do that, the Nvidia drivers do.

Something truly rots.
Either intentional or software incompatibility, I hope it's the latter.

>>59233571
Don't disbelieve you, but wanting a source

Seems like Gigabyte boards still post better results than ASUS one, BIOS issues aren't fixed yet.

Man there are fuckups everywhere, I wonder how everything will look in a month or two.

>>59231783


4 cor higher clocked cpu. The problem with the 5ghz 7700 ate the 100c Temps and the 225watt power consumption

>>59233599
https://forums.anandtech.com/threads/official-amd-ryzen-benchmarks-reviews-prices-and-discussion.2499879/page-130#post-38774826

There's screenshots earlier in the thread, I'll try to find it but the search tool on the forum is junk.

File: laugh.jpg (77KB, 600x449px) Image search: [Google]

77KB, 600x449px

>>59233345
>chinese source

File: Screenshot from 04-03-2017 15:11:34.png (24KB, 940x265px) Image search: [Google]

24KB, 940x265px

What the fuck is even going on with these chips anymore?

File: 1467055276405.jpg (26KB, 251x242px) Image search: [Google]

26KB, 251x242px

First Generation Ryzen CPUs are totally buggy.

Good thing there are so many moronic beta tester on the internet paying a load of money to test all the bugs.

>>59233934
This, should be interesting to see what's the situation after a month or two.

>>59232042
>At $175, the 1300 with 4cores and 4 threads would destroy the 7600k if SMT is working as well in games as HT at that point.
So what you're saying is that the AYYMD processor would destroy the Intel one if it was as good as the Intel one.

>>59233934
>meanwhile at AMD HQ
>man we're out of time to validate Naples for the Q2 launch, get working firmware and microcode, wat do
>release Ryzen to the unwashed gamer masses to debug
>???
>PROMOTIONS

>>59233991
Literally getting paid to have people bug test for you.

>>59233991

Its more or less what Intel did with the P4. X99 needed quite a few BIOS updates after release to sort out its memory going mental.

File: 1473117752927.png (799KB, 741x725px) Image search: [Google]

799KB, 741x725px

>https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/page-5
>yfw Win7 performs better on Ryzen than Win10

File: image010.png (14KB, 450x417px) Image search: [Google]

14KB, 450x417px

>>59232042
>-B-B-BUT IF WINDOW SCHEDULER GETS PATCHES RYZEN WILL DESTROY INTEL!

This is from 2012 soon after the Faildozer launch:
http://www.tomshardware.co.uk/windows-7-hotfix-bulldozer-performance,review-32365.html

Windows after the SMT patches. ZERO DIFFERENCE IN PERFORMANCE.

Keep dreaming AYYMD fags. Ryzen is a flop. Perhaps if you wait™ some more.

File: shot0001.png (711KB, 1280x532px) Image search: [Google]

711KB, 1280x532px

>>59234055
>Microsoft
Is there something these guys can't fuck up?

>>59234075
Isn't startcraft 2 notorious for being poorly optimized and poorly threaded? How would SMT affect it at all?

>>59234075
Wasn't there some sort of update to the Linux Kernel that improved performance from SMT because the old Bulldozer fix was interfering with it?

File: image016.png (14KB, 450x417px) Image search: [Google]

14KB, 450x417px

>>59234116

Try reading the linked article. There are more benchmarks there.

File: REKd.png (52KB, 720x289px) Image search: [Google]

52KB, 720x289px

>>59232378
year of linux gaming

>For those who came in late, the Intel x99 chipset and the Haswell-E CPUs have been out for almost 2 months now. In this past time, there have been multiple reports about motherboards having fatal errors resulting in a permanent CPU malfunction as well [1, 2]. Since then, everyone has been playing the blame game- DDR4, PSU, CPU and motherboard all being the culprit.


>new platform has bugs
Unbelievable, this is an outrage

>>59233934
>First Generation Ryzen CPUs are totally buggy
And so was the first generation of i7s when it launched, you fucking imbecile. Babby's first time following the release of a new CPU architecture huh?

>>59234195
Only Intel is allowed to have bugs, you shouldn't talk about the SATA bugs, USB bugs, 8MB bugs, BIOS bugs(not 100% Intel's fault) shutting off during high load bugs. it's not fair it's not like Intel has a list of errata on their website or something.

File: wrong.jpg (29KB, 396x400px) Image search: [Google]

29KB, 396x400px

>>59234075
>ryzen and bulldozer are the same thing

>>59234191
Fuck gaming, I play 3 games a year at most, everything released has been shit for so many years I've started to read books and shitpost more than I used to play game.

>>59234232
>wait™

>>59234231
Atom processors mysteriously failing after a few months/years

>>59234258
Ah, the newest issue the unwashed gaymur faggots never hear about.

>>59229801
This actually explains the poor performance perfectly. Being given incorrect information about the CPU by windows is catastrophic for optimizing for SMT, and would definitely result in performance drops. I'm frankly surprised it only drops by 10-20%; similar mistakes with Intel penalize performance even worse than that.

t. engine developer

File: days without jewish tricks.png (61KB, 905x633px) Image search: [Google]

61KB, 905x633px

>>59234116

Starcraft II flatout gimps any non Intel cpu. Run it through a VM and spoof the cpu ID to an intel one and you'll see a huge increase in performance.

>>59234255
CMT and SMT are not even the same thing; the former is literally a poor imitation of the latter. Which is why if you check the CPU specifications the FX8350 is marked as 8C/8T. This means the charts you posted are useless for any sort of comparison right now.

>>59234258
>>59234283

Best bit is Intel is refusing ot take responsibility for it.

>>59234385

>the former is literally a poor imitation of the latter

This is wrong - CMT has absolutely no relation to SMT. All SMT is - at a simple level - is a smart hardware scheduler. CMT is about getting more cpu cores in a given die space by sharing resources.

>>59234415
You're right, though it's still supposed to be an alternative to SMT.

>>59234385
CMT like in Bulldozer is a partial implementation of SMT while Ryzen has a full implementation. CMT duplicates hardware where as full SMT simulates the hardware. Full SMT is done more for efficiency reasons, it uses hardware that would normally go idle to prevent wasted cycles. CMT was meant more for performance reasons, it was designed to cram more cores into a smaller space by sacrificing some of the front-end hardware and an FPU.

File: watch_v=nsDjx-tW_WQ-00_02_42.png (1MB, 1280x720px) Image search: [Google]

1MB, 1280x720px

>>59233620
> There's screenshots earlier in the thread, I'll try to find it but the search tool on the forum is junk.
>>59233599
> Don't disbelieve you, but wanting a source

They are right in OP video, 2:42, picrel.

>>59234458
POWER and Niagara use both at the same time. SMT is better for power efficiency, but it's much more prone to going sideways in corner cases.

>>59234190
Not him, but really, what matters most for R7 is how good it is at it's primary role/market and that's not gaming, that's the server market, where it does perform very well.

>>59233502
Assuming they landed bug fixes for this in linux right around month ago, guessing microsoft got them also, but didn't implement yet.

https://archive.fo/uJuX6

>>59234924

Wouldn't be the first time MS held back critical bug fixes just to fuck AMD.

>>59234924
>MS literally cannot find the time to publish 5 lines of code

>>59233812

Not fully baked yet. Wait a month.

>>59234962
What do they gain from doing that, though?

>>59234989

Keeping Intel happy is still vitally important for MS.

>>59226282
1%low and 0.1%low is useless test for cpu.

File: forhonor4CPU.png (104KB, 582x1048px) Image search: [Google]

104KB, 582x1048px

>those minimums

Goddamn at least for this game ryzen is a fucking monster.

>performs like a 6900k with same amount of cores and clocks
>but can lose in gaming from 5% to 40%

How is this even possible? Do people actually think there's dedicated fixed function gaming blocks in these chips and AMD doesn't have them?

>>59235101
>minimum in 10% of average

Now this is how a proper game is made, not minimum dropping to 7% of average unless it's a frame or two totla

>>59235157
Just wait until AMD and mobo manufacturers will fix most Ryzen fuckups.

>>59232103
Actually it didn't, those E-celebs you're watching who keep saying that are wrong
HT never had this level of issues when launched. Sure it was a little rugged but not this bad

>>59235101
>still beaten by haslel for 500 bucks
Sure is a monster

>>59235245

Haslel is a framedropping mess in that chart.

>>59235229
There's more than just BIOS fixes here.

SMT, CCX, power modes in windows, software compatibility like the Nvidia VRAM thing.

There's no small amount of work, I don't care about gaming all that much but hopefully all these fixes make Zen even more impressive for heavy stuff and can go head-to-head dual Broadwell Xeons in something like NAMD that scales to half a million cores.

https://www.servethehome.com/amd-ryzen-7-1700x-linux-benchmarks/

>>59235261
>still gets beaten tho
Really makes my brain use more than 10%

>>59235285
You don't seem to have one.

>>59235261
Can't measure minimums properly in a new Ubisoft game bud, it has issues across the board but is still pretty optimized

File: A7D79A53-64D9-4AC3-9C35-469E2C74D404-13214-00000D2D2EB98424_tmp.jpg (46KB, 1280x720px) Image search: [Google]

46KB, 1280x720px

>>59235313
Upvote XD

>>59224334
>All benchmarks are shit until AMD is on top in them
seems legit

>>59236610

Even with the fixes AMD won't beat an.overclocked 7700.

But it sure as fuck isn't "i5 in gaming" like that bitch at gamers nexus insinuated.

>>59236638
it isn't bc it's worse

>>59224650
>you literally cannot fully multithread most individual tasks
thanks, object-oriented design

>Ryzen is BTFO Intel on enterprise benchmarks.
>Meanwhile /g/ gets shitposted with gaming benchmarks.

FUCK OFF BACK TO /v/!

>>59236801
that has nothing to do with it. for example if you have more than one chef working on a soup, you might get some improvement if they're well coordinated, but they still need to wait for the water to heat up. having 8 chefs could get you 8x the amount of soup (more throughput in a concurrent task) but they can't get you 1x soup 8x quicker (lower latency in a non-concurrent task).

File: Screenshot_2017-03-03-11-59-59.png (278KB, 2560x1440px) Image search: [Google]

278KB, 2560x1440px

>>59236798
wrong. and this isn't the best it can do.

>>59231183
maybe gigabyte mobos don't have bios bugs like others do?