[News] Shares of AMD fall after gaming performance of new processors disappoints

[vario]

Lastly, i would like to point out.Ryzen cpus, are pretty small, at 210mm2 or so.They dont cost that much to produce .They are smaller than both X6`s or FX.
I mean if they do 1 CCX chips , theyre gonna be tiny.

[drmrlordx]

Lastly, i would like to point out.Ryzen cpus, are pretty small, at 210mm2 or so.They dont cost that much to produce .They are smaller than both X6`s or FX.
I mean if they do 1 CCX chips , theyre gonna be tiny.

Actual cost of production will be a component of fab operation costs and R&D costs associated with developing the process. Cost per wafer on 32nm SOI has got to be a lot lower than 14nm LPP right now.

[vario]

Actual cost of production will be a component of fab operation costs and R&D costs associated with developing the process. Cost per wafer on 32nm SOI has got to be a lot lower than 14nm LPP right now.

Well of course ,right now.But 32nm soi is a no go now for cpus. Back in the day SOI also cost more than regular.
Anyhow what i meant is thats always much cheaper to build much smaller chips, and ryzen isnt big.
Also R&D costs are up to glofo not amd, (of course indirectly they do cover them somewhat) .
Also as AMD are now producing everything in glofo im pretty sure they have a nice deal .
Polaris die size is 232mm2, and they are selling it as a part of graphics card for far less.
Of course we dont know really how much amd pays for chips, and how , if they pay for wafers, or only functional units etc etc. There are million things to consider.
BUT, its always better to make something relatively small ;-)

[BenchZowner]

Gaming performance in Linux is bad as well ( bad compared to the top performance from Intel CPUs in 1080p, 1440p and 4k is ok with both platforms as it's very GPU limited ) as shown by the tests Phoronix ran.

[Ket]

Game optimizations for new CPU? What is that?

You are refusing to see the obvious. This cpu is not bad but also not good enough to compete with intel. Admit it or not this was the last bullet of AMD for x86 cpu market and they just hit intel at left arm and left it with 6 bullet remaining in its revolver on its right hand (I am also left handed as AMD).

With this Ryzen AMD left a lot of opportunities to intel for making amd not to shoot again. Intel will not kill amd but will left it crippled for ever. We all know Intel can play very dirty without any hesitation or fear of punishment.

When a new architecture operates sufficiently different enough from existing designs / code this is when performance gets impacted and needs to be addressed with a update. There are examples of this with Zen staring you in the face, I even mentioned them. So again, only in more detail this time;

Zen architecture is sufficiently different to what MS programmed for in W10 that SMT isn't working quite right, a update is being worked on to fix it. SMT on the other hand works exceptionally well in W7, so obviously MS also sufficiently changed their code enough from W7 to W10 (likely in their own attempts to optimise) that now some of the changes they made on the AMD side of things needs reverting back to get SMT working as well as it does in W7 in W10.

Any serious hardcore overclocker knows that for one gain there is a loss in doing so.

Higher cpu clock tuning tends to exhibit lower cpu scores.

Higher ram clocks tend to exhibit lower bandwidth

This is why you may see a cpu validation done on one brand and actual benchmarks on another brand by same overclocker.

Exactly. For sake of keeping the bullet point short and simple I was essentially saying AMD have probably chosen to go aggressive for better performance at lower clocks (weird how memory sub-timings are apparently so conservative though, suggests to me there hasn't been the time to spend on that area up until now). Although I think people would of been more accepting of code tuning that allowed clocks to be pushed higher. There is a lot of tuning to do and things to finish then add to current UEFIs, manufacturers had to rush the code for the boards that are out right now so its only sensible to give them a month or two to finalise code properly then look at how things have turned out.

[vario]

Gaming performance in Linux is bad as well ( bad compared to the top performance from Intel CPUs in 1080p, 1440p and 4k is ok with both platforms as it's very GPU limited ) as shown by the tests Phoronix ran.

So just to recap , 7700K running OC of 4.5ghz on all cores (4.2 def) , 2133 mhz ram, playable framerates at every game, vulkan result is abysmal tho, but it seems pretty obvious theres some kinda bug/beta there, because drop is big for ryzen AND for kabylake, its like 3x slower than opengl.
Put both cpus on 4ghz, give them decent ram/timings, maybe turn off smt and the diff will be 10% .Yes, kaby can hit ~5ghz, in this instance it will be more like ~20%.
Some of the tests seem pointless to me. Whats the difference between 200 and 160fps ?
Or 400 and 300 ?
In portal ryzen is faster 335fps vs 321fps haha.
They have some weird scores there. 6800K is slower than 7700K in blender
It totally does not align with this test
http://www.legitreviews.com/wp-conte...yzen-1800x.jpg

[zanzabar]

since the amd chips wont overclock past 4.1ghz i would count the low clock speed against it for the gaming market. they even have problems getting all cores to 4ghz ATM.

amd also has gone back to ADP as TDP in the ryzen parts. in the operons they give them separate but the 1800x looks to use around 120-130W with XFR and 2 cores working high clocked and the 6900k tends to use around the same at 4ghz with all cores.

[vario]

since the amd chips wont overclock past 4.1ghz i would count the low clock speed against it for the gaming market. they even have problems getting all cores to 4ghz ATM.

amd also has gone back to ADP as TDP in the ryzen parts. in the operons they give them separate but the 1800x looks to use around 120-130W with XFR and 2 cores working high clocked and the 6900k tends to use around the same at 4ghz with all cores.

Well, truth be told, if someone bought R7 Cpu to do competetive gaming at 144hz or SLI`d 1080`s .They did not do their homework.
As for power conumption, toms hardware always does very good measurings, and this slide paints a more accurate picture
https://img.purch.com/w/711/aHR0cDov...UtQmFycy5wbmc=

[chew*]

I just hope that in R5 review bundles AMD will note that cpus should be tested at least at 2666mhz ram which is its official max value.And have oses with proper settings, and better ram support.

Fwiw the ryzen reviewers box had 3000 ram that loaded xmp just fine in gaming-5 also in box.

Amd did not screw up. Reviewers chose to deviate. I did myself but i got better memory. Stuff that my viweres are most likey to use.

My reviews/reports are different at at a totally diff lvl and cater to a different aspect and im not a paid site or paid employee.

Amd was not trying to get miracles out of a review with pc3000 corsair vengeance 15-17-17 memory. Its bottom of the barrel stuff sold at microcenter.

So the real question that needs to be asked who really screwed up here and why and how. Theres no way they could have been oblivious to the results in memory bandwidth...if they are that ignorant or naive in testing they should be fired.

[NEOAethyr]

Oh this is totally on amd .
You can tell from the last few years of some of there vga card releases.
In that sense I also believe they created this interconnect to compete against nv, thinking it was ok for so called next gen computing, and advertising it as such essentially.

But anyways, they have been slacking in the memory department for a while.
It's kinda no surprise that they would just ignore the whole memory thing all together.

It's in my honest opinion that this one may not make it quite where it needs to be.
If this gen eventually gets to or the next one does, 4266 as a rating on the oc speed, (at least has a working 3600 div), then I think it'll fit where it needs to be.
It'll be worth the money.
Plus in the future it should clock better.

I think amd rushed it, and the board makers weren't quite ready.
One can argue the opposite, but that's how I think it went.

However because of this board makers are going out of there way it seems to make this work.
Like adding pstate config to the bios for ex., even if it's just a chunk of the total, it's still nice.

It would be nice to see...
Lets see 2 ccx groups, the mhz for each pstate and the voltage for each one too.
It's just one row I think missing plus voltages.
In whatever board that was :\.
Oh and multipliers..., weird kinda of config if you ask me, only 2 cpu multipliers that need to adjust.

Oh btw I guess these cpu's have onboard power reg's but they are disabled by default.
I was kinda curious about it because I guess there's this setting called soc voltage, and it controls voltage for nvmm and pcie and bunch of i.o stuff lol, it's all locked together...

I don't know what board makers can get out of amd for config info, like 3rd timings for example.
Or even secondary timings...
But I'm sure they'll try.
Otherwise amd has a platform that has a crappy bios.

I'm kinda interested because right now whatever updates they are pushing, it allows you to see the differences.
And lets say they adds a new bclk ratio or ram div sorta thing, just means there's potential to either mod or addon to the tables with you own loader.
Though the oddness to it, imaging a cpu all setup for a super oc on an asus board.
And then you plug it in a biostar board.
Will it still have certain features avail and will it clock better lol?
I just there was an updated tool that doesn't cause the checksums to be invalid on ami bios'es.
I miss the modbin days were even if the code was trashed up from it (like the ami tools..), it still worked as is.


Oh I wanna say to some of you that try to push the cpu that haven't noticed already, the boards have less caps then there mainstream intel equ around the cpu.
If you really wanna push them, even though I don't think it'll help much, you can add caps or replace them.
I would get tantalum caps, I think they make up to 1200uf, it's been a while since I looked (they can be piggy backed).
Hmm I only see 1000uf and 470uf :\.

[chew*]

Oh i dont doubt it was rushed at all thats clearly evident. But early adoption is what it is. I dont mind helping out on board i got for review but in current states I am not buying a board less they iron things out. Would not be surprised to see rev 1.1 1.2 etc etc. Im no stranger to the game but you know that.

I cant afford to bin 500 chips either so boards need to not take out my cpu on ln2 which means they need to be somewhat overbuilt/reliable. Currently that does not seem to be the case.

Fools rush in. I got plenty of time am4 be here for 4 years and i don't NEED points on hwbot so once again no rush. Ill explore platform for now.

[BenchZowner]

chew, there are various reviewer's kits, not all came with the same mobo

Regarding the screw ups in general, I can't help but laugh when "reviewers" use bad windows settings, or transfer windows installations, or don't know how to check benchmarks for consistency, nor how to benchmark a game without a proper built-in benchmark consistently, etc.

It doesn't take a genius to figure that a fully scripted game benchmark test that gives you totally different min and even average figures in a 10 run pool is inconsistent
Same goes for tests of Battlefield for example done only in single player, while most of the CPU stress is put on multiplayer ( 64p load preferred ).

It is really sad to see how the "reviewer's standards" has gone down under these days.

[chew*]

I was unaware of different kits. My bad then.

Totally agree. I ran rotr tomb raider. And i was like wtf? Are you serious i cant use this in a test if i sneeze resilts are different. If i itch my ass different again.

Bf1 lol how could they not know that. I play it i immediately noticed fps in 64p servers different from single player mission. Hell even maps have a big impact.

Thats part of the reason i went legacy. Im familiar with the high or low bugs albeit not games im aware of cpu power impact.

I still use my drive duper for bare os with all tests needed to run and identical drives without any drivers.

Drivers get installed after. Tests done wipe dupe from master install competitor software drivers repeat.

Things have changed. Alot.

[kromosto]

When a new architecture operates sufficiently different enough from existing designs / code this is when performance gets impacted and needs to be addressed with a update. There are examples of this with Zen staring you in the face, I even mentioned them. So again, only in more detail this time;

Zen architecture is sufficiently different to what MS programmed for in W10 that SMT isn't working quite right, a update is being worked on to fix it. SMT on the other hand works exceptionally well in W7, so obviously MS also sufficiently changed their code enough from W7 to W10 (likely in their own attempts to optimise) that now some of the changes they made on the AMD side of things needs reverting back to get SMT working as well as it does in W7 in W10.

There is no real thing like an optimization as you mentioned. CPUs are not like GPUs. CPUs may have different design but they still have x86 architecture. A game can be optimized for a GPU. GPUs are bond to directx (or opengl) but they may have very different architecture because directx does not state the low level design. It just defines the api calls that a GPU needs to implement. So a different GPU may become faster by just changing the order of api calls because of its architecture. But x86 is a different thing. because it defines the low level. Like C and C++ calling conventions. For example "Thiscall" is a convention for calling class functions. You push parameters of the function to stack in reverse order and move your class instance's pointer to the ecx register of the cpu, then call sub procedure from its functions address and retrieve the return value (if there is any) from the acx register of the cpu. If you are slow on this there is not way to optimize it. Of course there are exceptions to what I call and things are not simple as this but these exceptions wont give a noticable performance increase.

Old days we used to optimize our code oftenly but these days compilers are doing there job very well. I cant even remember when I last optimized my code. Of course I am referring to an optimization as you mentioned not optimization of my algorithms or design or common sense of the code of course.

SMT is a different thing which is related to OS as you mentioned. It is different because it changes the was of OS uses core resources. You can see which applications will gain performance when ms fixed this by looking smt off and smt on tests. For games performance will be similar todays smt off results as we know froms intels ht on off results. Which I know games get real performance increases. For other multithreaded applications look its smt on off results. You can also guess what it will be like after fix.

[kromosto]

Of course its good enough to compete with intel.
Newsflash, its not everything about ABSOLUTE PERFORMANCE of the highest end cpus.
Its about price to performance ratio.
AMD still has a lot to launch, whole R5 and R3 series, then there will be raven ridge apus (which will inevitably be better balanced than the intel counterparts for gaming).
If you only understand "Its not as fast in ST as kaby lake! and its also slower in MT than the 6950x!" you have a very bad perspective on this whole thing.
People were buying phenom II`s when the nehalem came out, people were even buying FX-s when sandybridge and ivybridge were out.
Ryzen is a much better alternative than both phenom X6 and FX.

You are right about perspective. But what is bad for AMD today is you are also right about peoples buying decisions because it works for both sides.

Problem for AMD is AMD needs money. How long did it required for AMD to recover and get to today after how intel dealed with them about athlon. 10 years? 8 years? If money flow does not increase dramatically there is a very huge possibility that AMD may not recover from this. Also dont forget Athlon was way better than competing with intel than ryzen.

[chew*]

If amd made it through deneb which saw decent sales considering not very competitive in mainstream and survived through bd and now have a better product than they have had since 939 they will live.

There is a vast market out side of our little 10% world that does not overclock dont want to dont care. If the system does it automatically and the platform as a whole is cheaper...there mommy's will buy it.

The 6 core will be a hot seller. I have a sandy i game smt on. I have a real computer with real apps running in background. Gaming 4 core sux. Gaming 4c/8t is smoother. I probably only need 6 cores when gaming no threading to run background crap and game smoothly.

Ok so warhammer or whatever multithread that i dont play might use more...once again i dont play it so not effected.

Not to mention had i upgraded i would have been pissed. Need new board to officialy support oc ivy. Then kaby so upgrade again...thats in what last 4 years 3 platform upgrades.

Amd is planning on using that one for next 4 years.

In the real world people have to justify a build. Gaming alone is not justification as you get older and wiser. A console games...its cheaper. Not saying its better...better in price for a specific task.

Want a gamer only use one of those which btw AMD silicon is inside.

[NEOAethyr]

In that sense I also believe they created this interconnect to compete against nv, thinking it was ok for so called next gen computing, and advertising it as such essentially.

I found this a min ago:
https://www.youtube.com/watch?v=oZVBMfgGVb8

I haven't watched it all yet but it describes that the interconnect, the fabric thing is essentially a interconnect that uses arm and there gpu along with x86.
Arm was dropped a while ago, at least for now.
It just gives me a better insight into what they were thinking...
It's essentially to advertise as a neural network, that's what I meant to say before.
I think to compete against the nv drive px2...

I know amd can't afford a fail, which is why it bothers me that they released it without the board makers being ready (It's really deeper then that but it can work...).
I think they're lucky that intel hasn't put 6 core dual chan in the main stream yet.
If they wait to long though intel will release that 6 core, and if they don't have there mem inline with what intel has at mainstream they're in trouble :\.

If they can get on top of things, then it will eventually be a choice between intel 6 core and amd 8 core, just that you'll have to fiddle with the mem more on amd.
I really don't think 6 and quad cores are gonna help much here, 200&400 mhz?
And neither will game enhancements..., it's just setting up a compilier to use avx and such..., which has been around for some time...
If they games don't have say avx2, just saying don't expect much...

Also even if the app/game is compiled to use everything you have, and no debug, inlining and other ops enabled (-02, -03, -0fast, etc)...
It's usually a minor increase in perf, anywhere from 1-4% on avg, and up to say maybe 40% in lucky cases.

It appears, supposedly..., that the win7 scheduler is better then win10's.
Make's sense to me... (I'm pretty sure ms is hiring millennials lol)
One thing though is, if the scheduler needs updated, then people on intel's will have added bloat to there code to support amd.
For ex., new kernel and bench's, amd gains perf, intel is not leading as much (intel lost a few points for larger code and amd got a fix).

Anyways, I think it's as simple as fixing up the scheduler to be a bit more liberal in numa mode.
It seems when you tell it to separate the 2 groups it refuses to fill up the 2nd half of threads using a single app.
Which may be fine actually for gaming, however, what about about say cinebench...?

I think games may have to focus on 8 threads, and no more.
And aren't some games now using up to 12threads or something?

I think this is kind of a screwy situation and I actually hope to see near ddr5 speeds with it in the future lol.

[Kal-EL]

They needed to release at the soonest possible moment in the release season. Every day is a pay day. I think they're already a success story considering the circumstances over the last decade.

We're all accustomed to crap ass bios revision at release and even circuit print jobs on rev 1 motherboards. The key is AMD sustaining thru the rocky release as its faint jab punch and pulverizing the market with its right hand power shot in Rev 2. boards, chips, bios support and industry adaptations and high end GPU release.

Then its another year for the market to consume product and push AMD into the black, upfund R&D and push deeper into Intel and Nvidia's pockets.

At this junction, any support from the consumer market only benefits the consumer in the end, death to the $1700 HEDT Chiprape. /4cents