AMD CPU speculation... and expert conjecture

[Cpus]

Hey guys. I found a site claiming to know the AMD Carrizo release date and specs (next gen kaveri)

http://allabouttech911.blogspot.com/2014/02/amd-carizo-release-date-and-rumors.html

 

[jdwii]

Mantle is not for hedt just as a apu is not for that purpose. The big gains mantle is producing is for lower-mid range systems. I am thinking a pll 965 is going to run mantle very good with that extra l3 cache or is it too slow to make a difference?

Me too want to see performance boost provided by mantle in case of phenom 2, is there any review in which they used any phenom chip

http://www.anandtech.com/show/7728/battlefield-4-mantle-preview
18% boost according to anandtech at least for my CPU

 

[jacobian]

If Mantle works the way it sounds, it's pretty smart move for AMD considering AMD's CPU's are slowly slipping behind Intel year after year. However, to get game developers interested in Mantle API, AMD needs to get at least Nvidia on-board. If Mantle is not something that's very AMD specific, I think Nvidia will provide its Mantle drives too.

 

[Cpus]

Hey guys. I found a site claiming to know the AMD Steamroller release date and specs

http://allabouttech911.blogspot.com/2013/04/amd-steamroller-release-date.html

 

[Master-flaw]

Hey guys. I found a site claiming to know the AMD Steamroller release date and specs

http://allabouttech911.blogspot.com/2013/04/amd-steamroller-release-date.html

LOL...here are the rest of this dudes blogs...
http://allabouttech911.blogspot.com/

 

[Cpus]

^ jared wilson has had a busy 24hrs, scraping all the post off the web. My av says malicious code being run on this blog.

I checked my virus scanner and went on the site but no malicious code showed up?

 

[-Fran-]

Now that I've given more thoughts to MANTLE and since one of you pointed out, but it seems no one noticed...

MIN FRAME RATES!

What happened with those? We can all agree that if we see an avg increase of 7% (for example), it will be amazing if the min frame rates increase in a significant amount. In other words, we can forget (a little) about averages and start asking about minimum ones, right?

So, given that, any numbers on those?

Cheers!

 

[palladin9479]

Now that I've given more thoughts to MANTLE and since one of you pointed out, but it seems no one noticed...

MIN FRAME RATES!

What happened with those? We can all agree that if we see an avg increase of 7% (for example), it will be amazing if the min frame rates increase in a significant amount. In other words, we can forget (a little) about averages and start asking about minimum ones, right?

So, given that, any numbers on those?

Cheers!

Mantle seems to be significantly reducing single thread congestion. Typically a game with have one to two heavy threads with a bunch of lighter threads doing different things on demand. The two primary threads are what drive the game and in scenes with intense effects going on can cause a bottleneck to appear. By optimizing those threads and coding them more effectively (seems to move some of the work off them and onto the lighter threads) you prevent those sudden bottlenecks from happening. Of course if your CPU's single threaded performance isn't causing your sporadic bottlenecks then Mantle won't do anything for you.

Essentially it's for fx6 and APU chips, to remove their greatest weakness (single threaded performance bottlenecks).

 

[noob2222]

@yuka

I posted the chart showing a regular interval of horrid lag spikes when using the mantle version of bf4 on the 7850k APU.

The pclab review was interesting because it shows the actual numbers instead of just the gains. Even under mantle, the 7850k is holding the 290x back by 22 fps, aka 27%.

 

[de5_Roy]

Now that I've given more thoughts to MANTLE and since one of you pointed out, but it seems no one noticed...

MIN FRAME RATES!

What happened with those? We can all agree that if we see an avg increase of 7% (for example), it will be amazing if the min frame rates increase in a significant amount. In other words, we can forget (a little) about averages and start asking about minimum ones, right?

So, given that, any numbers on those?

Cheers!

i haven't seen much of min. fps measurements. that's because the sites that posted early previews mainly offered shallow analyses. one of the possible sites that measured min. fps/frame time were
http://www.hardcoreware.net/battlefield-4-mantle-kaveri-performance/
for igp gaming.
and pcper's bf4 testing (slightly more in-depth)
http://www.pcper.com/reviews/Graphics-Cards/Battlefield-4-Mantle-Early-Performance-Testing
i don't know if it's been posted already.

the overzealous people keep forgetting that it's a beta release, so some issues will remain. despite the long awaited release, there's just too little to infer.... other than the observations that entry level cpus are the biggest gainers and the balanced/gpu-limited systems are gaining much less(using the weaker systems' gains as baseline). that's for one game only. for mantle to be worth considering outside it's niche, it should have at least 3-5 games from different genres, or total 10-20 games under it's optimization support so that people can buy/build a pc with proper mantle-boosted gaming in mind. that'll take at least a year to realize. amd will have to maintain a consistent debugging and maintenance cadence on top of that.


edit: (i'm not gonna use a seperate post, lazy i kno...)
the biggest takeaway from the catalyst 14 beta is that phase 2 of amd's frame pacing tweaks are now available for everyone
http://www.anandtech.com/show/7729/amd-catalyst-141-beta-drivers-now-available-mantle-frame-pacing-more
and that's a (very, very, very belated) good thing. compared to that, mantle seems subjectively negligible.

 

[gamerk316]

Now that I've given more thoughts to MANTLE and since one of you pointed out, but it seems no one noticed...

MIN FRAME RATES!

What happened with those? We can all agree that if we see an avg increase of 7% (for example), it will be amazing if the min frame rates increase in a significant amount. In other words, we can forget (a little) about averages and start asking about minimum ones, right?

So, given that, any numbers on those?

Cheers!

Mantle seems to be significantly reducing single thread congestion. Typically a game with have one to two heavy threads with a bunch of lighter threads doing different things on demand. The two primary threads are what drive the game and in scenes with intense effects going on can cause a bottleneck to appear. By optimizing those threads and coding them more effectively (seems to move some of the work off them and onto the lighter threads) you prevent those sudden bottlenecks from happening. Of course if your CPU's single threaded performance isn't causing your sporadic bottlenecks then Mantle won't do anything for you.

Essentially it's for fx6 and APU chips, to remove their greatest weakness (single threaded performance bottlenecks).

You're working on the assumption that the number of threads, and not the total workload, is at fault. I suspect CPU's like Pentiums and i3s are going to be FAR more effected then FX. Which would be ironic, for obvious reasons.

 

[gamerk316]

It would be interesting to see mantles effect on 2 cores, and hyperthreading

Already know the answer:

i3 gains about 25% performance, which likely puts it ahead of the FX-6350 again.

Kinda highlights how high level API's can choke performance, doesn't it? The more abstraction, the easier it is to code, but the more performance you lose. Hence why consoles like the PS3 and their glorified 7800 GTX can still pump out decent visuals.

Also... What could nVidia be thinking as an answer to MANTLE? It seems like they won't back AMD on this one (no surprise there), so will they push their own API now?

I hope not, otherwise you run the risk of going back to the days where some games simply don't support some hardware vendors. API wars are bad for the end user. More likely, they are hounding MSFT to do something in regards to DX, which it sounds like they are doing as we speak.

 

[-Fran-]

Well, nVidia can always pester MS to push a "close to metal" codepath in DX, but I really doubt MS will want regular PCs (specially cheaper ones) to be close in performance (read, visuals) to their own console. So it's a really weird thing for MS to accept so open heartedly an improvement like that in desktop. Also, remember nVidia bought 3DFX, the creators of Glide. What else does nVidia need to actually push a close to metal API like MANTLE? I do know they have one, but they don't push it to devs (that I know of). Maybe with TWIMTBP titles?

On the OGL side of things... I don't know if OGL has a "close to metal" path, but since OGL has industrial adoption, it would make sense to have something of the like. Too bad Khronos group keeps OGL so tied to the past... Will we ever see a OGL ES for desktop, replacing OGL as a gaming API? Man I wish we could... Maybe AMD can join Khronos, haha.

Anyway, from the pcper review posted by de5_Roy, I really believe MANTLE is a great thing to have. For instance, where they don't show improvements in FPS, they show improvements in smoothness, which is actually in the same scale of importance (if the graphs they use show that, of course) than pure FPS improvements. And min FPS were improved as well.

Cheers!

 

[Embra]

Preliminary results:

http://www.guru3d.com/articles_pages/amd_mantle_preview,1.html

 

[gamerk316]

Well, nVidia can always pester MS to push a "close to metal" codepath in DX, but I really doubt MS will want regular PCs (specially cheaper ones) to be close in performance (read, visuals) to their own console. So it's a really weird thing for MS to accept so open heartedly an improvement like that in desktop. Also, remember nVidia bought 3DFX, the creators of Glide. What else does nVidia need to actually push a close to metal API like MANTLE? I do know they have one, but they don't push it to devs (that I know of). Maybe with TWIMTBP titles?

On the OGL side of things... I don't know if OGL has a "close to metal" path, but since OGL has industrial adoption, it would make sense to have something of the like. Too bad Khronos group keeps OGL so tied to the past... Will we ever see a OGL ES for desktop, replacing OGL as a gaming API? Man I wish we could... Maybe AMD can join Khronos, haha.

Anyway, from the pcper review posted by de5_Roy, I really believe MANTLE is a great thing to have. For instance, where they don't show improvements in FPS, they show improvements in smoothness, which is actually in the same scale of importance (if the graphs they use show that, of course) than pure FPS improvements. And min FPS were improved as well.

Cheers!

Heres the issue: You can't get close to metal without exposing how the HW works, and NVIDIA is NEVER going to do that, at least in open drivers. And I don't want NVIDIA to give a new API, since we'll have MAJOR API wars, which are bad for the consumer.

It comes down to this: The more hardware you support, the more abstraction you have, and the more performance you lose. THIS IS NOTHING NEW. What Mantel does, essentially, is target ONE spec of hardware, allowing you to squeeze out all the optimizations you can. Same concept as Glide for 3dfx. The downside is you can only support ONE spec of hardware using the API. Hence why, when OGL and DX matured, and Glide mattered less, 3dfx faded away.

 

[gamerk316]

Preliminary results:

http://www.guru3d.com/articles_pages/amd_mantle_preview,1.html

Kinda surprised how little APU's got, especially compared to dGPU's. GPU performance bottleneck is in clear effect for APU's.

I'm REALLY interested in seeing how the i3 does now. i3's may make a comeback if these numbers hold.

 

[jdwii]

It would be interesting to see mantles effect on 2 cores, and hyperthreading

Already know the answer:

i3 gains about 25% performance, which likely puts it ahead of the FX-6350 again.

Kinda highlights how high level API's can choke performance, doesn't it? The more abstraction, the easier it is to code, but the more performance you lose. Hence why consoles like the PS3 and their glorified 7800 GTX can still pump out decent visuals.

Also... What could nVidia be thinking as an answer to MANTLE? It seems like they won't back AMD on this one (no surprise there), so will they push their own API now?

I hope not, otherwise you run the risk of going back to the days where some games simply don't support some hardware vendors. API wars are bad for the end user. More likely, they are hounding MSFT to do something in regards to DX, which it sounds like they are doing as we speak.

Now gamer come on their is no 6 core fx(its anandtech of course) and for another thing a 6 core has more overall performance when all cores are used.

 

[de5_Roy]

Kinda surprised how little APU's got, especially compared to dGPU's. GPU performance bottleneck is in clear effect for APU's.

that should be obvious. from the early tests, gpu bottleneck nearly erases mantle's so called advantage over directx, even in discreet gfx.

I'm REALLY interested in seeing how the i3 does now. i3's may make a comeback if these numbers hold.

pclab.pl used core i3 with a radeon r9 290x for bf4 mp. it gaiend so much that it went over stock fx8350 and kaveri 7850k.... or something. i only looked at the benchmark images, not the text.
the article is on the first page of pclab.pl site.

 

[-Fran-]

I don't think MANTLE is exposing the inner parts of GCN's hardware, since it's a programmable uarch (much like X86 and ARMv7 is, AFAIK) making the API abstract enough to hide the actual hardware. This is just conjecture out of what AMD has explained MANTLE to be, but I'm pretty sure nVidia won't have to open their cards legs to devs if they don't want to with MANTLE.

And given that, I think we need to see some lines of code from a MANTLE project to keep talking how close or far it is to "tha metal", heh.

Also, yes, that's what happened to Glide, among another important thing that is being said to be the real nail in the coffin for them: 32bits and textures. This was in the final years of the 90's and my memory is not that good, but I do remember reading about the GeForce 256 and how it trashed the Voodoo 3 and even 4 and 5 parts for a good 2-3 years, specially since it had hardware T&L (first shader-like spec? not sure). OGL and DX catching dev's attention was just a side effect of 3DFX not having all the goodies and relying solely on Glide to shine. If AMD makes that same mistake, they'll deserve the same destiny as 3DFX IMO, since what better example of what not to do than Glide's downfall

Cheers!

 

[truegenius]

Now that I've given more thoughts to MANTLE and since one of you pointed out, but it seems no one noticed...

MIN FRAME RATES!

What happened with those? We can all agree that if we see an avg increase of 7% (for example), it will be amazing if the min frame rates increase in a significant amount. In other words, we can forget (a little) about averages and start asking about minimum ones, right?

So, given that, any numbers on those?

Cheers!

anyone noticed yuka's signature (keeping mantle in mind)
"The era of 'you just need a dual core' is now ending." .- palladin9479

 

[gamerk316]

It would be interesting to see mantles effect on 2 cores, and hyperthreading

Already know the answer:

i3 gains about 25% performance, which likely puts it ahead of the FX-6350 again.

Kinda highlights how high level API's can choke performance, doesn't it? The more abstraction, the easier it is to code, but the more performance you lose. Hence why consoles like the PS3 and their glorified 7800 GTX can still pump out decent visuals.

Also... What could nVidia be thinking as an answer to MANTLE? It seems like they won't back AMD on this one (no surprise there), so will they push their own API now?

I hope not, otherwise you run the risk of going back to the days where some games simply don't support some hardware vendors. API wars are bad for the end user. More likely, they are hounding MSFT to do something in regards to DX, which it sounds like they are doing as we speak.

Now gamer come on their is no 6 core fx(its anandtech of course) and for another thing a 6 core has more overall performance when all cores are used.

Ahh, but if the CPU isn't a bottleneck, it isn't going to see as much performance gain. The i3, with far fewer resources, therefore has the most to gain by lessening CPU load...

pclab.pl used core i3 with a radeon r9 290x for bf4 mp. it gaiend so much that it went over stock fx8350 and kaveri 7850k.... or something. i only looked at the benchmark images, not the text.
the article is on the first page of pclab.pl site.

http://pclab.pl/art55953.html

Going to post this:

Versus:

PClab also made another good point: DX11.2 did a bit to decrease the API overhead of DX, and BF4 gains SIGNIFICANT performance on Win8:

So benching should ideally be done on Windows 8, to get a fair comparison.

 

[gamerk316]

You can see on my above where the GPU bottleneck comes into play, at about 80ish FPS or so. Note how the i3 is now competitive with with the OC'd 4770k? Core i3 is the big winner here.

Big loser of Kaveri, which still falls behind the Phenom II X4.

 

[de5_Roy]

yeah, it's nice that fx is finally in the same league as core i5 thanks to mantle. LOL

kaveri's gain is quite impressive, stock a10 7850k gains almost 2x avg. fps with mantle - 31.9 fps vs 58.9 fps. o.c. 7850k is actually faster than o.c. ph ii x4 965.

smoothness should be the important factor here, instead of avg. fps. i saw in the pcper article where frame times decreased with mantle. higher fps won't mean much if it still stutters.

 

[logainofhades]

pclab has some bf4 mantle cpu benches. look at the haswell pentium and core i3 go after 7850k and fx8350!
http://pclab.pl/art55953.html
i wish i could translate this... i'll read it after i fix my browser.

^ 7% @ 4000
"In the graphics limit the 4K resolution (3,840 x 2,160 pixels) overloaded the Mantle API the Radeon R9 290X little better, the average frame rate increases by 7 percent."

http://www.golem.de/news/amds-mantle-api-im-test-der-prozessor-katalysator-1402-104261-3.html

might mean that mantle won't help in a typical cpu-bound situation like online mp shooters if the graphics subsystem becomes bottleneck, no matter how high end the gfx card is. you must lessen the load on the gfx hardware for mantle to deliver.

I would take PClab's results with a grain of salt. They have been known to be quite the intel fanboy website.

Sounds like Mantle will be good for the budget crowd, at the very least. That I see is a good thing. The more PC gamers there are the better.

 

[de5_Roy]

I would take PClab's results with a grain of salt. They have been known to be quite the intel fanboy website.

that can be said for any website that shows, correctly or otherwise, intel cpus ahead of amd cpus.