AMD CPU speculation... and expert conjecture

[Cazalan]

Special mention : Broadwell chips will only have PCIE2.0-x4 lanes. So big dGPU's will be completely gimped.

It was to be expected for the mobile parts. It's part of the natural progression moving to SoC and a power conscious consumer. Expect the system RAM to be integrated next.

I think Charlie overstated the sinisterness of Intel's choices for that segment. APUs already put a hefty dent there anyway so the add-in market has been on the decline for a while now.

The workstation/server versions of Broadwell will have to have more lanes to support Intels own Xeon Phi and other high performance PCIe cards. That segment won't go away but it will move some. Desktops will likely be workstation/server only by mid-2014.

This could end up being good for AMD. Even if their CPU is slower they can offer cheap dual/triple/quad crossfire. You can bet where the budget enthusiasts will go.

 

[mayankleoboy1]

quote me where i said its pointless to update software for old driver.

Makes it a little hard to update software for hardware that doesn't exist yet don't you think?

Yeah.Right. AMD driver writers have to order HD89xx from newegg before they start writing driver for it. :lol: . You really are a noob, arent you.

OK. Let me say it loud and clear, because you dont read whats written: I believe the only reason AMD have come up with this rewrite of VM manager for GCN based products, so late on the cycle, is because :

a) new code is a part of their ongoing work for GCN+ drivers
and/or

b) this driver will be reused in their GCN+ refresh.

I cant make it any simpler, Noob. Seriously, try the book link i gave above. Will work wonders for you.

 

[de5_Roy]

this has been the least amusing argument about semantics this week on teh internet.
i have a piece ready on how to champion amd gpus over nvidia and intel gpus (very helpful for calfs). i know this is not the topic of this thread. slow week and all....

AMD Introduces Modular "Roadrunner" Server Platform That Simplifies Data Centers.
AMD’s Open 3.0 Platform to Let IT Professionals to Quickly Optimize Servers for Specific Workloads
http://www.xbitlabs.com/news/cpu/display/20130116220710_AMD_Introduces_Roadrunner_Server_Platform_That_Simplifies_Data_Centers.html
AMD details Open Compute server ecosystem
http://semiaccurate.com/2013/01/16/amd-details-open-compute-server-ecosystem/
because it has 'global foundries' in it
http://semiaccurate.com/2013/01/17/st-ericksson-puts-out-a-2-5ghz-phone-chip/

 

[noob2222]

Whats the point in bitching about it then?

Quite late in the product lifetime cycle to release a new Video memory manager, no?

so you made up something on your own thats not even in the article, seroiulsy what are they supposed to say? "We are updating the drivers for our radeon 7000 and radeon 8000 cards" that way people like you won't get confused as to why they are releasing an updated driver?

then again, you can't coprehend:

Dodd says additional improvements are coming down the pike, including a rewrite of the software memory manager for GPUs based on the Graphics Core Next architecture that should bring a more general improvement.

Exactly why do you need it pointed out that radeon 7000 and radeon 8000 are both GCN architecture? Anone with a grade school reading level would know that. Instead you come up with the assumption that GCN does not include GCN refresh. And that the only possible way to ever write a driver is to support only gcn refresh.

Or maybe GCN Refresh can benefit from that too...

Wow, a GCN card benefiting from a GCN update? you actually needed that pointed out to you that a GCN card uses GCN drivers regardless of whether its gcn 1.0 or gcn 1.5 or gcn 2.0? But they didn't write an entire article about that and left you confused.

 

[-Fran-]

Uhm... Didn't AMD release the Video Memory Manager before the 7K series launch? Too bad (like gamerk says) they took so long to improve it anyway.

You guys should resolve your sexual tension in PMs, really. hahaha

Cheers!

EDIT: Forgot to add:

 

[JAYDEEJOHN]

Yea, enough guys
I declare 2 winners, now kiss n make up, or theyll be 2 losers

 

[blackkstar]

I'm more worried that AMD didn't even bother to investigate these performance anomalies until TR brought the topic up. Heck, apparently even NVIDIA started only two years ago.

Still, glad to see this is FINALLY getting attention.



And in response to Blackstar: Doesn't the latest round of benchmarking by TR now put AMD ~ NVIDIA in these titles, minus 2-3 FPS or so? Maybe, I don't know, the DRIVER was the problem the entire time? If AMD has a driver problem (they did), I'd rather it be exposed and they fix it, rather then blaming NVIDIA and continuing to push a broken driver.

I definitely think the drivers have a lot of room for improvement. From what I have heard, AMD driver team was rushed to make dead lines, and that's where a lot of this crappy code in the drivers comes from. Not because AMD has crappy programmers or anything, but because they had inept marketing guys breathing down their necks telling them to hurry up.

FPS-wise Tahiti looks very good compared to GK104 and Pitcairn looks great compared to GK106. However, this whole frame time thing is some what new. No one else other than Tech Report really knew about this and for as long as I've known, FPS has been the deciding factor in a gaming benchmark.

So, AMD updates their drivers and they see FPS go up, and they think they're doing good. I don't really think it's their fault. It's like we've got a more accurate way of measuring a qualitative attribute of something that was just invented, and people say it's AMD's fault for not knowing about it?

It's like living in a world where we have no way of measuring radiation, giving someone x-rays, discovering a way to measure radiation, and then going, "you're so stupid for exposing people to all this radiation from x-rays! Why didn't you do something earlier about it!"

You can't solve a problem when you don't know the problem exists.

As a little FYI, NVIDIA and ATI/AMD have always had different approaches to GPU's. ATI/AMD always had the edge in memory bandwidth, NVIDIA always had the edge in shader performance. Problem for ATI/AMD, prior to DX9, performance was dominated by shaders, hence NVIDIAs dominance. But while NVIDIA was busy making the Xbox GPU, ATI/AMD successfully lobbied to get several features added to the DX9 spec whose performance was dominated by memory bandwidth. Low and behold, as a result, the NVIDIA FX series STANK in any DX9 benchmark, due to its poor memory bandwidth compared to AMD.
[/quote]

And there's the problem. Reviews are just, "oh look this has the bigger graph!" There is very little technical explanation. Hardly anyone says, "well, GTX 670 has more shader power so it did better here, however 7950 has more bandwidth so it'll handle higher resolution and MSAA better."

It's just "WELL THE WINNER IS XXXX BECAUSE THE AVERAGE FPS OR SPF WAS HIGHER IN THE GAMES WE RAN!" It's completely unscientific and unreliable.

I mean, I could make an FX 8350 review that was Civ5 render, BF3, AvP, x264, etc and say FX 8350 is better than 3570k. The average of all those benchmarks would be better than Intel's more than likely.

As a result, when you see similar performing GPU's (say, 7950 and 660 TI), and one easily outperforms on game "X", you can usually assume that game is driven by Shader performance (if NVIDIA is better) or memory bandwidth (if ATI/AMD is better). For the most part, there aren't that many tricks in the DX API you can use to move performance one way or the other. Aside from games that are dominated by one type of performance metric, you aren't going to see a lot of variance between similar GPU's of the same generation, regardless of vendor, unless theres an underlying driver issue.[/quote]

I definitely agree, and how review sites handle these issues and the tendency I've noticed with armchair enthusiasts to look at graphs only and go "OMG THIS HAS BIGGER BARS SO ITS BETTER!!!!" needs to be resolved for AMD.

I realize this feels unrelated to Steamroller, but it plays a much bigger part than people are giving credit, because these same kinds of practices apply to CPU reviews as well. Intel has its strong points and AMD has their own.

But AMD just gets bullied around. I wonder if AMD's reluctance to push desktop steamroller has to do with the fact that even if they do make a good product, it's just gonna get relegated to budget because of how reviews work.

 

[JAYDEEJOHN]

It seems, as both companies went into a more compute direction in their archs is when they started working on this.
Early on nVidia also had these problems with a more compute HW solution.
As time went by, they concentrated on this smoothing.
AMD, having their first true heavy compute GPU, is also now going down this path.
No ones at fault here, it just takes time with new HW and therefore all new drivers as well.
Just compare this to Intels struggle with their drivers, and the resources they have.

Look here
http://techreport.com/review/22653/nvidia-geforce-gtx-680-graphics-processor-reviewed/8
to see, nVidia had problems of its own in the past.
Im glad this is being done, as it only enhances our usage

 

[sarinaide]

I think this is going off topic but what really annoys me about GPU reviews is that they attempt to pidgeonhole GPU's because it is intended competition at a price point.

All AMD GPU's are lower clocked on the GPU component and active VRAM speed these are significant in factoring into the equations. A HD7970 GE is lower clocked than a Vanilla GTX680 but they are reviewed against each other despite the price difference as well. The AMD cards are generally cheaper as well unless you take a Vanilla GTX against a XFX DD in which case its more expensive but thats all ancillary.

The basic crux here is that Nvidia didn't comprehensively win despite a much faster GPU, AMD though closed the gap to the point its neck on neck just on drivers alone which is to me more impressive.

It may seem like a stupid reason to people but I chose GCN over Keplar for "Power Core technology" Run a HD7970GE at full tilt using up to 10-20%(manufacturer model dependant) less power on the drop of a slider. Tried a similar thing with the GTX680 lowering its power trying to run at full capacity it crashes another testiment to the architectural level of GCN

 

[gamerk316]

I definitely think the drivers have a lot of room for improvement. From what I have heard, AMD driver team was rushed to make dead lines, and that's where a lot of this crappy code in the drivers comes from. Not because AMD has crappy programmers or anything, but because they had inept marketing guys breathing down their necks telling them to hurry up.

FPS-wise Tahiti looks very good compared to GK104 and Pitcairn looks great compared to GK106. However, this whole frame time thing is some what new. No one else other than Tech Report really knew about this and for as long as I've known, FPS has been the deciding factor in a gaming benchmark.

So, AMD updates their drivers and they see FPS go up, and they think they're doing good. I don't really think it's their fault. It's like we've got a more accurate way of measuring a qualitative attribute of something that was just invented, and people say it's AMD's fault for not knowing about it?

It's like living in a world where we have no way of measuring radiation, giving someone x-rays, discovering a way to measure radiation, and then going, "you're so stupid for exposing people to all this radiation from x-rays! Why didn't you do something earlier about it!"

You can't solve a problem when you don't know the problem exists.

The first point shows me AMD has a corporate culture that will eventually collapse the company.

As for the second, my point wasn't so much the way AMD is benchmarking, but the point they assumed those high latency frames were the fault of the OS, and didn't even bother to investigate the problem (which likely would have shown exactly what TR reported). When dealing with something this complicated (OS/Driver/HW interaction), any anomalies must be investigated and understood; this is basic testing methodology. The fact this was not done is very, very worrying for me. [FYI: I do this type of stuff on a daily basis. Its not enough to see a problem, it must be dissected and understood, not shoved away into some dark corner].

 

[-Fran-]

I think this is going off topic but what really annoys me about GPU reviews is that they attempt to pidgeonhole GPU's because it is intended competition at a price point.

All AMD GPU's are lower clocked on the GPU component and active VRAM speed these are significant in factoring into the equations. A HD7970 GE is lower clocked than a Vanilla GTX680 but they are reviewed against each other despite the price difference as well. The AMD cards are generally cheaper as well unless you take a Vanilla GTX against a XFX DD in which case its more expensive but thats all ancillary.

The basic crux here is that Nvidia didn't comprehensively win despite a much faster GPU, AMD though closed the gap to the point its neck on neck just on drivers alone which is to me more impressive.

It may seem like a stupid reason to people but I chose GCN over Keplar for "Power Core technology" Run a HD7970GE at full tilt using up to 10-20%(manufacturer model dependant) less power on the drop of a slider. Tried a similar thing with the GTX680 lowering its power trying to run at full capacity it crashes another testiment to the architectural level of GCN

To be fair, when AMD released the 7970, the GTX680 was like 3 months away at the time and they used the "i'm alone at the top" pricing scheme. They're a company, they should do it, right? Anyway, my point it when the GTX680 came up, it came with better drivers and performed nicely on par with the 7970 at stock speeds, being withing strike distance in price. Also, when you review "flagship vs flagship", price doesn't play a big factor in it; you just want to know how far from each other are the best effort card for each camp. Same with CPUs and most hardware out there.

If you take a look now, AMD has come a long way with GCN and like I've always known, ATI/AMD hardware has always been top notch, being superior from nVidia in a lot of important areas over the time, but their drivers and software have always lacked horrendously (although all the Radeons I've had before were hassle free) compared to nVidia, but mostly games though (where it matters to 99% of folks); TV and Video is a whole world of difference in favor to ATI/AMD IMO. When CUDA came out though, video processing took an important step in favor of quality thanks to nVidia, I'll give and thank them that.

Anyway, regarding AMD as a CPU company, with games, they have to put more effort in the "Gaming Evolved" initiative. Deus Ex and Dirt 3/Showdown run GREAT on AMD hardware; or at least, better than Starcraft and most Blizzard games on AMD though, hahaha.

Cheers!

 

[mayankleoboy1]

Gaming Evolved is not to optimize games for AMd, but to not to optimize games for Nvidia.

GPU superiority in many games last only till 2 or 3 driver versions. So most TWIMTBP and Gaming evolved programs mean that the game devs gave the game betas/alphas to the GPU maker, so that the driver team can optimjze their drivers.

 

[esrever]

Gaming Evolved is not to optimize games for AMd, but to not to optimize games for Nvidia.

GPU superiority in many games last only till 2 or 3 driver versions. So most TWIMTBP and Gaming evolved programs mean that the game devs gave the game betas/alphas to the GPU maker, so that the driver team can optimjze their drivers.

Actually both gaming evolved and twimtbp tries to help developers add dx11 functionalities and pc oriented techniques to improve game performance and visual quality. Nvidia has tried to sabotage AMD with using physx and absurd amounts of tessellation. I have yet to read anything about AMD making the game run worse on the competitor's platform. AMD as far as I know have been pushing directcompute and other general compute techniques into PC gaming.

 

[esrever]

http://www.eurogamer.net/articles/df-hardware-orbis-unmasked-what-to-expect-from-next-gen-console

Both the next generation PlayStation - and its Xbox competitor - feature eight-core CPUs clocked at 1.6GHz according to sources trusted by Digital Foundry. The main processor architecture driving both consoles is said to be derived the new "Jaguar" technology currently in development by Intel's arch-rival, AMD.

 

[Cazalan]

http://www.eurogamer.net/articles/df-hardware-orbis-unmasked-what-to-expect-from-next-gen-console

Both the next generation PlayStation - and its Xbox competitor - feature eight-core CPUs clocked at 1.6GHz according to sources trusted by Digital Foundry. The main processor architecture driving both consoles is said to be derived the new "Jaguar" technology currently in development by Intel's arch-rival, AMD.

Good news for AMD if true, but those clock speeds are not very impressive. We know how inefficient software devs have been at supporting 4+ CPUs.

 

[JAYDEEJOHN]

I thoughr I saw an estimated 8x perf all around on this theoretical setup.
Its supposed to fit into a tighter TDP for lessor/no heat problems

 

[viridiancrystal]

Good news for AMD if true, but those clock speeds are not very impressive. We know how inefficient software devs have been at supporting 4+ CPUs.

If MS and Sony are putting in 8 cores at 1.6, devs either told them they were okay with doing that, or, if they didn't have a say, they will have to be. They don't really have a choice.
Huge news for AMD if it holds true though. Nearly a clean sweep of this console gen's gpus and cpus.

 

[esrever]

Good news for AMD if true, but those clock speeds are not very impressive. We know how inefficient software devs have been at supporting 4+ CPUs.

They dealt with programming for the cell. Programming for symmetric x86 cores should be a cake walk.

 

[sarinaide]

Gaming Evolved is not to optimize games for AMd, but to not to optimize games for Nvidia.

GPU superiority in many games last only till 2 or 3 driver versions. So most TWIMTBP and Gaming evolved programs mean that the game devs gave the game betas/alphas to the GPU maker, so that the driver team can optimize their drivers.

Nvidia and Intel have been using gaming partners for a very long time, why is AMD "GamingEvolved" partners supposed to be sabotaging competitors. All it does is maximize the strength of AMD Graphics using GPGPU performance under MS direct compute where AMD is very strong, but in no way limits Nvidia or Intel performance.

Partnerships are important and this is no less different to what team green has been doing for a very long time, I don't buy this belligerent sabotage nonsense.

 

[de5_Roy]

If MS and Sony are putting in 8 cores at 1.6, devs either told them they were okay with doing that, or, if they didn't have a say, they will have to be. They don't really have a choice.
Huge news for AMD if it holds true though. Nearly a clean sweep of this console gen's gpus and cpus.

afaik, these cpus and gpus will be customized enough so that sony, ms and nintendo will get larger parts of the revenues.
i think amd fired the guy who got them the console design wins. :lol:

 

[mayankleoboy1]

amd hasnt earned much from these console wins, have they?

 

[-Fran-]

From all the people I read about the subject, yeah; that's the overall impression.

It's funny though. The person who managed to get the "wins" for the designs will have on his resume (curriculum vitae?) "successful negotiation for important CPU company and International companies in a console design win", when all he/she did was put down the pants and open the legs / bend over, lol.

Anyway, at least it's good to have hardware out there, you can at least show your stuff to people. The "ah, this doesn't suck as people said on articles!" and stuff. And maybe AMD CPUs will get a little more love from devs/publishers.

Cheers!

 

[montosaurous]

If AMD can get within 10% of Intel in terms of per core performance then MAYBE I will consider upgrading instead of buying an Intel chip in the next 2 years. The most I will ever use is 4 threads, so I will only buy a Quad core CPU. Any AMD CPU is not worth the upgrade in per core performance from my current chip, so unless Steamroller offers near Haswell/Broadwell per core performance my next chip is probably going to be Intel. Don't get me wrong I don't hate AMD, in fact I'd like to see them beat the *** out of Intel again like they were years ago but I just can't see that happening.

 

[de5_Roy]

If AMD can get within 10% of Intel in terms of per core performance then MAYBE I will consider upgrading instead of buying an Intel chip in the next 2 years. The most I will ever use is 4 threads, so I will only buy a Quad core CPU. Any AMD CPU is not worth the upgrade in per core performance from my current chip, so unless Steamroller offers near Haswell/Broadwell per core performance my next chip is probably going to be Intel. Don't get me wrong I don't hate AMD, in fact I'd like to see them beat the *** out of Intel again like they were years ago but I just can't see that happening.

steamroller will be 28nm chip while haswell is 22nm and broadwell will be 14nm chips. amd is already at a process disadvantage. if intel sticks with current trend of single digit cpu performance improvement while improving power efficiency, then amd might have a chance to catch up to intel's cpu performance if they're willing to compromise power efficiency. that's sorta what happened with piledriver and ivy bridge. intel didn't improve cpu performance much, but amd tweaked their bd arch and had it manufacture on a more mature 32nm fabrication process, stuck with their current power limits thus closed the gap a bit (which drove amd fanboys to rejoice at the top of their voices).
amd is already incapable of competing with intel which is why we have $500 and $1000 6 core performance cpus. it won't change any time soon....
amd knows this and they're competing against intel in the laptop markets where they succeeded briefly and then managed to screw themselves over. long story short, amd knows they're better equipped to compete (they're depending on radeon) in the lower-mid to entry level segment and portable (apus, lappy, tablet etc) and that's where they have been aiming for the past two years or so.

 

[gamerk316]

Good news for AMD if true, but those clock speeds are not very impressive. We know how inefficient software devs have been at supporting 4+ CPUs.

Well, the next Xbox still has a LOT of roumers on the CPU front; IGN says PPC, S/A claims X86.