AMD Wants To Stop Being Known As The “Cheaper Solution”

[rtsurfer]

It got nothing to do with AMD. But everything to do with performance/watt and the 2:1 ratio for designs.

Remember, mobile chips. Not desktop chips.

But wouldn't more Cache help on mobile too.?

Or are you suggesting that more Cache would consume more power. ?

Even if that were true, they could just add more to desktop chips. They don't have the same die as mobile chips, right.? So if you are designing a chip for mainstream desktop, there is no reason not to put more Cache. The HEDT platform chips are server based & so they have more Cache like server chips.

Also this reminds me that previous gen 4930K & 4960X differed solely on the amount of Cache they had. But the 4960X still wasn't much faster, if at all.

 

[ShintaiDK]

But wouldn't more Cache help on mobile too.?

Or are you suggesting that more Cache would consume more power. ?

Even if that were true, they could just add more to desktop chips. They don't have the same die as mobile chips, right.? So if you are designing a chip for mainstream desktop, there is no reason not to put more Cache. The HEDT platform chips are server based & so they have more Cache like server chips.

Also this reminds me that previous gen 4930K & 4960X differed solely on the amount of Cache they had. But the 4960X still wasn't much faster, if at all.

Cache is relatively power hungry

There is no such thing as desktop chips. Havent been since Core 2. Its the exact same dies.

4930K vs 4960X may be other limitations. Cache speed still matters, not just size.

But you still have acses with relatively large difference:

 

[rtsurfer]

Cache is relatively power hungry

There is no such thing as desktop chips. Havent been since Core 2. Its the exact same dies.

Okay. Makes sense. I'd say, screw you mobile, but than I am itching to buy an Ultrabook, so that would be Hypocritical of me.

And how about 4930K & 4960X. The Cache differences there didn't net to much tangible.

 

[ShintaiDK]

See above. But the question is uncore speed, cache speed, ringbus path etc. And the relatively small difference 12 vs 15MB.

Remember in 4930K its just 0.5MB of each L3 slice thats disabled. While on 5930K its 2 complete slices.

But yes, mobile is the evil nemesis.

 

[rtsurfer]

See above. But the question is uncore speed, cache speed, ringbus path etc. And the relatively small difference 12 vs 15MB.

I see.
But it still appears to me that in term of raw gaming FPS, the benefit of adding cores would be more than Cache. Actually productivity & computation work are more capable of using the cache, hence the reason Xeon chips have large Cache.

So I think more cores for gaming is still a better strategy, provided we get multi-threaded games.

 

[ShintaiDK]

I see.
But it still appears to me that in term of raw gaming FPS, the benefit of adding cores would be more than Cache. Actually productivity & computation work are more capable of using the cache, hence the reason Xeon chips have large Cache.

So I think more cores for gaming is still a better strategy, provided we get multi-threaded games.

But its not. The 5960X vs the 4790K is a prime example

And I wouldnt have any faith in the coding part. The amount of serial code is simply too great by nature. It may have some effect for large multiplayer games. But else no.

The main purpose of Broadwell-E for desktop is to increase the clock of the 8 core to avoid any embarrassing cases where the cache cant offset the lower core performance against lower core amount models.

 

[rtsurfer]

But its not. The 5960X vs the 4790K is a prime example

And I wouldnt have any faith in the coding part. The amount of serial code is simply too great by nature. It may have some effect for large multiplayer games. But else no.

5960X has both more Cores & Cache.

Which just proves that the games are too badly coded to take advantage of either of architectural improvements..

 

[Pilum]

It wasn't the first time that AMD gave false predictions to investors regarding performance of their processors, Barcelona, Bulldozer, Richland, Kaveri, just to name a few, so I think this 40% number should be taken with huge quantities of salt, and then there's the issue of their foundry partner, that screwed every single node they tried to develop or implement.

Where did AMD talk about performance? AFAIK they only talked about IPC (which is completely meaningless without a frequency/performance target). Personally, I think +40% IPC @"we're not telling" GHz is completely believable.

 

[mrmt]

Where did AMD talk about performance? AFAIK they only talked about IPC (which is completely meaningless without a frequency/performance target). Personally, I think +40% IPC @"we're not telling" GHz is completely believable.

IPC is one metric of performance.

 

[nyker96]

I personally think this is a lost cause and not forsaken market. Unless AMD is thinking about using a high performance core to scale down all the way to CPUs that can be used in tablets etc. Nothing wrong w/ high performance but nowaday what drives the sells number isn't personal PCs but rather smaller mobile devices. I haven't seem too many people putting their money on PCs in the past 5 years, everyone I know mostly have a desktop for web browsing, but not even that, most are using a smaller tablet device for movies/web/pictures/social media.

I'd say maybe 10 years back CPU and desktop is important markets, not any more, to dive back into a shrinking market in my opinion is wrong move. They should be investing more on low power, high performance cpu and platforms for mobile devices.

 

[Atreidin]

IPC is one metric of performance.

OK but that's not the point, AMD slides specifically state 40% higher IPC, not general performance. I'd rather make this clear before everyone starts running with "40% better (absolute) performance" and then if AMD doesn't hit that target, have the forum flood with a bunch more negativity about AMD missing a target they never set for themselves in the first place.

 

[mrmt]

OK but that's not the point, AMD slides specifically state 40% higher IPC, not general performance. I'd rather make this clear before everyone starts running with "40% better (absolute) performance" and then if AMD doesn't hit that target, have the forum flood with a bunch more negativity about AMD missing a target they never set for themselves in the first place.

So we are supposed to take the overall performance number with a grain of salt because AMD lied before, but we should take the 40% IPC number at face value only because it's not an overall performance number, but rather a component of it?

 

[LTC8K6]

OK but that's not the point, AMD slides specifically state 40% higher IPC, not general performance. I'd rather make this clear before everyone starts running with "40% better (absolute) performance" and then if AMD doesn't hit that target, have the forum flood with a bunch more negativity about AMD missing a target they never set for themselves in the first place.

Perhaps they should put out accurate slides then?

The arrow labeled 40% actually indicated about 250%, I think.

 

[ElFenix]

40% better IPC at what clocks? and on what sort of test?

 

[ShintaiDK]

40% better IPC at what clocks? and on what sort of test?

I dont think they will tell before late 2016/early 2017. It looks like its the pivot to drive the company until then. But I have a feeling its not going to be the 40% thats hoped for.

 

[TeknoBug]

AMD isn't the "cheaper solution" to me. My athlon 500MHz was faster than a P3 550, and my athlon xp 1900+ was the fastest box at my college. And they held a lead for many years after.

They still have a niche, and may play second fiddle these days, but anyone who only knows them as the "cheaper option" doesn't know much about PCs.

So you're stuck in the past, everyone that has built and used both AMD and Intel 10+ years ago knew AMD was ahead but this is today, Intel is ahead in many ways but AMD's latest APU's has been an improvement- the integrated GPU with enough muscle for midrange gaming is surprisingly good. So hopefully AMD catches up this year or next year.

My i7 3770K I recently sold has been an amazing machine since I undervolted it and still overclocked it. My FX 8320 wasn't anything all that impressive but still did the job.

 

[PG]

I don't think that the 40% claim is unrealistic at all.

It was pretty well known that there was a 20% penalty due to CMT, and ZEN has moved to SMT, so that's half the 40% figure right there.

Zen is also said to have cache improvements, which was another sore spot for Bulldozer/Vishera. I'm not sure what this will gain, but it should really help.

There are other improvements I'm sure, and it's hard to know what is rumor and what will be real, but add this all together and it should mean a massive jump in performance.

When people talk IPC I think Cinebench 11.5 single thread performance. AMD has been stuck at around 1 to 1.1 with Phenom II and even Bulldozer/Vishera/Kaveri.
Bump that up by 40% and we still aren't at Haswell, but it's a lot more competitive.
Haswell i5 or i7 is around 1.7 or so.

To be honest I hope for more than a 40% jump, but we will see.

 

[DrMrLordX]

I don't think that the 40% claim is unrealistic at all.

It was pretty well known that there was a 20% penalty due to CMT, and ZEN has moved to SMT, so that's half the 40% figure right there.

Zen is also said to have cache improvements, which was another sore spot for Bulldozer/Vishera. I'm not sure what this will gain, but it should really help.

Bear in mind that AMD's statements are relative to Excavator, not Bulldozer; Piledriver; or bugfixed-Bulldozer-pretending-to-be-Piledriver. Some of the more glaring deficiencies of Piledriver have been smoothed out on Steamroller and (presumably) Excavator. 40% over Excavator is significant.

I agree that the cache improvements are probably a lot of what gets them there. Excavator in its present/near future form has no L3 (Steamroller and Excavator managed to improve on Piledriver despite having no L3, ha ha). If AMD can re-introduce L3 to their chips without making it the slow, bulky liability it was on Vishera, then hey, great. They've already stated they're shifting to an inclusive cache hierarchy which ought to speed things up nicely. Maybe the 14nm process has helped them to deal with poor cache density (an area where Intel has been kicking AMD's butt for years).

When people talk IPC I think Cinebench 11.5 single thread performance.

Don't. Cinebench, especially 11.5 and 15, are closed-source binaries that have unknown optimizations for various microarchitectures. You should be looking at open-source software that is optimized as well as can be for any given modern uarch. Or at least something closed-source that you can trust has been tuned for anything and everything under the sun (y-cruncher is an excellent example of this).

AMD has been stuck at around 1 to 1.1 with Phenom II and even Bulldozer/Vishera/Kaveri.

. . . but in other applications, AMD has actually managed higher IPC with Kaveri than Bulldozer, Vishera, or Stars. People just keep going back to Cinebench for whatever odd reason.

 

[SlowSpyder]

Piledriver's IPC falls somewhere between a CD2 and Nehalem, right? I bet jumping 40% performance per clock isn't as difficult as it might sound, they're starting with the bar set fairly low. If they can get mid - upper 3GHz clocks they can match current core / module performance at what would likely be much lower power use.

AMD has some good IP to share and build from. They should market to casual gamers, let them know that AMD powers the graphics in all three consoles as well as the CPU's in the two big ones. I think they're in a good position to gain a lot of ground as long as Zen is good an the R9 390X (presumed name?) is the part many of us hope it will be.

 

[senseamp]

If (when?) AMD liquidates, I believe that anti-trust authorities will force Intel to license x86 on FRAND terms. Then you will see meaningful competition in the x86 space, with big boys like Samsung putting their resources into the game. Until then, enjoy never ending hail-Mary passes from a cripple and fantasizing about competition in the x86 CPU space.

 

[mrmt]

Piledriver's IPC falls somewhere between a CD2 and Nehalem, right? I bet jumping 40% performance per clock isn't as difficult as it might sound, they're starting with the bar set fairly low.

If it's so easy why didn't AMD engineers do it before?

 

[frozentundra123456]

Piledriver's IPC falls somewhere between a CD2 and Nehalem, right? I bet jumping 40% performance per clock isn't as difficult as it might sound, they're starting with the bar set fairly low. If they can get mid - upper 3GHz clocks they can match current core / module performance at what would likely be much lower power use.

AMD has some good IP to share and build from. They should market to casual gamers, let them know that AMD powers the graphics in all three consoles as well as the CPU's in the two big ones. I think they're in a good position to gain a lot of ground as long as Zen is good an the R9 390X (presumed name?) is the part many of us hope it will be.

Actually, they are phasing out the cpus that power the two top end consoles, and I am not sure what having GCN in consoles has to do with Zen cpu performance.

If you are talking about dgpus, I think they have a good chance to gain back a fair amount of market share. It just depends on whether they introduce a new lineup of more efficient cards. If they only introduce one or to top end monsters and the rest of the lineup is rebrands, they are screwed, no matter how much the AMD fans on these forums will love a top end powerhouse and will tout price/performance for the rebrands. They need a whole new lineup with more efficient cards top to bottom, no matter how much some people like to deride efficiency.

As far a Zen, even if it is a technically successful product, at least until 2017 or later, they are betting the farm on servers, because by not including an igpu, they are ceding basically the entire consumer and enterprise market to intel except for high end desktop users.

 

[ShintaiDK]

If (when?) AMD liquidates, I believe that anti-trust authorities will force Intel to license x86 on FRAND terms. Then you will see meaningful competition in the x86 space, with big boys like Samsung putting their resources into the game. Until then, enjoy never ending hail-Mary passes from a cripple and fantasizing about competition in the x86 CPU space.

Plenty of x86 license holders that are not willing to make any CPUs. Its an illusion to think companies stand in line to throw billions down an empty hole.

Remove the foundry R&D from Samsung and you sit with an R&D budget around AMD. Not sure why people think Samsung is anything outside the foundry/memory business.

 

[TurtleCrusher]

Not sure why people think Samsung is anything outside the foundry/memory business.

Do you live near civilization? Samsung is in everything. Washers, dryers, phones, tablets, computers, lighting, printers, stoves/ranges, small appliances, TVs/monitors, high end audio, cameras and more. That's a heck of a lot more market penetration and product diversification than Intel, AMD, nVidia, and IBM.

 

[ShintaiDK]

Do you live near civilization? Samsung is in everything. Washers, dryers, phones, tablets, computers, lighting, printers, stoves/ranges, small appliances, TVs/monitors, high end audio, cameras and more. That's a heck of a lot more market penetration and product diversification than Intel, AMD, nVidia, and IBM.

Check the IDC chart again and then reconsider your reply.