AMD's Future Chips & SoC's: News, Info & Rumours.

Page 94 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.

rigg42

Prominent
Oct 17, 2018
555
165
640
14
Intel beating 12nm AMD CPUs isn't exactly news. I fully expect to see the 3200g/3400g to get slayed by 14nm Intel quad cores on Desktop as well.
 
I wonder which 12nm mode AMD is using for the APUs... High Performance? Intel I guess is stuck to a single 10nm node (low power?), so the disparity does not surprise me at all, but the key would be to test on equal clocks and performance without looking at power just yet. That would compare Ice Lake to Zen+ (aka Zen 1.5)

Cheers!
 

jdwii

Splendid
Last Amd laptop i owned was a Llano(A8-3500m with 8GB of ram) model still runs 10 perfectly today with a SSD and some TLC. Don't use it much as i'm mainly on my main rig.

So crazy Amd changed so much in the past 10 years, we all remember them being seriously obsessed with APU's. Now they want to avoid the term 10 years later. FPU performance they were downplaying now they are doubling the compute power.

Very hard to recommend a laptop without knowing the usage or budget.

10 years ago we had the I7 965 and 5870 ATI been thinking about what we might have in 2029 something tells me the upgrades will be less then it was in 1999 to 2009.
 
Last edited:

InvalidError

Titan
Moderator
Makes me wonder if 10 years later at 1080P we are far off from 10 years ago then think about 2009 vs 1999
Progress has slowed down considerably since ~2010. CPU-wise, 1999 to 2009 is a 20-25X increase in processing power: 3-3.5X in clocks, 4X in core count, 1.6-2X in IPC. From 2009 to 2019, it is about 5X: 1.3X for clocks, 1.5X for IPC, 2X in core count, 1.3X for SMT2.

Over the next 10 years, I'm not expecting much over a 3X increase in mainstream ($150-300) CPU-power: 2X on cores, 1.1X from SMT4, 1.2X on clocks and an optimistic 1.2X on IPC.
 
Progress has slowed down considerably since ~2010. CPU-wise, 1999 to 2009 is a 20-25X increase in processing power: 3-3.5X in clocks, 4X in core count, 1.6-2X in IPC. From 2009 to 2019, it is about 5X: 1.3X for clocks, 1.5X for IPC, 2X in core count, 1.3X for SMT2.

Over the next 10 years, I'm not expecting much over a 3X increase in mainstream ($150-300) CPU-power: 2X on cores, 1.1X from SMT4, 1.2X on clocks and an optimistic 1.2X on IPC.
I think your overestimating clocks but underestimating IPC; I wouldn't be shocked if IPC goes up by virtue of Intel putting out a new core design (if not a new architecture). Clocks are simply limited by thermals, and you aren't going to see the 5Ghz barrier breached for that reason alone.

But on the whole, I wouldn't expect much farther performance gains. Hell, I wouldn't be shocked if CPUs today are still considered viable a decade from now; I really think we're heading for a performance cliff.
 
So crazy Amd changed so much in the past 10 years, we all remember them being seriously obsessed with APU's. Now they want to avoid the term 10 years later. FPU performance they were downplaying now they are doubling the compute power.
hmm...if only someone on these forums argued that FP performance was the wrong thing for AMD to focus on. If only such a person existed!

(/s, if you couldn't tell)

But seriously, APUs as a whole only made sense in very power limited form factors due to having a very weak CPU core due to power-draw requirements. As a result, APUs now are synonymous with "weak CPUs", which is why AMD would want to avoid the term.
 

InvalidError

Titan
Moderator
I think your overestimating clocks but underestimating IPC; I wouldn't be shocked if IPC goes up by virtue of Intel putting out a new core design (if not a new architecture).
IPC is limited by how many opportunities the CPU has to parallelize instruction from a thread's execution stream. How much IPC is possible even on a theoretical basis with infinite execution resources is still limited by data dependencies and missed conditional branches. At some point, IPC will hit a brick wall and we're very close to that, just look at how AMD had to double L2 and L3 sizes, improve branch prediction, widen the architecture by 33%, all to achieve only a 10-12% IPC increase. Doubling transistor count mainly from doubling cache sizes to get only 10% more single-threaded performance out of Zen 2 isn't sustainable and being that much closer to the single-threaded IPC brick wall means even more effort will be necessary for every additional IPC inch. (Yes, I know AMD says 15%, but those are multi-threaded benchmarks, meaning that part of the IPC gains come from SMT having a wider back-end to fill. The single-threaded IPC that most games depend heavily on won't benefit from this anywhere near as much.)

Now, Zen 3 is expected to increase SMT from two to four threads. With more threads per core, chip architects can focus on aggregate per-core throughput instead of single-thread IPC, which opens up the possibility of actually seeing a single-threaded IPC regression in favor of larger SMT gains and we may end up with 30-40% more performance out of SMT at the expense of 10% single-thread IPC. (Still comes out to about the same ~30% more performance per clock per core as my previous +20% SMT +10% SMT.)
 

TechyInAZ

Titan
Moderator
Wait Zen 3 is adopting Intel Itainum quad thread hyperthreading?? That's cool!!!

Plus, in the future, as companies keep adding more and more cores to their architectures, memory is going to play an even bigger role in performance. Even now, memory can be a big bottleneck for high core count CPUs in terms of raw speed and latency.
 

InvalidError

Titan
Moderator
Wow that's cool! Are there ARM chips that run SMT?
Not in the mainstream. At the moment, the Cortex A65E (automotive) is the first and only SMT ARM core design. Give ARM a few more years to run into all the same IPC brick walls as everyone else, then they'll go SMT across most of the product stack as a more power-efficient way of getting more work done per clock per core.
 

jdwii

Splendid
I think your overestimating clocks but underestimating IPC; I wouldn't be shocked if IPC goes up by virtue of Intel putting out a new core design (if not a new architecture). Clocks are simply limited by thermals, and you aren't going to see the 5Ghz barrier breached for that reason alone.

But on the whole, I wouldn't expect much farther performance gains. Hell, I wouldn't be shocked if CPUs today are still considered viable a decade from now; I really think we're heading for a performance cliff.
That's where i was getting at back in 2009 we had the I7 965 and a Radeon 5870 as the top products and today which aged better?

Then take a look at 1999 the Athlon 1Ghz CPU wasn't even released until 2000 1H and the best GPU was Nvidia Geforce 256 in 99.


Point being a 10 year old PC if one bought top of the line can still function perfectly and do everything for most people and with the 965 upgrading to a RX580 or RX 570 and of course a SSD and you will probably be ok for the 2-3 years trying saying the same thing in 2009 with the Athlon 1ghz CPU lol.
 

goldstone77

Distinguished
Aug 22, 2012
2,250
13
19,965
85
Intel Ice Lake performance has appeared in a leaked PassMark result, showing faster performance than the AMD Ryzen 5 3500U, despite Team Red's chip having a higher base clock speed.
A few things about the test...
https://pbs.twimg.com/media/D9bJldjXkAE-Nqw.png
  1. PassMark takes about ~1 min. 30 seconds to complete give or take, so it's possible the i7 could boost at 3.9GHz for the entire test, 200MHz higher than the boost of the 3500U.
  2. Why compare an i7 to the R5 3500U instead of the R7 3700U?
  3. A comment was made on twitter that the source of the test was fake, I'm checking on that now.
Edit: 3. "Chiphell user posted it then it was deleted eventually" So, likely a fake.
 
Last edited:

jdwii

Splendid
Ice lake what is that 6+ months away before desktop parts are out or longer? 7-7-19 launch for Ryzen 3000 7nm series.

When is Zen2+ or Zen 3 coming again? 2020? So when Ice Lake comes out i guess we can all say lets wait 6+ months for Zen 3 lol.

If we get IPC wars again i will be so happy life nostalgia coming back.
 
Reactions: jaymc

InvalidError

Titan
Moderator
Ice lake what is that 6+ months away before desktop parts are out or longer?
There are no plans for desktop Icelake on Intel's leaked roadmaps. In all likelihood, if there ever are desktop Icelake parts, they will be just like Broadwell-DT parts: never planned to reach the desktop and nearly impossible to get before the next-gen after it launches once Intel did decide to announce the desktop variants, making it almost completely irrelevant for desktop.
 
Reactions: remixislandmusic

goldstone77

Distinguished
Aug 22, 2012
2,250
13
19,965
85
Intel doesn't like to cut price on existing chips, they like to offer slight promotional discounts ~$15 and "free" octane. Here in the recent history they did off cheaper prices on newly announced products. Intel isn't stupid, if sales drop by significant numbers, and they can't figure out how to limit AMD's ability to sell their processors then they will be forced to drop their prices as supply and demand dictate. They have share holders!
 

InvalidError

Titan
Moderator
Intel doesn't like to cut price on existing chips
Until around Core2's launch where AMD and Intel were in periodic price wars, 30+% price cuts happened nearly every year if not twice a year in some cases.

The biggest difference between then and now is LEAN/JiT manufacturing: Intel has large enough manufacturing of other parts and large enough provisioning contracts for CPUs that it can schedule end of manufacturing to match whatever outstanding orders it has left when it decides to discontinue products and never have to worry about a meaningful amount of surplus inventory, hence hardly ever any need to lower MSRPs on existing parts. AMD does not have this much supply chain flexibility, still has to pay for all of its minimum wafer volume commitments regardless of sales, so it has to make sure those sales keep happening and also lacks the large long-term OEM contracts to smooth out that demand.
 

jdwii

Splendid
Intel dropping prices right after "maybe" fixing their shortage issue doesn't make a lot of sense to me, and all their CPU's will sell to oem's and us a smaller market.

Something tells me Intel will just release their 5ghz all core 8 core CPU and keep marketing that all cores are not the same and maybe hold off until they release their newer architecture. At 5Ghz Intel will beat Amd probably in games and other lightly threaded apps. Intel will simply use that as a statement that they believe their CPU's are worth their price.

Personally if i was Intel I would do that above while getting a new CPU architecture(3-5 years) built.
 
Reactions: remixislandmusic

InvalidError

Titan
Moderator
Intel dropping prices right after "maybe" fixing their shortage issue doesn't make a lot of sense to me, and all their CPU's will sell to oem's and us a smaller market.
If AMD is starting to eat into OEM orders and Intel has sorted out its shortage issues, then Intel may be seeing some surplus inventory. There are no miracles to mitigating surplus stock, you either reduce production or lower prices to stimulate demand and make the surplus go away. Extra manufacturing capacity does you no good if the extra chips only end up hogging warehouse space.
 

ASK THE COMMUNITY

TRENDING THREADS