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

Page 89 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
I'd love to share your optimism, jdwii, but AMD is not going to "flat out" beat Intel in IPC with Zen2. There's too much optimization done towards AVX that AMD just doesn't have (yet) and needs to play catch up on. I would relegate CB11 and 15 scores to "legacy" as more and more software is heavily using AVX. I believe CB20 is now using AVX. So, what I've seen so far is great, but Intel will still have the top spots. I said a looooong time ago that reaching parity is still good and I still stand by that. AMD needs to stand on the same level and they're pretty much doing that now, so they'll stay afloat and give Intel much needed pressure to deliver.

I also like what AMD has done in terms of the platform. Even though it's not fully backwards compatible, you can get away with using most of Ry3K in 3xx and 4xx gen chipsets.

EDIT: I just noted the CB scores were indeed using CB20. That's neat and more likely than not, paint a reaaaaally good picture for AMD.

Cheers!
 
Last edited:
https://www.techspot.com/article/1616-4ghz-ryzen-2nd-gen-vs-core-8th-gen/

2600X IPC vs 8700K

For the 5 applications Zen+ is 4.8% weaker on average vs coffee-lake

On the 6 games being shown Zen+ was 10.3% weaker in IPC compared to the coffee-lake on average.

So I do indeed expect to see Amd on average to win in terms of IPC by around 5-10%. Average doesn't mean everything Intel will still win some but they will lose some too at least until ice lake comes out and i'm purely talking about IPC not frequency where i expect Intel to hold their dominance for probably the entire Zen family.

This is if Amd isn't lying which with Zen's IPC improvement and Zen+ IPC improvement Amd i feel was truthful with their statements.

In terms of frequency 9900K's can pull off 5Ghz to even 5.3Ghz overclocks on 360mm AIO's and high end heatsinks like the noctua nh-d15 pretty easily.

I expect Zen 2 to pull off around 4.4-4.5Ghz as with the 1800X and the 2700X max overclocks where typically 100-200mhz lower then XFR boosts as Amd pushes their flagship chips to their limit out of the box. That means at 5.3 vs 4.5 Intel's frequency will be up to 18% higher for users who OC so for enthusiasts who absolutely demand 144+ fps Intel will still be the better option despite Amd winning in IPC for the short time.

Also I have to admit i'm kind of excited to see Ice lake bring 18% IPC improvements compared to coffee-lake! So happy to see single threaded performance continue to be improved as it makes everything faster and more responsive not just applications that can use more cores. Adding more cores is cool and all but its like adding memory unless the application or game you use uses it you won't see the benefit from it.
 
jdwii said:
Intel will still win some but they will lose some too at least until ice lake comes out and i'm purely talking about IPC not frequency where i expect Intel to hold their dominance for probably the entire Zen family.
Click to expand...
At the moment, Icelake is only a 4C8T mobile/embedded CPU and the top-end SKU clocks 400MHz LOWER than its 14nm predecessor. Doesn't bode too well for any hypothetical desktop variant of which there are no signs of on leaked roadmaps.
 
  • Like
Reactions: digitalgriffin and jdwii
InvalidError said:
At the moment, Icelake is only a 4C8T mobile/embedded CPU and the top-end SKU clocks 400MHz LOWER than its 14nm predecessor. Doesn't bode too well for any hypothetical desktop variant of which there are no signs of on leaked roadmaps.
Click to expand...

You are not wrong that does indeed have me worried about overclocks as i said previously the continuation of shrinking die size will make higher frequency's harder to get do we all remember ivy bridge vs sandy in terms of overclocks?

Not a fanboy of Intel (or Amd despite me being so excited for Zen 2) but something tells me 5ghz on icelake will not be impossible to do but perhaps harder then coffee-lake.

Overall i'm a fanboy of performance per watt and performance per dollar as well as low noise components(so yeah i'm a major fanboy of Noctua). Also that is why i'm on Nvidia as they offer higher performance per watt then Amd so far anyways plus i own a G-sync monitor so i'm kind of locked into Nvidia's ecosystem.
 
Yuka said:
Some context of the speeds and gaps between the i7 8700K (I still consider this a great CPU, BTW) and the engy sample:

https://cpu.userbenchmark.com/Compa...2-3734-N-vs-Intel-Core-i7-8700K/m697865vs3937

So it's clocked at ~3.4Ghz and the 8700K is 28% ahead in single core speed at ~4.7Ghz. I'd say that is pretty damn impressive, as the page says it didn't go beyond 3.4Ghz.

EDIT: This is what I'm talking about (napkin math):
  • 137 / 107 - 1 = 0.280 -> 28.0%
  • 4.7 / 3.4 - 1 = 0.382 -> 38.2%
The 10 core part is going to boost up to ~5Ghz on its own, so it's going to be above the i7 8700K in single core speed. Well, for whatever they test in that place, that is.

MORE EDIT: The big fat assumption I'm surprised no one has pointed out is about how linear (or not) the test score in that site is. And that score looks oddly similar to CB scores. I'll find more information against other engy samples from the Ryzen 2K gen era and extrapolate a bit more information.

2700X @3.5Ghz: https://cpu.userbenchmark.com/UserRun/16847690

  • 107 / 101 - 1 = 0.059 = 5.9% (or 1 - 101 / 107 = 0.056 = 5.6%)
  • 3.5 / 3.4 - 1 = 0.029 = 2.9%
Interesting numbers and looks like there is a slight IPC increase (~7%). The ~30% jump in performance looks more real now with these tidbits of information if they'll have > 4.7Ghz turbo speeds. Now, the 2700X was using slower RAM, so that could also play against it in the benchmark.

But hey! I still like what I'm seeing.

Cheers!
Click to expand...
I'll quote myself, because I believe this calculation holds water with what AMD presented.

I'm saying it'll be a ~7% IPC increase in mixed workloads over the 2700X and not 15% as Miss Su stated, but that's to be expected as IPC is dependent on code-path used (shocker! lol).

So, I said ~30% jump in single threaded performance if AMD came up with 4.7Ghz speeds and up, but they'll actually be lower, so my modified statement will be to "up to" 30% increase of ST performance. MT could be higher, as doubling cache will help tremendously there.

Cheers!
 
InvalidError said:
If your thread gets bumped between cores, then it got evicted from that core by the OS scheduler so the OS could run something else on that core, its state got moved from core to RAM, then back from RAM to whatever other core it got scheduled to next. Two full context swaps like this cost a couple of microseconds each, which is ~100X worse than the worst ringbus latency. Ringbus is the least of your worries if the OS bumping your threads between cores is causing performance issues.

The OS shouldn't be swapping threads around while there are enough under-used cores available to keep everything running. If you don't want the OS constantly playing musical chair with your threads, use a few less threads than there are cores available. Alternatively, you can use thread processor affinity to force the OS to associate threads with specific cores and button things up yourself, handy if you want to share data through the nearest L2/L3/L4 or RAM in NUMA environments.
Click to expand...

Tying threads to specific cores causes all sorts of issues the minute you run into another program that does the same exact thing. It's very bad practice to do so on a preemptive OS, though this is generally the preferred way to schedule threads on an embedded system.

The problem (focusing on Windows here) is at some point, your thread is getting bumped for a thread that has higher priority. And when it goes rescheduled, it is the lowest priority thread at that instant that gets bumped; it is not guaranteed (and statistically unlikely) that the bumped thread ends up back on the same core it started on.

Linux handles this better by using thread pools, but runs into the same problems if overall system workload changes over a short period as thread allocation would have to be reshuffled.
 
gamerk316 said:
Tying threads to specific cores causes all sorts of issues the minute you run into another program that does the same exact thing. It's very bad practice to do so on a preemptive OS, though this is generally the preferred way to schedule threads on an embedded system.

The problem (focusing on Windows here) is at some point, your thread is getting bumped for a thread that has higher priority. And when it goes rescheduled, it is the lowest priority thread at that instant that gets bumped; it is not guaranteed (and statistically unlikely) that the bumped thread ends up back on the same core it started on.

Linux handles this better by using thread pools, but runs into the same problems if overall system workload changes over a short period as thread allocation would have to be reshuffled.
Click to expand...
Don't forget Windows doesn't use I/O Scheduling, but time slices for all things I/O. That also impacts threading a lot.

Cheers!
 
jdwii said:
https://www.techspot.com/article/1616-4ghz-ryzen-2nd-gen-vs-core-8th-gen/

2600X IPC vs 8700K

For the 5 applications Zen+ is 4.8% weaker on average vs coffee-lake

On the 6 games being shown Zen+ was 10.3% weaker in IPC compared to the coffee-lake on average.

So I do indeed expect to see Amd on average to win in terms of IPC by around 5-10%. Average doesn't mean everything Intel will still win some but they will lose some too at least until ice lake comes out and i'm purely talking about IPC not frequency where i expect Intel to hold their dominance for probably the entire Zen family.

This is if Amd isn't lying which with Zen's IPC improvement and Zen+ IPC improvement Amd i feel was truthful with their statements.

In terms of frequency 9900K's can pull off 5Ghz to even 5.3Ghz overclocks on 360mm AIO's and high end heatsinks like the noctua nh-d15 pretty easily.

I expect Zen 2 to pull off around 4.4-4.5Ghz as with the 1800X and the 2700X max overclocks where typically 100-200mhz lower then XFR boosts as Amd pushes their flagship chips to their limit out of the box. That means at 5.3 vs 4.5 Intel's frequency will be up to 18% higher for users who OC so for enthusiasts who absolutely demand 144+ fps Intel will still be the better option despite Amd winning in IPC for the short time.

Also I have to admit i'm kind of excited to see Ice lake bring 18% IPC improvements compared to coffee-lake! So happy to see single threaded performance continue to be improved as it makes everything faster and more responsive not just applications that can use more cores. Adding more cores is cool and all but its like adding memory unless the application or game you use uses it you won't see the benefit from it.
Click to expand...

You are grossly overstating the overclocking potential of the 9900k vs the caliber of cooling required. 9900k's that even hit a stable 5.2-5.3 at "safe" voltage are rare as hens teeth to begin with. Even then keeping a 9900k stable and cooled at 1.4 (+) v @ 5.2 (+) ghz with a 360 AIO or a D-15 is not realistic. Trust me I know. I've tried with a binned chip. 4.8 - 5.0 ghz all core is about as good as you can hope for using non custom cooling with your average 9900k. Even then heavy loads are probably degrading the chip in many cases because It's almost certainly running above 85c when loaded hard.

We can't really assume that the new ryzen CPU's will behave similarly in regards to overclocks as the previous gen CPU's did. While not a completely unreasonable assumption it's still an assumption. They've made some pretty drastic changes in these new CPU's. The chiplet design is going to vastly streamline the binning process and the more expensive chips will likely be capable of better overclocks this time around. I'm holding out hope that AMD is confident they can match Intel in gaming at the stock clocks and left OC headroom. This way OC enthusiasts can take them to the next level while throwing power efficiency and reasonable cooling requirements out the window. This might be wishful thinking but It does make some logical sense that this could be a possibility.

I realize its fun to speculate, but we are really left with as many new questions as we are with answers after the Computex announcements. Do they match Intel at stock settings in games? What is the deal with XFR and PBO this time around? Do we have overclocking headroom? How much? 5 ghz? Are the 8 cores using single chiplets or dual chiplets? Is this SKU by SKU? Is the 16 core coming? When? Is there going to be a non x 8/12/16 core? Are there going to be 7nm quad cores?

We might get some answers at E3 in 2 weeks but i suspect there is a whole lot we won't know until shortly before they hit shelves in July. I personally can't wait to see the independent reviews and benchmarks.
 
Last edited:
  • Like
Reactions: jdwii
I still hold my statements in regards to Amd getting 15% improvements in IPC as i personally feel they have been honest when regards to Zen I also don't think we will see 5ghz overclocks on Ryzen i truly don't see anything past 4.4-4.6 all core being possible and 4.6 will probably be rare when i say overclocks i mean 100% stable not a single run test.

I would be happy to be wrong and i would love to see 5ghz overclocks on ryzen i'm personally buying the 3800X myself i wish it was cheaper haha.

I do expect memory frequency to be improved probably 3600mhz or so will be the new easy target with low timings. I managed 3400mhz with 14 timings with my 2700x using Samsung b-die memory.

Also 5Ghz in the overclock community is easy to pull off on a 9900K heck Intel's new chip stays at 5ghz all core at stock though i admit it probably runs hot lol.

8 core models will be a single chiplet design by the way but i do want to see more of the actual design sadly we have to wait. I miss paper launches but i will link the site below where the author confirmed the 8 cores are a single chiplet design which is fantastic news.

https://www.anandtech.com/show/1440...-cores-for-499-up-to-46-ghz-pcie-40-coming-77

I also enjoyed this interview with Robert Hallock

View: https://www.youtube.com/watch?v=ml7rHtCHoFE&feature=youtu.be
 
Last edited:
  • Like
Reactions: rigg42
Yuka said:
Additionally, they mentioned the 15% jump is in comparison to the Ry1K gen and not 2K, so my 7% makes total sense now after the napkin math.
Click to expand...
If it really is 15% over first-gen, then we'd be down to Intel-style single-digit improvements for two Zen generations in a row. Between the cache doubling and other structure buffs, that would be a rather horrible silicon area increase to performance gain ratio.

We'll see in six weeks.
 
  • Like
Reactions: NightHawkRMX
No its mistaken out of context its 15% over Zen+ if you would watch his interview you would hear him say that.

2min in the interview
"So our first generations zen was 52% higher and then the middle one was 3 to 5 and then this one is another 15% on top of that" -Robert Hallock

Also i could be wrong like i hope i was lol seems overclocks might be possible at higher frequency's.

https://wccftech.com/amd-ryzen-3000-zen-2-cpu-5-ghz-overclock-4-5-ghz-all-core-boost/
 
Yuka said:
The plot thickens!

https://twitter.com/x/status/1134130341122232320
View: https://twitter.com/IanCutress/status/1134130341122232320


That 15% IPC increase is based in SPECint testing, so it doesn't necessarily mean it really is across the board. In fact, it's bound to be lower than that. Additionally, they mentioned the 15% jump is in comparison to the Ry1K gen and not 2K, so my 7% makes total sense now after the napkin math.

Cheers!
Click to expand...

I remember when the Amd community was claiming that this benchmark was bias towards Intel.
 
I'd love to be wrong, but they have a bit of an argument going on at the moment over the 15% figure. And SPECint is an enterprise benchmark, so what any "community" thinks about it is crap, garbage and biased thinking, as AMD needs to measure itself against it if they want to have a shot at the enterprise market (and all other "enterprise" class benchies, obviously).

I have to say I've always understood the 15% figure was on top of Zen+, but now I'm doubting.

Cheers!
 
jdwii said:
In terms of frequency 9900K's can pull off 5Ghz to even 5.3Ghz overclocks on 360mm AIO's and high end heatsinks like the noctua nh-d15 pretty easily.
Click to expand...

I'm almost 100% sure that most 9900K's can't pull 5.3Ghz. There is silicon lottery binning data has only around top 8% of 9900k hitting above 5.1 Ghz. Getting even higher clocks like 5.2 and 5.3 Ghz are even less likely than that, probably significantly less so for 5.3 Ghz. I'm guessing around less than 3% of 9900k hit 5.3 Ghz. This is just taking a gander at silicon lottery i9-9900k listings and some past historical binning stats so take it with a grain of salt.
 
  • Like
Reactions: rigg42
Seriously. At least you have hyperthreading.
Although I have a newer, more efficient Zen CPU, my 4c4t ryzen 3 1200 @3.7 is even slower than your 2600k both in per-thread and multithreaded performance and yet I can still game fine on it for most titles.
 
boju said:
Man, id still be happy if i had a gimped 9900k @ 4.5GHz seeing the value im still getting out of my old 2600k. What it takes to please people these days.
Click to expand...
If you are still happy with the value you are getting from your i7-2600k, then it means that what is on the market today still isn't good enough for you to bother with an upgrade,. That's the essence of what I find disappointing about the 3k series. Yes, it is better, but still not so much so for the price that I can be bothered to upgrade my i5-3470 yet.

Until about ten years ago, keeping the same PC for 5+ years was nearly unthinkable for most people including myself with most people upgrading every 2-3 years, today it is a matter of course for a growing number of people as good 5+ years old PCs (something like i5-2400 and up) are still adequate for the vast majority of everyday tasks and most casual to somewhat competitive gaming.
 
I kept upgrading for I/O. Sata II to Sata III, Sata III to NVMe. Not much point going beyond that though. PCIe 4.0 and ludicrous speed drives won't do much for most people. Not until everyone is swapping around 8K video files or some nonsense.

They kept adjusting Moore's law so they would have a new expectation to hit. It really helps with everyone's planning to have it in place. Should really be renamed though, law makes it sound absolute. But being able to predict the average computing power a few years ahead of time lets programmers target their user base without having said future hardware in hand.
 
remixislandmusic said:
Im not sure how much of this can be blamed on the end of moores law
Click to expand...
At the moment, software developers deserves the bulk of the blame IMO: if more software was heavily threaded, CPUs could be designed to favor total throughput over single-threaded performance and this would make it possible to scale performance using much simpler, smaller and more power-efficient cores. Since 3D rendering is embarrassingly parallel, this is exactly what GPUs do: hundreds of simpler extremely power-efficient cores running thousands of threads.

We'll probably see some sort of big.LITTLE arrangement at some point in the future where we'll have a couple of high-cost cores to run heavily sequential blobs like threads handling user interactions and a bunch of heavily threaded cores to handle bulk processing. AMD wanted to do something sort of like that when it hyped up heterogeneous compute a couple of years ago, likely only a matter of time before the idea resurfaces. There may already be a hint of that in rumors of quad-threaded Zen 3.
 
InvalidError said:
At the moment, software developers deserves the bulk of the blame IMO: if more software was heavily threaded, CPUs could be designed to favor total throughput over single-threaded performance and this would make it possible to scale performance using much simpler, smaller and more power-efficient cores. Since 3D rendering is embarrassingly parallel, this is exactly what GPUs do: hundreds of simpler extremely power-efficient cores running thousands of threads.

We'll probably see some sort of big.LITTLE arrangement at some point in the future where we'll have a couple of high-cost cores to run heavily sequential blobs like threads handling user interactions and a bunch of heavily threaded cores to handle bulk processing. AMD wanted to do something sort of like that when it hyped up heterogeneous compute a couple of years ago, likely only a matter of time before the idea resurfaces. There may already be a hint of that in rumors of quad-threaded Zen 3.
Click to expand...
Not developers per se, but the big companies pushing entire ecosystems that were hard to upgrade and expensive (coughMicrosoftcough). If you talk about individual Indies, they've made wonders with Unreal back in the day. Hell, even ID Tech has been superb in terms of multi-threaded support for DECADES. Even Java has had proper MT support since 1.0 and that is JAVA.

The big corporation building gaming engines and other complex frameworks have all the blame as much as the companies using them not wanting to move with the times as quickly as we'd want them to.

Cheers!
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom