Ryzen 4000 IPC boost could be around 20%

[Loote]

I think 10900K will still lead in some games and specific usage cases, but I wonder, will AMD nerf their lower tiers? Right now 3600X has more or less the same (gaming/single core) performance as higher tier chips, heck, imagine the 4300X.

I wonder how long will they be able to continue with 10+% IPC jumps, assuming this is true of course.

[mr moose]

I jist feel sorry for AMD sometimes, I mean we all know that kid, the one who always sits on the chair with the broken leg, wins a trip to the USA only to get sick an unable to fly, finally develops a decent CPU that is a hands down winner and covid comes along and screws up potential sales. 

 

Seriously this is awesome news even if it's only half true, because it feels good to actually have a viable alternative when upgrading.

 

 

Also, Intel are not screwed, rekt, roasted, etc, they'll be just fine.

 

.

[mr moose]

14 minutes ago, Loote said:

I wonder how long will they be able to continue with 10+% IPC jumps, assuming this is true of course.

 

According to jim keller,  moores law is not dead in it's essence and its sense of what it was also an observation of,   I would guess given that we are likely a fair way from IPC jumps being limited by either node advances or architecture advances.

 

Great video that is worth the 26 minutes if your into CPU design etc.

 

https://www.youtube.com/watch?v=c01BlUDIlK4

 

[Loote]

1 minute ago, mr moose said:

Great video that is worth the 26 minutes if your into CPU design etc.

I'm not but I'll give it a try anyway. Good to hear we're not likely to slow down soon.

[Guest]

If only Intel would have some sort of other revenue stream or more products / contracts to sell.

[Arika]

i'm secretly hoping AMD takes top spot in everything just to see all the surprised Pikachu faces when AMD becomes just like Intel was before the launch of Ryzen.

 

people shouldn't want intel to be "doomed" 

people shouldn't want AMD to be the best at everything

because then there's no competition. 

 

Without competition the consumer loses. and if you think that AMD wouldn't be like Intel, you're an insane person, there has never been a company in the top position in any industry that hasn't abused their power.

[Loote]

I think a year or two of AMD's dominance could allow them to bring the percentages to a much closer state, you know, so that their financial results aren't less than 10% of Intel's. Maybe I'm mistaken, but I feel having a more even split would make it easier for AMD to compete in the future.

[mr moose]

11 minutes ago, RorzNZ said:

If only Intel would have some sort of other revenue stream or more products / contracts to sell.

Without the gamer market and some of the content creator workstation market Intel have only got their NUC/thinclient, server/data, lidar, AI, GPGPU, Memory and SSD markets left to draw revenue from.   I guess they have to consider filing for bankruptcy.

 

 

[5x5]

49 minutes ago, comander said:

I could be mistaken as I'm going off of "IPC gain" figures put out by the companies... Zen1 was around 7% behind SKL/CFL and Zen2 was around 7% ahead of them overall... Intel says ICL is 18% ahead of CFL in terms of IPC (but its clock speed is around 20% lower so... same-ish performance)

 

Could you link to your source? I checked a handful of reviews and saw nothing showing Renoir ahead of ICL in terms of IPC (but tons showing that Renoir performs awesome in general). 

Are you sure you're not referring to the CFL based laptop parts? Those are around 7% behind on IPC and are overall the fastest parts intel has by a hair. 

Perfectly sure - Ice Lake's IPC gains do exist but they only narrowed the gap between Zen 2. Keep in mind, Renoir does appear to post more impressive results than base Zen 2. I'm not sure why that is but it's definitely somewhat better overall and I fully expect the upcoming 4700G will outperform the 3700X/3800X(T).  The exact percentages are in an older NBC article but I don't have the time to dig it up.

Though you can see here - https://www.notebookcheck.net/The-AMD-Ryzen-7-4700G-holds-its-own-against-the-Intel-Core-i9-10900K-in-initial-benchmarks.481216.0.html

The initial 4700G results are on par with a 9900K/10700K despite the large clockspeed gap and that's no small feat for Zen 2. Zen 3, well, that'll be a good 10% ahead of Intel in virtually every task from the looks of things. The only exception being AVX-512 I imagine.

[Trik'Stari]

I don't understand how AMD has gotten so far ahead of Intel in the CPU market, but still lags painfully far behind Nvidia in the GPU market.

[GOTSpectrum]

I think sometimes we on this forum can lose sight of what is really important for AMD. Yes gaming performance will improve(If this is functional IPC gain) But even if it is only FP, AMD will become THE choice for big data and HPC.

 

More cores, more IPC, less power, more IO. 

 

Yes it's fun to talk about gaming and desktop performance but the money is with the titans, Google, AWS, Microsoft, Cray, IBM, Alibaba, et al.

 

IF this is true and the power envelope remains mostly the same, the Performance Per Watt of next gen EPYC is going to be insane compared to Intels line up. 250W with only 28 cores on the 8380H compared to 64 cores at as low as 200w. 

 

Now yes, I know, clock speed is different and things. But, regardless, getting another 20% on top of EPYC will cement AMDs place in the market. PCIe 4.0 was a good move, reduced IF latency, higher performance. There are improvements we have already seen, and with ZEN being so young there are still many low hanging fruits to be targeted for getting better performance.  It will be interesting to see how it goes in the next few years as the low hanging efficiency/performance/latency gains have dried up. I really hope AMD don't stagnate.

 

Anyway, this is where the big money is at and it is where AMD need to make headway if they plan for this good run to keep going. Consoles will give then a boost of income in the next couple years as people buy the new gen, let's hope AMD builds a war chest, no point better the house by spending everything they make. 

 

With that said, they already have contracts for some serious next gen HPC projects, now they need to capitalise on that and keep the ball rolling in the right direction. 

[GOTSpectrum]

3 minutes ago, Trik'Stari said:

I don't understand how AMD has gotten so far ahead of Intel in the CPU market, but still lags painfully far behind Nvidia in the GPU market.

Patents, dev costs, technical ability.

 

Nvidia have some amazing patents regarding memory compression, this is one of the reasons we have seen AMD cards as memory starved as they have been. Nvidia have many more patents too, plus gameworks which likes to bog down team red cards.  There's also money, Team Green can spend such a great amount for advertising alone, forget about their much higher budget for actual product development. A massive issue for AMD is driver development, AMD don't have enough people with the right technical skills to get it all running smoothly. 

 

And finally, AMD is fighting a war on two sides with the two largest silicon engineering corporations out there, while being tiny compared to either one of them. They have to pick their battles. RTG really only exists for HPC and to ensure they have the best in class semi-custom offerings. 

 

I'd love to see AMD create a semi-custom EPYC, with aspects of a GPU on packaged for lighting fast large scale FP workloads. But I'm sure the smart people over at AMD have done the maths and worked out it isnt worth the cost or hassle. 

[Trik'Stari]

2 minutes ago, GOTSpectrum said:

Patents, dev costs, technical ability.

 

Nvidia have some amazing patents regarding memory compression, this is one of the reasons we have seen AMD cards as memory starved as they have been. Nvidia have many more patents too, plus gameworks which likes to bog down team red cards.  There's also money, Team Green can spend such a great amount for advertising alone, forget about their much higher budget for actual product development. A massive issue for AMD is driver development, AMD don't have enough people with the right technical skills to get it all running smoothly. 

 

And finally, AMD is fighting a war on two sides with the two largest silicon engineering corporations out there, while being tiny compared to either one of them. They have to pick their battles. RTG really only exists for HPC and to ensure they have the best in class semi-custom offerings. 

 

I'd love to see AMD create a semi-custom EPYC, with aspects of a GPU on packaged for lighting fast large scale FP workloads. But I'm sure the smart people over at AMD have done the maths and worked out it isnt worth the cost or hassle. 

You'd think they have more money to play with, considering how hard they've been stomping Intel.

[GOTSpectrum]

5 minutes ago, Trik'Stari said:

You'd think they have more money to play with, considering how hard they've been stomping Intel.

So for reference, this is the income for 2019FY;

 

AMD :- 6.73b USD(up 4% year on year)

 

Nvidia :- 11.72b USD(up 21% year on year)

 

Intel :- 72.0b USD(up 2% year on year)

 

As you can see, simply looking at revenue for the fiscal year of 2019, The raw income for team red is far smaller. They have be super efficient with their cash, and that means making some hard choices like software dev team sizes. Also bare in mind that Nvidia has more than just GPUs, and Intel has so much more than CPUs it isn't even funny. But still, you get the picture. AMD is fighting two behemoths on a comparatively shoestring budget.  It's a real David and Goliath, except there are two Philistines giants in this story.

[cj09beira]

36 minutes ago, Trik'Stari said:

I don't understand how AMD has gotten so far ahead of Intel in the CPU market, but still lags painfully far behind Nvidia in the GPU market.

all it takes is a look at R&D, nvidia is currently at 700M, amd at 400M, intel 13B, so amd is up agaisnt 2 juggernauts, 

and they aren't far behind 

[D13H4RD]

I'll believe it when I see it. Not saying it won't be good, just that I always take any rumor with a grain of salt. 

 

Also, if all this pans out to be true, Intel still won't be crushed. No one should wish for Intel to be defeated to the point where they can't compete.

 

They will be shaken but they will also fight back. We want competition, not an utter massacre. 

[LAwLz]

6 hours ago, zeusthemoose said:

If this is true, Intel is doomed (unless they also have a similar performance boost coming soon)

5 hours ago, Enderman said:

They literally just released their 10000 series CPUs, so obviously not...

 

Intel do have major IPC improvements in the pipeline. It's just that they haven't released it for desktops yet.

Sunny Cove has a 18% higher IPC than Skylake, and around 7% higher IPC than Zen 2.

 

 

  

2 hours ago, Arika S said:

i'm secretly hoping AMD takes top spot in everything just to see all the surprised Pikachu faces when AMD becomes just like Intel was before the launch of Ryzen.

 

people shouldn't want intel to be "doomed" 

people shouldn't want AMD to be the best at everything

because then there's no competition. 

 

Without competition the consumer loses. and if you think that AMD wouldn't be like Intel, you're an insane person, there has never been a company in the top position in any industry that hasn't abused their power.

I think Gamers Nexus said it best in this video:

That type of marketing breeds a lot of immature fanboys that love cheap shots at the competition and that doesn't really put much though into things other than "Durr X bad Y good hurr". 

[Loote]

39 minutes ago, GOTSpectrum said:

Yes it's fun to talk about gaming and desktop performance but the money is with the titans, Google, AWS, Microsoft, Cray, IBM, Alibaba, et al.

 

Everyone always blabs about *the datacenter*, but PC market is huge enough. AMD already leads in many aspects of server CPU needs and more and more decisions are going in their favor, I'd even say the things Intel's still better at won't get solved by IPC increase.
In terms of PCs, in laptops they were still inferior in most use cases and the offers with AMD CPUs were scarce, now that mobile Zen 3's out we can see them making a bigger dent. In case of desktops AMD was a good choice, but definitely not an only choice, your enterprise Dells and HPs could meybe benefit from AMD CPU, but would also work perfectly competently on Intel, whereas in games Intel does still hold the lead.

[GOTSpectrum]

2 minutes ago, Loote said:

-snip-

Although I will agree with you on some points, just a few things.

 

1. Profit margins are greater on HPC products, so although the revenue may not be insignificant, the profit can be many times greater.

 

2. The thing about the HPC world is they hate to change, so getting them onboard with EPYC will help to maintain a more steady high profit income stream.

 

3. Taking recent changes out of the picture for a min due to COVID, the desktop segment is shrinking to some degree or another long term, laptops are a massive market but profit margins are significantly lower than HPC. 

 

4. revenue is great, but profit is what is needed, and at the end of the day if you can make more profit, you can afford to spend more on dev. Meaning better products, shorter dev times and more competition. 

 

I know some say they don't want inel crushed, but for me, in the short term, I do. I want AMD to gain meaningful market share in this high profit segment, and to do that they need to blow intel out of the water. At least for now. I don't want intel to die though, because then we will just end up with team red sand bagging, anfd ain't no one wanting to have that! 

 

TLDR; HPC Profit is king.

[porina]

 

6 hours ago, Fasauceome said:

could have sworn this was an official statement, not a leak

This is the IPC for Zen 3, I also have vague recollection they said something along that line somewhere, even without hard numbers.

 

5 hours ago, VegetableStu said:

man, intel better have CPUs on their own 7nm stuff ready by next year (as far as they've forecasted). they're totally sitting ducks right now

The current best rumour for Rocket Lake is that it will still be 14nm, although it will finally move beyond Skylake architecture. Beyond that, it sounds like Intel might go hybrid 10/14nm depending on the product, at least until 7nm is ready.

 

5 hours ago, Deslumo said:

yeah, there was something someone said about how Intel should accept defeat for a few years and hopefully come back with something competitive later.

I think it was Intel's interim CEO who said that, without trying to dig it up. In essence, they know with their manufacturing broken leg it will take them some time to reach parity, before they can try to charge ahead once again. 

 

5 hours ago, VegetableStu said:

namely them not sinking more R&D costs to 10nm

Wasn't 10nm originally due after Skylake? That failed so we ended up with Kaby Lake as a process refinement on Skylake. They know their problems, and how much value they get from fixing them. Feels like it is a hold out with what they have until 7nm, as fully fixing 10nm isn't going to be economic at this stage.

 

5 hours ago, Dylanc1500 said:

The important question is, which instruction/s?

When the likes of AMD or Intel give IPC indicators, they try to use a wide workload. That's not to say there wont be some bias potential in there, but it wont be focusing on best case niches.

 

4 hours ago, comander said:

This looks to be about on par with Ice Lake in terms of IPC. 

If the Zen 3 indicators are accurate, it should exceed Ice Lake/Sunny Cove. Ice Lake should be about ball park comparable to Zen 2.

 

4 hours ago, comander said:

Tiger Lake/Rocket Lake will be interesting. I'm expecting a bit of a boost over ICL, but I expect TGL/RKL to NOT clock as high as CML. 

The Rocket Lake rumours suggest it will be one of the Cove architectures backported to 14nm. So we have a bigger more complex core on older process. It'll be interesting to see how that goes. Either Cove will at least offer some IPC uplift from Skylake architecture.

 

4 hours ago, comander said:

Also at this point I'm sick of lake codenames. Most of these should be the EXACT SAME NAME. 

The way to cope with these is to understand the difference between product designations from architecture.

 

So on Intel mainstream architecture, we have Skylake (also used in Kaby Lake, Coffee Lake, Comet Lake), Sunny Cove (Ice Lake), and future Willow Cove (?).

 

4 hours ago, 5x5 said:

This is over 15% above Ice Lake. Currently, Renoir mobile leads Ice Lake by 7% as tested by notebookcheck (in terms of IPC).

IPC can be a complicated beast and I like to break it down into peak IPC (what the cores can do if not hindered by other limiting factors) and practical IPC (what you actually get). I'd like to run tests myself to determine peak IPC as I have done with previous Ryzen vs Skylake, but I don't have Ice Lake to compare there, and most info on the net is practical IPC. Especially on mobile platforms it is difficult to make comparisons on peak IPC. With a fair error margin, I do consider Sunny Cove (as used in Ice Lake) to be peak IPC comparable to Zen 2. The Zen 3 rumour levels should clearly exceed this. Who knows if Intel will have Willow Cove ready in time to go against it. Ice Lake mobile will be around a year old towards the end of the year, so that would put its successor potentially going out around then too, even if fighting in different markets.

 

 

Looking back at the OP's source links, even with their age there seems to be one part I hadn't heard before. A possible uplift in FP performance by 50%? That, combined with the expected bigger logical unit (CCD=CCX?) could make them fearsome and be enough for me to stop looking so much at AVX-512, which I'd expect Intel will not offer the good version on consumer when we eventually get it anyway. A 50% uplift implies they add a 3rd pair of mul/add units on top of the existing two. With the new unified cache, that could allow it to be fed adequately and offer some nice performance.

[SPCFHT]

Can't wait to see the new 4000 series ryzen and upgrade my dumb intel cpu bought 5 years ago

[CTR640]

Why is Intel still the best for gaming? If 1440p and higher, you don't really need an Intel cpu, right? Or am I missing something?

What I gathered from the others is Intel is great for 144+ fps and I don't game at 144 and higher.

 

And yeah, it's about time Intel gets their ass kicked. I mean, a fucking yearly 3-5% performance?? I game at 1440p and higher so highly likely AMD for me whenever I upgrade my cpu.

[Loote]

8 minutes ago, GOTSpectrum said:

1. Profit margins are greater on HPC products, so although the revenue may not be insignificant, the profit can be many times greater.

 

I wonder how much extra Client Computing Group gets from things other than PCs. Anyway it will of course look different for AMD, but I doubt the difference would make the proportions super different.

[leadeater]

6 hours ago, Dylanc1500 said:

The important question is, which instruction/s?

IKEA Book Shelves per Cycle. Up to 20% more confusing more quickly.

[Arika]

5 minutes ago, leadeater said:

IKEA Book Shelves per Cycle. Up to 20% more confusing more quickly.