Ryzen 4000 IPC boost could be around 20%

[BabaGanuche]

1 hour ago, Vishera said:

This time AMD will need to implement at least triple channel support into the consumer space.

Why do you say that? DDR5 will have twice the bandwidth of DDR4.

[Letgomyleghoe]

16 hours ago, GDRRiley said:

15-20% with better clocks then 300 chips. dam this sounds like intel about to get roasted hard

gaming advantage goes out the window if thats gonna happen, so there will be quite litteraly no reason to buy intel, unless you already have a z board that is.

[mr moose]

6 hours ago, TheTechWizardThatNeedsHelp said:

Intel does have those markets, but their cpu profit margins are the highest. In SSDs the big company is Samsung. In DRAM it's Samsung. Intel is largest in NUC and GPGPU, but those are much smaller markets than CPU, SSD, DRAM, ECT.  If it becomes fliped, so Intel becomes a CPU manufacturer you only get if you are a huge fanboy, and you don't care about greatest price to performance, that will hurt Intel. A lot. 

The whole post was a sarcastic response to the idea that Intel are going to suffer some how because of this.

 

 

[Helpful Tech Witch]

2 minutes ago, mr moose said:

The whole post was a sarcastic response to the idea that Intel are going to suffer some how because of this.

i know i was just pointing out while you are right, that this is hurting intel. a lot.

[mr moose]

1 minute ago, TheTechWizardThatNeedsHelp said:

i know i was just pointing out while you are right, that this is hurting intel. a lot.

This is barely a tickle for them.   This is good for AMD, but it is far from a dangerous for Intel.  You have to understand that Intel's business is so much bigger than gaming and small workstations.    The vast majority of their CPU's go to data centres and fleet PC's.  They already have the market where the money is. 

 

Intel have already had what amounts to an entire node failure, the rise of ryzen and several issues with security exploits yet they are still ticking over at max capacity and growing revenue.

 

https://www.macrotrends.net/stocks/charts/INTC/intel/revenue

 

 

[cj09beira]

6 hours ago, dfsgsfa said:

very wierd approach from AMD.

 

things we already know , 4000 is last gen supporting am4, ipc over 3000 is 12%- 20% if rumors arr true

 

while new am5 shall at least have 2 gen , new tech introduce?

 

so in amd marketing point of view, how much % ipc should 5000 be better than 4000 if the increment is already as high as 20%, can they really keep the trend?, thats very worrying

well we will probably get some sort of hbm on cpus pretty dam soon, that alone should help quite a bit, ddr5 will help, they should be able to keep this up, then from hbm on package they will pass to cache on top of the cpu with each core having direct access to part of it bypassing the memory controller (working in that case like a L2 cache almost) this should allow for great ipc gains, thermals will be an issue though.

3 hours ago, Vishera said:

This time AMD will need to implement at least triple channel support into the consumer space.

ddr5 will bring enough more bandwidth, and i expect us to see some sort of hbm cache sometime during am5's time frame

[StDragon]

2 hours ago, mr moose said:

This is barely a tickle for them.   This is good for AMD, but it is far from a dangerous for Intel.  You have to understand that Intel's business is so much bigger than gaming and small workstations.    The vast majority of their CPU's go to data centres and fleet PC's.  They already have the market where the money is. 

 

Intel have already had what amounts to an entire node failure, the rise of ryzen and several issues with security exploits yet they are still ticking over at max capacity and growing revenue.

 

https://www.macrotrends.net/stocks/charts/INTC/intel/revenue

 

 

Playing Devil's advocate here: When it comes to the stock market, perception is reality. That is to say, the stock market is outrageously disconnected from reality. So, it's possible that if Intel concedes the desktop CPU market (and they won't), their stock price could drop significantly.

 

That's the danger of a market that's valued on an illusion; you must maintain that illusion. Reality be damned.

[mr moose]

Just now, StDragon said:

Playing Devil's advocate here: When it comes to the stock market, perception is reality. That is to say, the stock market is outrageously disconnected from reality. So, it's possible that if Intel concedes the desktop CPU market (and they won't), their stock price could drop significantly.

 

That'st the danger of a market that's valued on an illusion; you must maintain that illusion. Reality be damned.

Which is true, however their revenue doesn't look to be much effected by AMD's rise over the last 3 years.  while stocks will go up and down over many different conditions (perception being one of them) revenue and profit tells us how well their products have been selling in reality.  

[RejZoR]

6 hours ago, mr moose said:

The whole post was a sarcastic response to the idea that Intel are going to suffer some how because of this.

 

 

They do, but not in scale of going out of business entirely or closing down the CPU division or anything like that. That just won't happen. Just like it didn't for AMD as they survived on GPU division alone during Bulldozer era. It's no different with Intel, they are just much much bigger with much wider portfolio of products that keep them in business. And Intel CPU's, while not being the best thing out there anymore, they are far from bad which wasn't the case with Bulldozer which was severely inferior on many levels so situation isn't quite the same this way either.

[5x5]

8 hours ago, comander said:

The 10900k is comet lake, not ice lake. 

 

https://en.m.wikipedia.org/wiki/Ice_Lake_(microprocessor)

 

Current ICL parts only go to 4 cores, have the best IPC of any x86 part and the worst clock speed characteristics of any current x86 part. 

Ice Lake isn't coming to desktop, thought and on mobile Zen 2 delivers better results than Ice Lake by a VERY wide margin. Even ULV Zen 2 CPUs outperform ice Lake.

[PineyCreek]

Mmm...might build a new rig once the new AMD CPUs and Nvidia GPUs come out...relegate this 1st Ryzen to a network file server.

[Helpful Tech Witch]

Just now, PineyCreek said:

Mmm...might build a new rig once the new AMD CPUs and Nvidia GPUs come out...relegate this 1st Ryzen to a network file server.

That's what I'm doing

[porina]

8 hours ago, cj09beira said:

ddr5 will bring enough more bandwidth, and i expect us to see some sort of hbm cache sometime during am5's time frame

I'm not sure HBM is the right technology to implement a consumer level CPU cache. It is relatively low clocked and makes up for it by having a very wide bus. GPUs are more parallel in nature so can make better use of that width feeding it. CPUs would probably prefer smaller chunks faster.

 

Also the ram bandwidth needs are a complicated thing. If we were to look at it like an Intel CPU, I think we should have gone to quad channel in enthusiast level already. With Zen 2, the generous L3 cache does help mitigate that need to a fair extent. And the rumoured bigger CCX would make the effective cache even more useful in Zen 3.

 

I think DDR5 on introduction would be effectively like a 3-channel DDR4 setup, since we're practically standardised on 3200 as a baseline speed for DDR4 sold today (at least for AMD systems), and entry level DDR5 is supposed to be 4800.

[Vishera]

12 hours ago, Escanor said:

He wants 300% more bandwidth 

Damn right!,It's time to move on from dual channel,it's especially beneficial for high core count CPUs like the 3950X.

[HumdrumPenguin]

Ryzen will likely show serious raw power when ddr5 comes in.

[Helpful Tech Witch]

4 hours ago, Vishera said:

,It's time to move on from dual channel,

Move on from dual channel? My PC boot loops in dual channel. I have to use single channel.

[AngryBeaver]

Intel is going to be in a tough spot for the gaming crowd and those needing high core counts. For the rest of the world though it isn't going to hurt them too bad. They still have the majority of server market share and that is where the money is. AMD is encroaching on them, but still hasn't taken it away despite having really good high core chips.

 

Intel really needs to kick the monolithic chips though. It is hurting them when it comes to yields and binning. Now there new 10xx processors with the new IHS is actually pretty nice for gaming, but I would rather have more cores for the similar price if we are only talking 10-20fps loss at levels that are already higher than a lot of monitor refresh rates.

 

So if AMD can either get clocks or IPC up enough to close the gap... then they have a true win. If dx12 moves away from single core draw calls then AMD has a win. I mean don't get me wrong... it is amazing what intel is able to milk out of this old architecture, but it is reaching it's limits so they need to move to a better process soon if they want to stay competitive.

[Vishera]

1 hour ago, TheTechWizardThatNeedsHelp said:

Move on from dual channel? My PC boot loops in dual channel. I have to use single channel.

That's because your hardware is faulty,don't blame dual channel for it...

[Helpful Tech Witch]

1 hour ago, Vishera said:

3 hours ago, TheTechWizardThatNeedsHelp said:

 

That's because your hardware is faulty,don't blame dual channel for it...

I had thermal paste on the cpu

[5x5]

32 minutes ago, comander said:

RocketLake is based on ICL and on the roadmap for a 14nm backport. Intel is positioning 8C RKL as an i9 alternative to 10C CML, which implies pretty substantial single core uplift, presumably via IPC. 

 

As stated, ICL has the best IPC of anything out right now and the worst clock speed. Both matter. 

Well, if you're hoping Rocket Lake will be better than what we have at the moment, you'll be disappointed. I've been right on quite a few things thus far and I really hope I'm wrong but from what I'm hearing, Rocket Lake is on the verge of getting cancelled outright.

EDIT: Without saying too much - look at the 5775C for a good indication of where things are going.

[porina]

34 minutes ago, 5x5 said:

look at the 5775C for a good indication of where things are going.

Coincidentally I just got my 5775C system updated today (hasn't been used in a while) and I'm updating my 5675C system right now.

 

For those who may not be familiar with these, these were Broadwell desktop parts, kinda. Broadwell was the process step update after Haswell, but the transition from 22nm to 14nm did not go well. Broadwell came out in mobile and server/HEDT versions, but was very late on desktop. It could be said it was a launch to say they shipped it, as it was not long before Skylake came along anyway. Intel didn't make a regular desktop CPU as such. It was essentially a mobile part repackaged for desktop. I actually wouldn't mind getting a mobile Ice Lake CPU in desktop package just to play with. With Broadwell, the unique feature it carried over from the mobile parts was the 128MB eDRAM (part of Iris Pro) which works great as a L4 cache for some compute uses. That's in part why I still have those systems. It essentially makes ram bandwidth irrelevant for some compute cases that are heavily ram dependant.

 

In trying to draw parallels to today, we have Ice Lake mobile around for a while now. Ice Lake server is meant to be available towards the end of the year. There is no Ice Lake desktop part, but Rocket Lake kinda sits in that spot. We can't say if it will be late or not until around this time next year which is when the generation refresh is due. Ice Lake uses Sunny Cove cores. I'm unclear from rumours if Rocket lake will be Sunny Cove or its successor Willow Cove. Unlike Ice Lake which is built on 10nm, Rocket Lake is rumoured to be on 14nm still. We're looking at next gen architecture on last gen process, and possible complications from that. The die size will be larger than it otherwise would be, and may running hotter, maybe not hit target clocks, whatever they might be.

[5x5]

36 minutes ago, porina said:

Coincidentally I just got my 5775C system updated today (hasn't been used in a while) and I'm updating my 5675C system right now.

 

For those who may not be familiar with these, these were Broadwell desktop parts, kinda. Broadwell was the process step update after Haswell, but the transition from 22nm to 14nm did not go well. Broadwell came out in mobile and server/HEDT versions, but was very late on desktop. It could be said it was a launch to say they shipped it, as it was not long before Skylake came along anyway. Intel didn't make a regular desktop CPU as such. It was essentially a mobile part repackaged for desktop. I actually wouldn't mind getting a mobile Ice Lake CPU in desktop package just to play with. With Broadwell, the unique feature it carried over from the mobile parts was the 128MB eDRAM (part of Iris Pro) which works great as a L4 cache for some compute uses. That's in part why I still have those systems. It essentially makes ram bandwidth irrelevant for some compute cases that are heavily ram dependant.

 

In trying to draw parallels to today, we have Ice Lake mobile around for a while now. Ice Lake server is meant to be available towards the end of the year. There is no Ice Lake desktop part, but Rocket Lake kinda sits in that spot. We can't say if it will be late or not until around this time next year which is when the generation refresh is due. Ice Lake uses Sunny Cove cores. I'm unclear from rumours if Rocket lake will be Sunny Cove or its successor Willow Cove. Unlike Ice Lake which is built on 10nm, Rocket Lake is rumoured to be on 14nm still. We're looking at next gen architecture on last gen process, and possible complications from that. The die size will be larger than it otherwise would be, and may running hotter, maybe not hit target clocks, whatever they might be.

Basically spot on. And it's almost certain to be on 14nm.

[dizmo]

 

On 7/16/2020 at 11:56 PM, GDRRiley said:

they haven't when it comes to IPC

intel's already being matched and the fall launches don't look to be good for them. AMD has a way to go on launch stability sure but given the final silicon (unless they need to update it) looks to be in testing now a late 2020 launch should be stable.

Like I said, I'll believe it when I see it. Intel isn't matched, their 10th gen is better than AMDs; the 10600k still smokes the 3600x in games.

I could see this bringing them to parity, but they're not at it right now.

[5x5]

5 minutes ago, dizmo said:

 

Like I said, I'll believe it when I see it. Intel isn't matched, their 10th gen is better than AMDs; the 10600k still smokes the 3600x in games.

I could see this bringing them to parity, but they're not at it right now.

Hyperbole.

 

Tuning memory and spending the time to optimize everything, a 3600(X) is within spitting distance (under 3%) of the 10600K even overclocked. Even a 10% IPC gain, half of what is actually coming, will be enough to push even the stock 4600 ahead.

[GDRRiley]

3 minutes ago, dizmo said:

the 10600k still smokes the 3600x in games.

really 2-5% win in some while loosing in others is smoking it?

 

4000 chips will be faster than what intel has in the fall.