Raptor Lake Leaks + Intel 4 developments

[Vince]

But the problem is that the Core i5 13600KF is faster in gaming and faster in non-gaming applications:

OcUK Intel Raptor Lake review thread

Because you can use 8000 ddr5 ram on the Intel, and you can't on AMD? With 3D memory of CPUs and Direct Storage on GPUs, next year is the year RAM goes to die doesn't have any affect on gaming anymore

forums.overclockers.co.uk

Look through the figures - once I compiled the review thread and started reading the reviews I was surprised.

You have to consider its not just 6 "big" cores,but 8 "small" cores with Skylake level performance. Its like having a downclocked Core i7 9700 plonked onto the six core Raptorlake CPU.

The Core i5 13600KF is as fast or faster than a Ryzen 7 7700X in many non-gaming benchmarks. It's frequently beating a Ryzen 9 5900X in video encoding benchmarks and at times can get close to a Ryzen 9 5950X!!

Its not even close in non-gaming scenarios at all,and the Ryzen 5 7600X is consistently slower overall. Even with DDR4 you are seeing at worst about 10% lower gaming and application performance,and a lot of times not even that.

The added problem,is that the Core i5 13600KF is £350ish. The Ryzen 7 5800X3D is £420. So even if you get a £100 AM4 motherboard,you can still get a very good B660 DDR4 motherboard for £150~£160. The total price is about the same.

The Core i5 13600KF still wins in value. I can see a Core i5 13600KF lasting much longer. Plus think about gamers who stream or capture video,having 8 Skylake level cores will make a big difference for multi-tasking.

So you really are spending more with a slower Ryzen 5 7600X to try and become "future proof" and as a person who is on B450,AMD tried to stop Zen3 on 400 series motherboards. The amount of arguments I had on here when that happened,and only because of the backlash they relented. But they still removed PCI-E 4.0 from later AGESAs meaning B450/X470 lost partial or full PCI-E 4.0 support.

There is no guarantee that with Zen5 AMD might not have a technical hitch which means it needs newer motherboards. But even it works,a Ryzen 5 7600X will run out of steam quicker anyway which means you need to upgrade quicker. You need to price upgrades very carefully on the same socket otherwise,it does not add up financially. Many don't so it's not always cost effective staying on the same socket. I did,and stuck to my upgrade budget very strictly. If not it is more cost effective to sell what you have and change platforms.

It's a tough one for sure. I'm not sure there is a right option tbh, no doubt the 13600k is attractive but any option is decent here at the mid range and I think you could make arguments either way depending on what you have or if you are starting from scratch. What a time to live in aye, a time where we can actually argue that both are viable for a number of different reasons. Back in bulldozer era I couldn't even imagine having this conversation just a few short years later.

 

[CAT-THE-FIFTH]

It's a tough one for sure. I'm not sure there is a right option tbh, no doubt the 13600k is attractive but any option is decent here at the mid range and I think you could make arguments either way depending on what you have or if you are starting from scratch. What a time to live in aye, a time where we can actually argue that both are viable for a number of different reasons. Back in bulldozer era I couldn't even imagine having this conversation just a few short years later.

I agree - but I think AMD has gotten a tad complacent because of its Zen2/Zen3 wins. They have given Intel a space to do this,and the Core i5 13600KF is still a $300 Core i5 which isn't a great price. But it looks good relative to the Ryzen 5 7600X. AMD has tried to do another Ryzen 5 5600X,but the issue is this time they haven't gone and beaten Intel(to justify the high price). Considering how small the dies in the Ryzen 5 are,they could easily have made this is a much cheaper CPU. I personally think they underestimated Intel. If anything it's concerning because Meteor Lake is on TSMC 5NM and uses Intel's answer to chiplets. The node advantage AMD will have,is going to shrink. Raptorlake is on Intel 10NM which inferior to TSMC 7NM. So AMD probably has a 1.5 node advantage here,and much greater transistor density.

 

[Vince]

I agree - but I think AMD has gotten a tad complacent because of its Zen2/Zen3 wins. They have given Intel a space to do this,and the Core i5 13600KF is still a $300 Core i5 which isn't a great price. But it looks good relative to the Ryzen 5 7600X. AMD has tried to do another Ryzen 5 5600X,but the issue is this time they haven't gone and beaten Intel(to justify the high price). Considering how small the dies in the Ryzen 5 are,they could easily have made this is a much cheaper CPU. I personally think they underestimated Intel. If anything it's concerning because Meteor Lake is on TSMC 5NM and uses Intel's answer to chiplets. The node advantage AMD will have,is going to shrink. Raptorlake is on Intel 10NM which inferior to TSMC 7NM. So AMD probably has a 1.5 node advantage here,and much greater transistor density.

Meteor lake they surely won't fab on TSMC... Somebody said just yesterday its taping out on Intel 4? Intel have a lot to prove with their answer to chiplets so I am super excited to see it. I just hope they have a better interconnect than EMIB that actually works and can scale, hopefully a working Foveros product that would be exciting. Also the intel speed is in the ringbus right now so how they approach that is again going to be super exciting... These next few releases will be all out war! We just need to grab some popcorn and sit back

I'm actually fairly convinced they wont fab the cpu on TSMC thinking about it... the packaging tech is generally part of the node etc so they would need to do a lot of work with TSMC to implement their interconnect and packaging at TSMC which im not convinced they would do... Ive been wrong before though, many times in fact so an intel chip on TSMC would be interesting to see.

 

[CAT-THE-FIFTH]

Meteor lake they surely want fab on TSMC... Somebody said just yesterday its taping out on Intel 4? Intel have a lot to prove with their answer to chiplets so I am super excited to see it. I just hope they have a better interconnect than EMIB that actually works and can scale, hopefully a working Foveros product that would be exciting. Also the intel speed is in the ringbus right now so how they approach that is again going to be super exciting... These next few releases will be all out war! We just need to grab some popcorn and sit back

I intend to hold out until Zen5,and certainly I would need dGPU prices to get better before replacing what I have.

 

[Robert896r1]

Meteor lake they surely won't fab on TSMC... Somebody said just yesterday its taping out on Intel 4? Intel have a lot to prove with their answer to chiplets so I am super excited to see it. I just hope they have a better interconnect than EMIB that actually works and can scale, hopefully a working Foveros product that would be exciting. Also the intel speed is in the ringbus right now so how they approach that is again going to be super exciting... These next few releases will be all out war! We just need to grab some popcorn and sit back

iGPU (chiplet) is TSMC. Main cores are Intel 4 for MTL.

 

[Vince]

iGPU (chiplet) is TSMC. Main cores are Intel 4 for MTL.

Yea id go with that. Intel 4 should be interesting they really need to smash that out though as we can't be dealing with another sapphire rapids.

 

[CAT-THE-FIFTH]

iGPU (chiplet) is TSMC. Main cores are Intel 4 for MTL.

That makes sense. So is Intel 4 better than TSMC 5NM? Or about the same?

 

[Curlyriff]

What? From cinebench numbers, the 7950x is 9% more efficient at 125w. That's not a lot is it?

Yeah it is. 9% when we have 70 computers running overall, it is hugely important especially with rising costs to run PC's. People have been having their minds blown that Intel 13900k is 3-4% faster in gaming to the 7900x/7950x but that is at efficiency cost.

 

[Vince]

That makes sense. So is Intel 4 better than TSMC 5NM? Or about the same?

Correct me if I am wrong but I believe intel 7 is intels 10nm node, intel 4 is their 7nm node. Intel nodes on the whole are generally slightly better at the same nm designation.

 

[CAT-THE-FIFTH]

Correct me if I am wrong but I believe intel 7 is intels 10nm node, intel 4 is their 7nm node. Intel nodes on the whole are generally slightly better at the same nm designation.

I wish there was an international standard for all this!

 

[Joxeon]

No idea what is right or wrong in chart overall but the idle usage is pure nonesense for the 5950x and I assume the rest of the 5000 series as per my posts where I am at 8-10watt idle and 25-40 watt with Revit, Spotify, Chrome etc running. Yes the data is wrong on the chart as shown by others.

While the core will run 8-10w I suggest you check the memory controller.

 

[Vince]

While the core will run 8-10w I suggest you check the memory controller.

The SOC's draw will depend on many factors including drives plugged in, memory speed / SoC voltage to attain that speed etc, etc. Much the same way that the controller on Intel works. A single zen core at total idle is like 1 to 3 w usage, less in EPYC... on a system like mine with 8 populated DIMM slots running overclocked, 4 populated nvme in raid and some 14 hard disks your uncore is going to be doing more than say a system with a single disk and two sticks of memory. That's just how it is regardless of platform. I'd suggest that idle power usage between the platforms is roughly inline so not really a positive or negative that is substantial or worth an argument on either side. In fact if you scale back to the lowest common denominator which would be laptop you would suggest that AMD has the advantage here when it comes to ultra low power draw cpus.

But yea the uncore on mine runs at around 20 to 25 watts but then its a 280w tdp chip with 8 dimms running at an overclocked 3466MT (so I'm shoving LLC and voltage at the SoC to get there over 8 DIMMS). Id say that it's likely a "stock" chip would be within the same realms as the intel chip at idle when you take all that into account...

Humbugs system above "sort of idle" is what 16/17w real world whereas mine with what I would consider idle but actually has 2 spreadsheets, 4 word documents, chia, discord, outlook, 20 odd browser tabs, 2 RDP instances, calculator, notepad and spotify running is "idling" at about 30 ish watts. Put simply both are within a hair of each other in the real world and both are perfectly adequate with nothing wild going on.

Id love to find that graph I once had that shows some interesting observations over the lifetime of an architecture. I will see if I can dig it out.

 

[Joxeon]

What if you don't have DDR4 already. And you do get a gaming performance hit with DDR4 compared to DDR5. So factor in £100 for RAM then.

You can pick up a B650m motherboard for £176.33 already so the cost difference there is £16 not £110! That gives you DDR5 and PCIE 5.0 support to last years just fine.

Also what price are the B760 boards from Intel. We haven't had info but if you want parity spec with PCIE 5.0 and M.2 at 5.0 also then expect the Intel boards to be similar price as well.

That means your saving if you are comparing is £155 not £325. It will be less if you go B760 and DDR5 for Intel too to get the performance shown in reviews gaming wise and back to only £16 difference.

If you don't have DDR4 already then yeah it's better to go with DDR5 but also your likely not a serial upgrader either and just want the best performance available today for your budget in which case Intel is still offering more performance for less money.

 

[Vince]

If you don't have DDR4 already then yeah it's better to go with DDR5 but also your likely not a serial upgrader either and just want the best performance available today for your budget in which case Intel is still offering more performance for less money.

Am sure the DDR5/Board argument will be basically a none argument in 6 to 8 months time when the retailer scalping stops and prices settle down. ID expect to see prices levelling off somewhat as the manufacturers and vendors adapt to the competition and prices get adjusted for the products place in the market. Honestly where we are right now Id really struggle on making a decision on what to buy. Intel seems the obvious choice at the mid range with a 13600, ddr4, z690... As you start going up the stack it seems to get a bit less obvious.

 

[Robert896r1]

Sharing 30mins of R23 on a custom loop

 

[Vince]

Sharing 30mins of R23 on a custom loop

Some very nice temps there dude and some pretty impressive core clocks as well. That average CPU package power is insane though. Is that actually right? An average of basically 400w during that test?

4000w PL1 and 2 power limits - Yea Boi!!! That does look like a very fun chip to play with if you have the right cooling!

 

[JAMIE_CORTEZ]

Some very nice temps there dude and some pretty impressive core clocks as well. That average CPU package power is insane though. Is that actually right? An average of basically 400w during that test?

4000w PL1 and 2 power limits - Yea Boi!!!

Some people would tell you that’s impossible

 

[Robert896r1]

Some very nice temps there dude and some pretty impressive core clocks as well. That average CPU package power is insane though. Is that actually right? An average of basically 400w during that test?

No and this is a common problem people have esp tech tubers. Cpu package power is vid x current.

Reality isn’t that. Where available, you want to look at the voltage regulator section. The power(pout) is the correct reading.

 

[Vince]

Some people would tell you that’s impossible

It's a beast my cooling can't handle my chip at full load at all and if I start overclocking it it just throttles to like 500mhz and makes me cry ... need me some of that cooling!

 

[Vince]

No and this is a common problem people have esp tech tubers. Cpu package power is vid x current.

Reality isn’t that. Where available, you want to look at the voltage regulator section. The power(pout) is the correct reading.

That isn't all that bad then basically up there at the 300 to 350 watts many of the reviewers were spouting. Random question... is there any benefit on cranking the ring frequency on these (if you can that is).