Discussion Intel current and future Lakes & Rapids thread

[Geddagod]

Those 56 cores at 5.2GHz are performing like 56 GC's in Raptor at 4.5GHz.

(13400f 4.1 GHz all core) 16038 x (56/6) = 149688
128391/149688 =~ 0.85 x 4.1 = 3.5GHz.
13100f 4.3 GHz all core, 8090 x (56/4) = 113260
113260/149688 = ~0.76 x 4.3 = 3.3 GHz.

Doesn't appear as if you can scale it like that, using different products with different core counts seems to show different scaling factors.

Regardless GLC server prob would have lower PPC in some tasks than RPC due to much lower L3 latency, and I think CB23 doesn't appreciate having way more L3 cache, so the better latency is prob more beneficial for the applications performance (I'm guessing).

 

[aigomorla]

(13400f 4.1 GHz all core) 16038 x (56/6) = 149688
128391/149688 =~ 0.85 x 4.1 = 3.5GHz.
13100f 4.3 GHz all core, 8090 x (56/4) = 113260
113260/149688 = ~0.76 x 4.3 = 3.3 GHz.

Doesn't appear as if you can scale it like that, using different products with different core counts seems to show different scaling factors.

Regardless GLC server prob would have lower PPC in some tasks than RPC due to much lower L3 latency, and I think CB23 doesn't appreciate having way more L3 cache, so the better latency is prob more beneficial for the applications performance (I'm guessing).

i honestly think you would MELT the socket and TOAST the vrm's on a 56core Raptor.
You'd probably also pop the house breaker while your at it too.

 

[itsmydamnation]

Its traded blows with a 96core AMD Epyc 9474F.
It displaced it with a little over half the cores.

I don't think AMD TR will even be able to clock that high with that many cores.
It gets exponentially more difficult to get higher clocks when you have that many cores.

.

im sorry but this is pure cope ,

1. a 9474F is a 48 core part , i assume you mean dual socket,?
2. 48x 3.6 x 2 = 345 "effective" ghz vs say 96 x 5.4 = 518 "effective" ghz
345 / 518 = 1.5 , 123,593 * 1.5 = 185,389
185,389 / 132,484 = 1.4 times better
3. the exact same Zen4 CCD's can hit 6.7ghz on LN2 in CB23 so i am being very generous
4. this is LN2 mate , and its just more CCD's , the CCD's are spaced apart so there is no increase in any host spots, heat density is the same and surface area is increased at a linear rate.


SPH is a bad, very late , power hungry and you are coping, your continued random proclamations like anyone in this target market would run on water etc are just additional stories you tell yourself to convince yourself your not on copium, but you are.

sorry to be the bearer of bad news.

 

[itsmydamnation]

(13400f 4.1 GHz all core) 16038 x (56/6) = 149688
128391/149688 =~ 0.85 x 4.1 = 3.5GHz.
13100f 4.3 GHz all core, 8090 x (56/4) = 113260
113260/149688 = ~0.76 x 4.3 = 3.3 GHz.

Doesn't appear as if you can scale it like that, using different products with different core counts seems to show different scaling factors.

Regardless GLC server prob would have lower PPC in some tasks than RPC due to much lower L3 latency, and I think CB23 doesn't appreciate having way more L3 cache, so the better latency is prob more beneficial for the applications performance (I'm guessing).

Cb23 runs mostly in L2 , it could be cache/fetch policy differences or maybe the huge memory latency of SPR when you do get the occasional miss. I dont think L3 has almost any impact.

cheers

 

[aigomorla]

sorry to be the bearer of bad news.

Look at the results..

Cinebench - R23 Multi Core with BenchMate overclocking records @ HWBOT

Overclocking records

hwbot.org

1. Intel
2. Intel
3. Intel
4. 2 x EYPC

And why are you thread bashing in an intel thread?
I am pulling facts.

Did Intel take World Record?
Yes

Did it outrank 2 x EYPC a 96core system? Yes

Did it use LN2?
Yes it did, but guess what so did the Thread Rippers which are ranked 5, and 6.

Again WHO is at the TOP with the World Record.
Intel.

For heavens sake why are you derailing the thread... do you want me to remind you what our Forum Director said about Derailing it with AMD?

Why are you guys who are all Salty about Intel actually having the crown trying to shoot its achievement.

So this is the the last guy that did exactly what you are doing now.... And our Forum Director's Comment.

Water cooled stock Server Genoa can do 120,000 points. Sapphire Rapids HEDT is a Joke at stock. Zen4 Threadripper will break 200,000 points no problem

Want to talk about AMD and their exploits?
Post in an AMD titled thread.
Not here.


esquared
Anandtech Forum Director

So i will tell you exactly what our forum director said to that last person that happened to do exactly what you did...

Take your AMD exploits out of here.

If StormPeak takes takes the World Record even with LN2, then lets talk.
Until then you are derailing this thread.

 

[Thunder 57]

Look at the results..

Cinebench - R23 Multi Core with BenchMate overclocking records @ HWBOT

Overclocking records

hwbot.org

1. Intel
2. Intel
3. Intel
4. 2 x EYPC

And why are you thread bashing in an intel thread?
I am pulling facts.

Did Intel take World Record?
Yes

Did it outrank 2 x EYPC a 96core system? Yes

Did it use LN2?
Yes it did, but guess what so did the Thread Rippers which are ranked 5, and 6.

Again WHO is at the TOP with the World Record.
Intel.

For heavens sake why are you derailing the thread... do you want me to remind you what our Forum Director said about Derailing it with AMD?

Why are you guys who are all Salty about Intel actually having the crown trying to shoot its achievement.

So this is the the last guy that did exactly what you are doing now.... And our Forum Director's Comment.


So i will tell you exactly what our forum director said to that last person that happened to do exactly what you did...

Take your AMD exploits out of here.

If StormPeak takes takes the World Record even with LN2, then lets talk.
Until then you are derailing this thread.

From your own link, 1, 2, 3 ane udner LN2 at 5.2.5.4GHz. Number 4 is AMD running at 4.0GHz and stock cooling. That's just apples and oranges. I'd rather just have an honest conversation than say "it's an Intel thread, get out!".




You need to pay attention to the mod warnings.


esquared
Anandtech Forum Director

 

[Thunder 57]

Its traded blows with a 96core AMD Epyc 9474F.
It displaced it with a little over half the cores.

I don't think AMD TR will even be able to clock that high with that many cores.
It gets exponentially more difficult to get higher clocks when you have that many cores.



Raptor doesn't even have 56 cores, so your pulling theoretical as we don't even know if Raptor can do 4.5ghz @ 56 cores. Which i highly doubt, as there is just no way you can get that much power to the socket even if there was one.

The 3495 was recorded to pull 1kw.
Which is absolutely insane.

And if you scale SPR down, i don't think you can do a core to core comparison even.

But i would really like to see what a 13900K can do vs a 3435X.
Here is some scores for a overclocked 3435x

SkatterBencher #61: Intel Xeon w5-3435X Overclocked to 5200 MHz - SkatterBencher

We overclock the Intel Xeon w5-3435X up to 5200 MHz with the ASUS Pro WS W790E-Sage SE motherboard and EK-Pro water cooling.

skatterbencher.com

View attachment 82511

Its really annoying how there is very little coverage on this cpu from more reputable sources.
Its like they are afraid to show it.



1KW! lol... it was stated somewhere else.
Your going to need a 240V 30amp socket if you have a 4090 along with it, and did some massive benching.

Horray for graphs that don't start at 0 and make the gains look much greater than they are! Reminds me of this gem:

That sure is some 6%!

 

[coercitiv]

That sure is some 6%!

Almost TWICE as big as the 30% in the other chart!

 

[Hulk]

i honestly think you would MELT the socket and TOAST the vrm's on a 56core Raptor.
You'd probably also pop the house breaker while your at it too.

A 56 core "mythical" Raptor would pull about 12 Watts per core at 4.5GHz running CB R23. Or 672 Watts. It's a moot point of course but it doesn't and never will exist.
Just pointing out something is going on with SPR IPC on CB R23 vs GC in Raptor.

 

[Thunder 57]

Almost TWICE as big as the 30% in the other chart!

And I'll bet that SPR LN2 overclocks use more than TWICE the power of Zen stock. And that fmax, what a joke. That graph is beyond stupid.

 

[Geddagod]

And I'll bet that SPR LN2 overclocks use more than TWICE the power of Zen stock. And that fmax, what a joke. That graph is beyond stupid.

The graph is dumb, but 6% fmax using the same arch and from just adding BPD is pretty impressive. Especially if Intel could implement this into their newest Intel 20A product (ARL) immediately, rather than using it for a 'refresh' gen.
Intel usually doesn't see large ST frequency uplifts, AFAIK, at least recently. ST freq had stagnated ~5.3 GHz for Intel for a while now until RPL. SKL, TGL, RKL, ADL, all saw around that same level of frequency. And it looks like RPL was just the combination of an extremely mature node with some amazing DTCO, based on where MTL frequency rumors seem to put it as.
The last really 'huge' ST freq uplift was Zen 4 from Zen 3- at 16%. BPD providing 6% freq uplift, just from that tech alone, not counting additional arch, node, and design freq uplifts seems to be a a pretty good deal.

 

[H433x0n]

The graph is dumb, but 6% fmax using the same arch and from just adding BPD is pretty impressive. Especially if Intel could implement this into their newest Intel 20A product (ARL) immediately, rather than using it for a 'refresh' gen.
Intel usually doesn't see large ST frequency uplifts, AFAIK, at least recently. ST freq had stagnated ~5.3 GHz for Intel for a while now until RPL. SKL, TGL, RKL, ADL, all saw around that same level of frequency. And it looks like RPL was just the combination of an extremely mature node with some amazing DTCO, based on where MTL frequency rumors seem to put it as.
The last really 'huge' ST freq uplift was Zen 4 from Zen 3- at 16%. BPD providing 6% freq uplift, just from that tech alone, not counting additional arch, node, and design freq uplifts seems to be a a pretty good deal.

I think you'll be surprised with RPL-R. The Intel 7 node got another uplift similar to the 3rd gen Intel 7 process used in RPL-S.

 

[eek2121]

I think you'll be surprised with RPL-R. The Intel 7 node got another uplift similar to the 3rd gen Intel 7 process used in RPL-S.

false.

 

[Geddagod]

I think the best I can see them doing is bringing the 14900k up the 13900KS frequency of 6GHz and then bringing a potential 14900k frequency up to 6.2 or 6.3 GHz. We already see OC'd 13900ks go up to 6.2-6.3GHz peak ST freq, so a limited edition 14900KS skus getting up to those frequencies isn't out of the ream of possibility IMO.

 

[Hulk]

I think the best I can see them doing is bringing the 14900k up the 13900KS frequency of 6GHz and then bringing a potential 14900k frequency up to 6.2 or 6.3 GHz. We already see OC'd 13900ks go up to 6.2-6.3GHz peak ST freq, so a limited edition 14900KS skus getting up to those frequencies isn't out of the ream of possibility IMO.

I agree but I care not about ST frequency. For me the Raptor Refresh lives or dies on how much less power than Raptor it uses on all cores at iso frequencies. Raptor can go 5.5GHz on all cores with ridiculous power/cooling. If the Refresh can do 5.5GHz all cores on something like 200"ish" Watts that would be something to consider for current ADL/RPL owners I think.

 

[H433x0n]

I think the best I can see them doing is bringing the 14900k up the 13900KS frequency of 6GHz and then bringing a potential 14900k frequency up to 6.2 or 6.3 GHz. We already see OC'd 13900ks go up to 6.2-6.3GHz peak ST freq, so a limited edition 14900KS skus getting up to those frequencies isn't out of the ream of possibility IMO.

I don’t know the performance nor the frequencies. At iso frequency it’s supposed to use less energy from an enhanced node. How that shakes out in a final product is anybody’s guess.

The only thing I know for sure is that techtubers are going to lambast it because motherboard vendors will default to an unlimited PL1/PL2.

 

[Exist50]

At iso frequency it’s supposed to use 10-15% less energy from an enhanced node.

I think you're confusing the process gains RPL had, if not just their lumped together "Intel 7" numbers.

 

[H433x0n]

I think you're confusing the process gains RPL had, if not just they’re lumped together "Intel 7" numbers.

It’s an enhanced/tweaked process coupled with DLVR. We’ll find out what efficiency gains it all adds up to when it releases.

 

[Exist50]

It’s an enhanced/tweaked process coupled with DLVR. We’ll find out what efficiency gains it all adds up to when it releases.

On what grounds do you assume either?

 

[H433x0n]

On what grounds do you assume either?

None. Take it for what it’s worth, which could be nothing. We’ll see when it’s announced.

 

[nicalandia]

Absolute Beast....

You do know that CBR23 will not scale past 256 Threads right? and that a single stock 96 core EPYC can do 110,000 points. Disabling SMT on a 2S Genoa EPYC should yield more than 180,000 points at stock.

Some one here with 2S Genoa Would like to test CBR23 With SMT off?

 

[nicalandia]

From YuuKi_Ans

https://twitter.com/x/status/1675921701392232449

 

[Henry swagger]

https://twitter.com/x/status/1675530238606942217

In 10 years raptor lake vs zen 4 will have the same gap.. wider cores age better 😁

 

[aigomorla]

In 10 years raptor lake vs zen 4 will have the same gap.. wider cores age better 😁

in 10 yrs my smart watch would be faster then a raptor lake.

 

[eek2121]

in 10 yrs my smart watch would be faster then a raptor lake.

Geekbench/Cinebench on Apple watch when? 🤣

Note that he is comparing a mid tier part to a high end part. I mean come on, this is an Intel thread and all, but that is really reaching…

At the very least, compare it to a 3800X. The 3700X has a 65W tdp, the 9900k…well…😏

EDIT: If @Henry swagger or someone wants to donate a 9900k + motherboard, I will resurrect my 3900X system to compare and post results in a new thread. 😉

Pretty sure I have at least 2 X570 boards collecting dust, and my 3900X is sitting in a coffee cup with a few other chips…

I do wish I still had my 2600k (donated to a family member). I would love to benchmark current games on that and all the other CPUs I have.