Intel's Future Chips: News, Rumours & Reviews

[YoAndy]

Coffe lake engineering sample doing 5GHz

You can OC it to 6Ghz using liquid nitrogen but the retail version will be stuck at poor 4.3Ghz single core and 4Ghz on all cores, what a lame cpu.

Are you trying to say that they will not be able to be overclocked? That is a patently false statement.

No, I wasn't trying to say that.

I speak from the point of view of those ones who want to buy a CPU without OC them, yes they do exist. Such persons, me included, would like to have the fastest CPU for gaming when changing rig every lot of years and they choose the quickest game CPU therefore I7 7700K.

We are sad to know that Coffee Lake will be worst in gaming than a year old i7 7700k.

I do not want to overclock CPUs, I hope you won't blame me for that

And you are making that assumption on what? There are no retail versions of the chips yet just a few engineering samples that could or could not be like the final retail chip. So you are just speculating at this point.

Statistics say that Engineering Samples are always toned down in terms of Ghz by roughly 10%.

Unless this is the very first case in history that an ES is toned down by a 25-30%, the final numbers are quite closely predictable as f.i. just happened for Skylake and Kaby Lake when ESs were roughly 10% toned down.

In addition being Coffee Lake 14nm the IPC will be the same, despite Intel labels it 14nm++ just to catch the marketing hype. Skylake IPC was the same of Kaby Lake IPC which is the same of Coffee Lake IPC because they share the very same 14nm old archtecture.

You are one of a few rare people ho spends premium on a K unlocked processor that doesn't want to overclock it. And then blames the company. Is like asking Ferrari to sell you a Minivan faster than a formula one car. Those processors are like formula one cars, you have to tweak them

You made my day mate.

Cause I am not an OClocker, in your opinion, I can't buy a K CPU and complain it is slower in gaming than a year older CPU. Funny.

I am entitled to buy a K CPU even if I don't overclock it, that's because the 8700 without K doesn't have the same frequencies of the K version, therefore your statement pertains to those snob elitarian guys which OC CPU and think the non oveerclockers ones have to be wiped out from earth.

Wrong.

I have fiddled with OC a while in my life, I am just too old too get fun out of it anymore.

Intel is producing IMHO another Skylake X fiasco if to have the most FPS out of a game you are forced to buy an year old CPU called 7700k.

The i7 7700K 4.5Ghz Turbo (at the moment of this writing) is $340 USD and the i7 7700 4.2 Ghz turbo is $305 USD. Is the extra 300 MHz worth the extra $35 USD?. NO. That's a chunk of money that would be better spent towards a better GPU, monitor, nicer case, prostitutes etc(whatever you are in to). The price/performance isn't worth it. The K processors are only worth getting if you're planning on overclocking, because they can overclock and beat the performance of more expensive CPU's. Gamers we benefit a lot using unlocked processors because (after overclocking) single core performance increases dramatically(example) The i7 2600k 3.8 GHz stock can be overclocked to 5GHz.

 

[-Fran-]

The i7 7700K 4.5Ghz Turbo (at the moment of this writing) is $340 USD and the i7 7700 4.2 Ghz turbo is $305 USD. Is the extra 300 MHz worth the extra $35 USD?. NO. That's a chunk of money that would be better spent towards a better GPU, monitor, nicer case, prostitutes etc(whatever you are in to). The price/performance isn't worth it. The K processors are only worth getting if you're planning on overclocking, because they can overclock and beat the performance of more expensive CPU's. Gamers we benefit a lot using unlocked processors because (after overclocking) single core performance increases dramatically(example) The i7 2600k 3.8 GHz stock can be overclocked to 5GHz.

What? 300Mhz not worth it? If you get a 300Mhz OC, you're a happy camper with any CPU. Why is it not worth it from a non-K to a K? You're forcing your own preconceptions onto someone else, specially with the "prostitutes" unnecessary snarky remark. "Value" has no objective measure and that one of the fundamental stones of Capitalism (well, egoism and negotiation are, but oh well).

And no, the 2600K did not reach 5Ghz. Mine, a 2700K even, goes up to 5Ghz with unsafe voltage and temps. I have it at 4.6Ghz, but Sandy is an outlier. See Ivy and Haswell for more "average" cases. 500Mhz is your average *on all cores and turbo*.

Cheers!

 

[YoAndy]

The i7 7700K 4.5Ghz Turbo (at the moment of this writing) is $340 USD and the i7 7700 4.2 Ghz turbo is $305 USD. Is the extra 300 MHz worth the extra $35 USD?. NO. That's a chunk of money that would be better spent towards a better GPU, monitor, nicer case, prostitutes etc(whatever you are in to). The price/performance isn't worth it. The K processors are only worth getting if you're planning on overclocking, because they can overclock and beat the performance of more expensive CPU's. Gamers we benefit a lot using unlocked processors because (after overclocking) single core performance increases dramatically(example) The i7 2600k 3.8 GHz stock can be overclocked to 5GHz.

What? 300Mhz not worth it? If you get a 300Mhz OC, you're a happy camper with any CPU. Why is it not worth it from a non-K to a K? You're forcing your own preconceptions onto someone else, specially with the "prostitutes" unnecessary snarky remark. "Value" has no objective measure and that one of the fundamental stones of Capitalism (well, egoism and negotiation are, but oh well).

And no, the 2600K did not reach 5Ghz. Mine, a 2700K even, goes up to 5Ghz with unsafe voltage and temps. I have it at 4.6Ghz, but Sandy is an outlier. See Ivy and Haswell for more "average" cases. 500Mhz is your average *on all cores and turbo*.

Cheers!

I had one i7 2600K and it was stable at 5Ghz. And I was trying to be funny don't take it so hard (but personally) I don't think 300 Mhz is worth 35.

Are you overclocking?
YES= buy it
NO= nope

 

[Yarvolino]

Interesting informations

https://hardforum.com/threads/coffee-lake-lga-1151-6c-12t-launching-in-august-several-k-models-planned.1930226/page-16

https://www.overclock3d.net/news/cpu_mainboard/leaked_intel_i7_8700k_specifications/1

To me those numbers ( overclock3d ) seem to be extreeeeeeeeeeemely optimistic, I would tone them down by a severe 20-30%

 

[Yarvolino]

Coffe lake engineering sample doing 5GHz

You can OC it to 6Ghz using liquid nitrogen but the retail version will be stuck at poor 4.3Ghz single core and 4Ghz on all cores, what a lame cpu.

Are you trying to say that they will not be able to be overclocked? That is a patently false statement.

No, I wasn't trying to say that.

I speak from the point of view of those ones who want to buy a CPU without OC them, yes they do exist. Such persons, me included, would like to have the fastest CPU for gaming when changing rig every lot of years and they choose the quickest game CPU therefore I7 7700K.

We are sad to know that Coffee Lake will be worst in gaming than a year old i7 7700k.

I do not want to overclock CPUs, I hope you won't blame me for that

And you are making that assumption on what? There are no retail versions of the chips yet just a few engineering samples that could or could not be like the final retail chip. So you are just speculating at this point.

Statistics say that Engineering Samples are always toned down in terms of Ghz by roughly 10%.

Unless this is the very first case in history that an ES is toned down by a 25-30%, the final numbers are quite closely predictable as f.i. just happened for Skylake and Kaby Lake when ESs were roughly 10% toned down.

In addition being Coffee Lake 14nm the IPC will be the same, despite Intel labels it 14nm++ just to catch the marketing hype. Skylake IPC was the same of Kaby Lake IPC which is the same of Coffee Lake IPC because they share the very same 14nm old archtecture.

You are one of a few rare people ho spends premium on a K unlocked processor that doesn't want to overclock it. And then blames the company. Is like asking Ferrari to sell you a Minivan faster than a formula one car. Those processors are like formula one cars, you have to tweak them

You made my day mate.

Cause I am not an OClocker, in your opinion, I can't buy a K CPU and complain it is slower in gaming than a year older CPU. Funny.

I am entitled to buy a K CPU even if I don't overclock it, that's because the 8700 without K doesn't have the same frequencies of the K version, therefore your statement pertains to those snob elitarian guys which OC CPU and think the non oveerclockers ones have to be wiped out from earth.

Wrong.

I have fiddled with OC a while in my life, I am just too old too get fun out of it anymore.

Intel is producing IMHO another Skylake X fiasco if to have the most FPS out of a game you are forced to buy an year old CPU called 7700k.

The i7 7700K 4.5Ghz Turbo (at the moment of this writing) is $340 USD and the i7 7700 4.2 Ghz turbo is $305 USD. Is the extra 300 MHz worth the extra $35 USD?. NO. That's a chunk of money that would be better spent towards a better GPU, monitor, nicer case, prostitutes etc(whatever you are in to). The price/performance isn't worth it. The K processors are only worth getting if you're planning on overclocking, because they can overclock and beat the performance of more expensive CPU's. Gamers we benefit a lot using unlocked processors because (after overclocking) single core performance increases dramatically(example) The i7 2600k 3.8 GHz stock can be overclocked to 5GHz.

As you correctly told above, better to go with prostitutes than buying a lame CPU

 

[YoAndy]

For the ones saying 2600K won't reach 5GHz or above, and because yours didn't that doesn't mean anything. Mine was stable at 5.15GHz with a custom liquid set up

5.15 GHz http://www.userbenchmark.com/UserRun/4337740

6.65 GHz http://www.userbenchmark.com/UserRun/4315571

5 GHz http://www.userbenchmark.com/UserRun/4310524

5.65 GHz http://www.userbenchmark.com/UserRun/4399674

5.30 GHz http://www.userbenchmark.com/UserRun/4384843

7.6GHz http://www.userbenchmark.com/UserRun/4327162

8.5 GHz http://www.userbenchmark.com/UserRun/4327867

5.05 GHz http://www.userbenchmark.com/UserRun/4311993

5.09 GHz http://www.userbenchmark.com/UserRun/4330211

5.70 GHz http://www.userbenchmark.com/UserRun/4425554

5.95 GHz http://www.userbenchmark.com/UserRun/4361694

 

[-Fran-]

For the ones saying 2600K won't reach 5GHz or 6GHz Mine was stable at 5.15GHz with a custom liquid set up

6.65GHz http://www.userbenchmark.com/UserRun/4315571

5GHz http://www.userbenchmark.com/UserRun/4310524

5.65GHz http://www.userbenchmark.com/UserRun/4399674

I'm using a TT Frio which is ~40USD HSF. To me, CLWCs are not worth it and custom ones even less even with the higher OC potential. Your were talking about things that "made sense" and not. You cost doubles with any CLWC/CustomLoop and the closer the CPU is to the limit, the less Hertz you will get (see cold bugs and/or Nitrogen cooling).

Also, see this: http://cpu.userbenchmark.com/Compare/Intel-Core-i7-4770K-vs-Intel-Core-i7-3770K/1537vs1317

EDIT: And this: http://cpu.userbenchmark.com/Compare/Intel-Core-i7-4770K-vs-Intel-Core-i7-2600K/1537vs621

The averages are exactly where I remember them for improved cooling. All the links you're providing are outlisers.

Cheers!

 

[TMTOWTSAC]

People pay for convenience in quite literally every single product that has ever existed. That's the origin of products, paying someone else to make/do something rather than doing it yourself. Just think of the 7700k as being a pre-configured 7700 set to run at +600 MHz stock for an extra $35. People pay more than $35 to have someone else replace a Hard Drive or add extra RAM. They pay more for 4 cheap case fans to keep turboboost at max. They pay hundreds for custom water loops that aren't even guaranteed to squeeze an extra 200 MHz out of their chip. If the 7700 is too slow, $35 more for a 7700k is a no-brainer.

 

[juanrga]

Now on SKL-X on the new review, I am ignoring power consumption figures because many SKL-X tests used buggy BIOS launch, which affected turbo and p-state policies, and thus power consumption figures and overclocking capacities. THe own review mentions something about BIOS updates.

And you haven't even made half the effort of taking into account the buggy BIOS'es and AGESA upgrades for/from AMD in articles when laying out your points that mostly favor Intel.

Two different situations.

Launch of X299 platform was rushed. Reviewers did test with beta BIOS that affected performance, overclocking and power consumption. The problem was in the mobo software, e.g. turbo3 not working, not in the microarchitecture. Some site retested with the updated BIOS and noticed nice improvements in performance

Launch of RyZen was also rushed. But the latency problems affecting performance in games weren't the result of preliminary BIOS/AGESA. The problem was in the microarchitecture. No new AGESA/BIOS is going to eliminate the huge latency from the CCX-CCX interconnect; only a complete redesign of the microarchitecture can do it. This is the reason why I said then that future new versions of AGESA could not solve the issue. And time just gave me the reason; as observed above the 1800X is even loosing to a 10C Broadwell 6950X in games.

 

[juanrga]

You are WRONG! Watch the video, because your comments I know you didn't. Overclocking the mesh to 3GHz caused random game crashes, and on boot failure to find the SSD. The 7700K wasn't limited in the vast majority of the games and neither was the 7800X, and those were the games he focused on. The mesh for the most part gave 3-6% gains over stock performance, but once overclocked those gains diminished to 1-3% overall. Plus system instability.

I have friends have gone above 3GHz stable mesh oc. And I gave benches with oc the mesh providing 25% performance gains.

The reason why this review only got 3% performance gains even when overclocking by huge 34% amounts is because the CPU is being bottlenecked.

It is ~20nm vs ~16nm.

The SKL-X review has changed the measurement procedure, but the main conclusions are the same than in the old review.

Older review
i7-6900k: 198W
R7-1800X: 238W
Gap: 20.2%

Newest review
i7-6900k: 155W
R7-1800X: 193W
Gap: 24.5%

On both cases the '95W' chip is consuming more power than the 140W chip, specially when we consider platforms variations as dual-channel vs quad-channel.

Now on SKL-X on the new review, I am ignoring power consumption figures because many SKL-X tests used buggy BIOS launch, which affected turbo and p-state policies, and thus power consumption figures and overclocking capacities. THe own review mentions something about BIOS updates.

You typically ignore things that don't help your point! Nevermind that 1st generation Ryzen is exactly that. The first generation. Intel has been optimizing the same uAch for the last ~10 years, and they have a superior process node. You pick a test that shows the greatest disparity between the two as an example which is ridiculous, because Ryzen as I stated before wasn't made to run excel it was made for multi-tasking. And when you compare the two on power consumption and performance Ryzen looks really good! And I think ASML who makes the machines for all the semiconductor foundries might have a better idea about how analyze the process nodes, and come up with an equation to do this better than random commentators on forums.

The only that I have ignored are both statistically irrelevant and wrong results. I explained why the methodology on AT review is wrong. They measured power with 256bit and 512bit workloads, but then measured performance on benches cannot use those vectors, which generated a false efficiency gap for RyZen that doesn't correspond with reality. It is just the contrary. RyZen is more inefficient than Broadwell.

I used a pair of alternative benches to demonstrate how the 95W RyZen consumes 20--25% more power than 140W Intel chips. I can bring a third bench

The R7 is consuming 26% more than the 6900k, but is only 16% faster. Therefore RyZen is less efficient, and that without considering the review is measuring total platform power, not CPU power.

Your mention that RyZen is the fist gen Zen doesn't change a bit what I am saying. I am not discussing about future muarchs. I am refuting the popular myth that RyZen is more efficient than Broadwell chips.

For SKL-X the situation is more complex, some reviews show a picture and other show another, due to huge variations on mobos and BIOS. For instance Hexus got

The 8C SKL is 12% faster than 8C Zen, but consumes 14% more. Could we conclude that RyZen is 2% more efficient? No, because the reviewer had problems with the mobo he used for the 8C SKL, which consumed a disproportionate amount of power at idle, even more than the 10C SKL

 

[juanrga]

Interesting informations

https://hardforum.com/threads/coffee-lake-lga-1151-6c-12t-launching-in-august-several-k-models-planned.1930226/page-16

https://www.overclock3d.net/news/cpu_mainboard/leaked_intel_i7_8700k_specifications/1

The 4.3GHz all core turbo implies that the engineering sample clocked at 5GHz wasn't using LN2 😉

[And no, the 2600K did not reach 5Ghz. Mine, a 2700K even, goes up to 5Ghz with unsafe voltage and temps. I have it at 4.6Ghz, but Sandy is an outlier. See Ivy and Haswell for more "average" cases. 500Mhz is your average *on all cores and turbo*.

The averages are exactly where I remember them for improved cooling. All the links you're providing are outlisers.

Average OC for i7-2660k is 5054MHz on air.

 

[-Fran-]

Average OC for i7-2660k is 5054MHz on air.

Sigh... You're really going to argue that when even the link I provided stated 4.8Ghz as average?

Plus, all other averages are lower than Sandy proving my point as well.

What the hell, Juan?

Do you even have a source for it?

EDIT: And in regards the AGESA stuff. That is just one part of the whole argument. So you're saying the Mesh topology Intel introduced is flawless? How can you be that biased?

 

[goldstone77]

You are WRONG! Watch the video, because your comments I know you didn't. Overclocking the mesh to 3GHz caused random game crashes, and on boot failure to find the SSD. The 7700K wasn't limited in the vast majority of the games and neither was the 7800X, and those were the games he focused on. The mesh for the most part gave 3-6% gains over stock performance, but once overclocked those gains diminished to 1-3% overall. Plus system instability.

I have friends have gone above 3GHz stable mesh oc. And I gave benches with oc the mesh providing 25% performance gains.

The reason why this review only got 3% performance gains even when overclocking by huge 34% amounts is because the CPU is being bottlenecked.

It is ~20nm vs ~16nm.

The SKL-X review has changed the measurement procedure, but the main conclusions are the same than in the old review.

Older review
i7-6900k: 198W
R7-1800X: 238W
Gap: 20.2%

Newest review
i7-6900k: 155W
R7-1800X: 193W
Gap: 24.5%

On both cases the '95W' chip is consuming more power than the 140W chip, specially when we consider platforms variations as dual-channel vs quad-channel.

Now on SKL-X on the new review, I am ignoring power consumption figures because many SKL-X tests used buggy BIOS launch, which affected turbo and p-state policies, and thus power consumption figures and overclocking capacities. THe own review mentions something about BIOS updates.

You typically ignore things that don't help your point! Nevermind that 1st generation Ryzen is exactly that. The first generation. Intel has been optimizing the same uAch for the last ~10 years, and they have a superior process node. You pick a test that shows the greatest disparity between the two as an example which is ridiculous, because Ryzen as I stated before wasn't made to run excel it was made for multi-tasking. And when you compare the two on power consumption and performance Ryzen looks really good! And I think ASML who makes the machines for all the semiconductor foundries might have a better idea about how analyze the process nodes, and come up with an equation to do this better than random commentators on forums.

The only that I have ignored are both statistically irrelevant and wrong results. I explained why the methodology on AT review is wrong. They measured power with 256bit and 512bit workloads, but then measured performance on benches cannot use those vectors, which generated a false efficiency gap for RyZen that doesn't correspond with reality. It is just the contrary. RyZen is more inefficient than Broadwell.

I used a pair of alternative benches to demonstrate how the 95W RyZen consumes 20--25% more power than 140W Intel chips. I can bring a third bench

The R7 is consuming 26% more than the 6900k, but is only 16% faster. Therefore RyZen is less efficient, and that without considering the review is measuring total platform power, not CPU power.

Your mention that RyZen is the fist gen Zen doesn't change a bit what I am saying. I am not discussing about future muarchs. I am refuting the popular myth that RyZen is more efficient than Broadwell chips.

For SKL-X the situation is more complex, some reviews show a picture and other show another, due to huge variations on mobos and BIOS. For instance Hexus got

The 8C SKL is 12% faster than 8C Zen, but consumes 14% more. Could we conclude that RyZen is 2% more efficient? No, because the reviewer had problems with the mobo he used for the 8C SKL, which consumed a disproportionate amount of power at idle, even more than the 10C SKL

It's obvious now that you will manipulate, skew, and change your point beyond reason to force some kind of hypothetical win. This is like playing thermal nuclear war. The only way to win is by not playing the game.

 

[juanrga]

Average OC for i7-2660k is 5054MHz on air.

Sigh... You're really going to argue that when even the link I provided stated 4.8Ghz as average?

Plus, all other averages are lower than Sandy proving my point as well.

What the hell, Juan?

Do you even have a source for it?

EDIT: And in regards the AGESA stuff. That is just one part of the whole argument. So you're saying the Mesh topology Intel introduced is flawless? How can you be that biased?

The average MHz value I quoted is from the hwbot database.

No one here said that the mesh interconnect is "flawless"; all the technological designs have both strong and weak points. What has been pointed here is that the original bad gaming performance of the SKL-X (with the i9-7900X chip loosing to the 10C Broadwell-E in all games) was a result of buggy BIOS launch, and the reason why retesting with the new BIOS changed the situation, as demonstrated in the benches given. It has been also demonstrated that overclocking the mesh gives nice performance gains of up to 25% on games, but even on stock settings, the 8C SKL-X CPU can be up to 27% faster than 8C RyZen on games

No magic AGESA/BIOS update for the AM4 platform is going to close that gap.

 

[-Fran-]

The average MHz value I quoted is from the hwbot database.

No one here said that the mesh interconnect is "flawless"; all the technological designs have both strong and weak points. What has been pointed here is that the original bad gaming performance of the SKL-X (with the i9-7900X chip loosing to the 10C Broadwell-E in all games) was a result of buggy BIOS launch, and the reason why retesting with the new BIOS changed the situation, as demonstrated in the benches given. It has been also demonstrated that overclocking the mesh gives nice performance gains of up to 25% on games, but even on stock settings, the 8C SKL-X CPU can be up to 27% faster than 8C RyZen on games

Spoiler

No magic AGESA/BIOS update for the AM4 platform is going to close that gap.

Look at the minimum FPS. Can you explain that?

At least you recognize the Mesh topology is not flawless, but you're selling short what AMD has with Infinity Fabric and you're not even looking at the good points of it's topology. What you see in the average/max FPS'es from Boiling Lake X is best case scenarios for the uArch. The big average of games is not that much better than Zen. Plus, I'd say the Zen uArch is holding it's own quite well being lower clocked and with less CPU resources in the in-core.

And no, I'm not implying AGESA is a "silver bullet", but the 3% uplift by fixing the buggy RAM support goes a long way when your *targeted* rival was within 5% or less from you. Put into context 2 BIG things: Intel has the better process AND more years in polishing the uArch. Even if this is the Mesh debut, the uArch behind the in-core is not.

Cheers!

 

[juanrga]

The average MHz value I quoted is from the hwbot database.

No one here said that the mesh interconnect is "flawless"; all the technological designs have both strong and weak points. What has been pointed here is that the original bad gaming performance of the SKL-X (with the i9-7900X chip loosing to the 10C Broadwell-E in all games) was a result of buggy BIOS launch, and the reason why retesting with the new BIOS changed the situation, as demonstrated in the benches given. It has been also demonstrated that overclocking the mesh gives nice performance gains of up to 25% on games, but even on stock settings, the 8C SKL-X CPU can be up to 27% faster than 8C RyZen on games

Spoiler

No magic AGESA/BIOS update for the AM4 platform is going to close that gap.

Look at the minimum FPS. Can you explain that?

At least you recognize the Mesh topology is not flawless, but you're selling short what AMD has with Infinity Fabric and you're not even looking at the good points of it's topology. What you see in the average/max FPS'es from Boiling Lake X is best case scenarios for the uArch. The big average of games is not that much better than Zen. Plus, I'd say the Zen uArch is holding it's own quite well being lower clocked and with less CPU resources in the in-core.

And no, I'm not implying AGESA is a "silver bullet", but the 3% uplift by fixing the buggy RAM support goes a long way when your *targeted* rival was within 5% or less from you. Put into context 2 BIG things: Intel has the better process AND more years in polishing the uArch. Even if this is the Mesh debut, the uArch behind the in-core is not.

Minimums look okay for one of the benches, one the other bench one of the datapoints could be an outlier or anything else.

AMD problem is not only on Infinity Fabric. Any other interconnect is not going to eliminate the latency penalty caused by a memory controller being optimized for throughput (as some of us said before launch), neither a new interconnect, is going to eliminate the huge die-die latencies on EPYC.

The latency gap wasn't 5%, but much more, that is why the new AGESAs with optimizations have almost changed nothing and recent reviews still put Zen well behind. On Ashes 8-core RyZen is still 20% behind 8-core Broadwell either on average or in minimum framerates.

 

[-Fran-]

The average MHz value I quoted is from the hwbot database.

No one here said that the mesh interconnect is "flawless"; all the technological designs have both strong and weak points. What has been pointed here is that the original bad gaming performance of the SKL-X (with the i9-7900X chip loosing to the 10C Broadwell-E in all games) was a result of buggy BIOS launch, and the reason why retesting with the new BIOS changed the situation, as demonstrated in the benches given. It has been also demonstrated that overclocking the mesh gives nice performance gains of up to 25% on games, but even on stock settings, the 8C SKL-X CPU can be up to 27% faster than 8C RyZen on games

Spoiler

No magic AGESA/BIOS update for the AM4 platform is going to close that gap.

Look at the minimum FPS. Can you explain that?

At least you recognize the Mesh topology is not flawless, but you're selling short what AMD has with Infinity Fabric and you're not even looking at the good points of it's topology. What you see in the average/max FPS'es from Boiling Lake X is best case scenarios for the uArch. The big average of games is not that much better than Zen. Plus, I'd say the Zen uArch is holding it's own quite well being lower clocked and with less CPU resources in the in-core.

And no, I'm not implying AGESA is a "silver bullet", but the 3% uplift by fixing the buggy RAM support goes a long way when your *targeted* rival was within 5% or less from you. Put into context 2 BIG things: Intel has the better process AND more years in polishing the uArch. Even if this is the Mesh debut, the uArch behind the in-core is not.

Minimums look okay for one of the benches, one the other bench one of the datapoints could be an outlier or anything else.

AMD problem is not only on Infinity Fabric. Any other interconnect is not going to eliminate the latency penalty caused by a memory controller being optimized for throughput (as some of us said before launch), neither a new interconnect, is going to eliminate the huge die-die latencies on EPYC.

The latency gap wasn't 5%, but much more, that is why the new AGESAs with optimizations have almost changed nothing and recent reviews still put Zen well behind. On Ashes 8-core RyZen is still 20% behind 8-core Broadwell either on average or in minimum framerates.

So you have no explanation for that. Figures. But then you go and make a "solid" argument for Intel having the lead without an idea why those numbers come to be. You know, it's ironic how the same data set you display tells 2 stories simultaneously and you can't see both.

Also, when I mention 5%, I'm talking about performance difference and not latency differences. Don't forget the data has to travel in the IF BUS for all cores. The IMC's are centralized, so the IF is used for all purpose transport; hell they even go as far as to make it explicit it can replace PCIe for GPUs much like nVLink does. Even with all the """problems""" Infinity Fabric has in Zen 1, it's doing quite well and holding it own. You're even in the *need* to compare it with the new Mesh Intel had to put forth to replace the Ring BUS (QPI?) in their *next gen CPU* just released with God knows how much extra money behind for R&D.

And all of this is not even seeing software taking all the NUMA capabilities they introduced with Zen. I can guarantee Zen is still not showing all the cards on the table and Intel knows that. Have you seen the AES throughput metrics?

Cheers!

 

[juanrga]

The average MHz value I quoted is from the hwbot database.

No one here said that the mesh interconnect is "flawless"; all the technological designs have both strong and weak points. What has been pointed here is that the original bad gaming performance of the SKL-X (with the i9-7900X chip loosing to the 10C Broadwell-E in all games) was a result of buggy BIOS launch, and the reason why retesting with the new BIOS changed the situation, as demonstrated in the benches given. It has been also demonstrated that overclocking the mesh gives nice performance gains of up to 25% on games, but even on stock settings, the 8C SKL-X CPU can be up to 27% faster than 8C RyZen on games

Spoiler

No magic AGESA/BIOS update for the AM4 platform is going to close that gap.

Look at the minimum FPS. Can you explain that?

At least you recognize the Mesh topology is not flawless, but you're selling short what AMD has with Infinity Fabric and you're not even looking at the good points of it's topology. What you see in the average/max FPS'es from Boiling Lake X is best case scenarios for the uArch. The big average of games is not that much better than Zen. Plus, I'd say the Zen uArch is holding it's own quite well being lower clocked and with less CPU resources in the in-core.

And no, I'm not implying AGESA is a "silver bullet", but the 3% uplift by fixing the buggy RAM support goes a long way when your *targeted* rival was within 5% or less from you. Put into context 2 BIG things: Intel has the better process AND more years in polishing the uArch. Even if this is the Mesh debut, the uArch behind the in-core is not.

Minimums look okay for one of the benches, one the other bench one of the datapoints could be an outlier or anything else.

AMD problem is not only on Infinity Fabric. Any other interconnect is not going to eliminate the latency penalty caused by a memory controller being optimized for throughput (as some of us said before launch), neither a new interconnect, is going to eliminate the huge die-die latencies on EPYC.

The latency gap wasn't 5%, but much more, that is why the new AGESAs with optimizations have almost changed nothing and recent reviews still put Zen well behind. On Ashes 8-core RyZen is still 20% behind 8-core Broadwell either on average or in minimum framerates.

So you have no explanation for that. Figures. But then you go and make a "solid" argument for Intel having the lead without an idea why those numbers come to be. You know, it's ironic how the same data set you display tells 2 stories simultaneously and you can't see both.

Also, when I mention 5%, I'm talking about performance difference and not latency differences. Don't forget the data has to travel in the IF BUS for all cores. The IMC's are centralized, so the IF is used for all purpose transport; hell they even go as far as to make it explicit it can replace PCIe for GPUs much like nVLink does. Even with all the """problems""" Infinity Fabric has in Zen 1, it's doing quite well and holding it own. You're even in the *need* to compare it with the new Mesh Intel had to put forth to replace the Ring BUS (QPI?) in their *next gen CPU* just released with God knows how much extra money behind for R&D.

And all of this is not even seeing software taking all the NUMA capabilities they introduced with Zen. I can guarantee Zen is still not showing all the cards on the table and Intel knows that. Have you seen the AES throughput metrics?

I don't have explanation for a single datapoint that seems to be an outlier. I have explanation why in general SKL-X is faster than BDW-E on games and much faster than RyZen.

I was also talking about performance. The original gap was about 20% clock for clock. A 5% improvement was not going to change things significantly. I said this before the new AGESA/BIOS were released and it turned to be right. On the other hand all those that believed the myth spread in certain forums about future magic AGESA/BIOS fixing gaming are still awaiting for the miracle.

Infinity Fabric is a derivation of the old SEAMICRO 'supercomputer' fabric, which has very little of supercomputer and it was more focused to microservers

AMD purchase all that by huge amounts of money, loosing money due to another wrong decision.

Intel has been developing the mesh interconnect during many years as part o the advanced R&D they do on manycores. And started using it in the commercial Phi line. Meshes are also used in the manycores of small companies.

 

[-Fran-]

Infinity Fabric is a derivation of the old SEAMICRO 'supercomputer' fabric, which has very little of supercomputer and it was more focused to microservers

AMD purchase all that by huge amounts of money, loosing money due to another wrong decision.

Intel has been developing the mesh interconnect during many years as part o the advanced R&D they do on manycores. And started using it in the commercial Phi line. Meshes are also used in the manycores of small companies.

Oh, I know what IF is. That is why I think it's not a problem at all. AMD chose a very smart implementation for it that fits all it's future endeavors with semi-custom designs and integration of SoCs. The concept behind IF is not new for the world, but AMD's implementation is and what they're aiming to do with it. I strongly suggest you keep an eye on how it develops alongside nVLink. Intel is going to have a fit when nVidia moves nVLink to the consumer arena and proves how good it is. AMD is taking IF along that path as well and I think their first serious integration effort will be displayed with the APUs and, probably, VEGA.

Cheers!

 

[juanrga]

Infinity Fabric is a derivation of the old SEAMICRO 'supercomputer' fabric, which has very little of supercomputer and it was more focused to microservers

AMD purchase all that by huge amounts of money, loosing money due to another wrong decision.

Intel has been developing the mesh interconnect during many years as part o the advanced R&D they do on manycores. And started using it in the commercial Phi line. Meshes are also used in the manycores of small companies.

Oh, I know what IF is. That is why I think it's not a problem at all. AMD chose a very smart implementation for it that fits all it's future endeavors with semi-custom designs and integration of SoCs. The concept behind IF is not new for the world, but AMD's implementation is and what they're aiming to do with it. I strongly suggest you keep an eye on how it develops alongside nVLink. Intel is going to have a fit when nVidia moves nVLink to the consumer arena and proves how good it is. AMD is taking IF along that path as well and I think their first serious integration effort will be displayed with the APUs and, probably, VEGA.

I also know what IF is and can do, and the reason why my interest is zero.

Nvidia is using nVlink now only as a stopgap to avoid PCIe limits. Their future SoC will use a mesh interconnect, like Intel, except it is more advanced that what we have now in commercial chips. Former Nvidia desing used a 16x16 mesh. Lately they increased to 16x32.

Intel didn't chose mesh fortuitously, but because it is the optimal interconnect in the design space of different options: buses, rings,... Fujitsu using mesh/torus for the future post-K supercomputer is not fortuitous either.

 

[jaymc]

The average MHz value I quoted is from the hwbot database.

No one here said that the mesh interconnect is "flawless"; all the technological designs have both strong and weak points. What has been pointed here is that the original bad gaming performance of the SKL-X (with the i9-7900X chip loosing to the 10C Broadwell-E in all games) was a result of buggy BIOS launch, and the reason why retesting with the new BIOS changed the situation, as demonstrated in the benches given. It has been also demonstrated that overclocking the mesh gives nice performance gains of up to 25% on games, but even on stock settings, the 8C SKL-X CPU can be up to 27% faster than 8C RyZen on games

Spoiler

No magic AGESA/BIOS update for the AM4 platform is going to close that gap.

Look at the minimum FPS. Can you explain that?

At least you recognize the Mesh topology is not flawless, but you're selling short what AMD has with Infinity Fabric and you're not even looking at the good points of it's topology. What you see in the average/max FPS'es from Boiling Lake X is best case scenarios for the uArch. The big average of games is not that much better than Zen. Plus, I'd say the Zen uArch is holding it's own quite well being lower clocked and with less CPU resources in the in-core.

And no, I'm not implying AGESA is a "silver bullet", but the 3% uplift by fixing the buggy RAM support goes a long way when your *targeted* rival was within 5% or less from you. Put into context 2 BIG things: Intel has the better process AND more years in polishing the uArch. Even if this is the Mesh debut, the uArch behind the in-core is not.

Minimums look okay for one of the benches, one the other bench one of the datapoints could be an outlier or anything else.

AMD problem is not only on Infinity Fabric. Any other interconnect is not going to eliminate the latency penalty caused by a memory controller being optimized for throughput (as some of us said before launch), neither a new interconnect, is going to eliminate the huge die-die latencies on EPYC.

The latency gap wasn't 5%, but much more, that is why the new AGESAs with optimizations have almost changed nothing and recent reviews still put Zen well behind. On Ashes 8-core RyZen is still 20% behind 8-core Broadwell either on average or in minimum framerates.

So you have no explanation for that. Figures. But then you go and make a "solid" argument for Intel having the lead without an idea why those numbers come to be. You know, it's ironic how the same data set you display tells 2 stories simultaneously and you can't see both.

Also, when I mention 5%, I'm talking about performance difference and not latency differences. Don't forget the data has to travel in the IF BUS for all cores. The IMC's are centralized, so the IF is used for all purpose transport; hell they even go as far as to make it explicit it can replace PCIe for GPUs much like nVLink does. Even with all the """problems""" Infinity Fabric has in Zen 1, it's doing quite well and holding it own. You're even in the *need* to compare it with the new Mesh Intel had to put forth to replace the Ring BUS (QPI?) in their *next gen CPU* just released with God knows how much extra money behind for R&D.

And all of this is not even seeing software taking all the NUMA capabilities they introduced with Zen. I can guarantee Zen is still not showing all the cards on the table and Intel knows that. Have you seen the AES throughput metrics?

I don't have explanation for a single datapoint that seems to be an outlier. I have explanation why in general SKL-X is faster than BDW-E on games and much faster than RyZen.

I was also talking about performance. The original gap was about 20% clock for clock. A 5% improvement was not going to change things significantly. I said this before the new AGESA/BIOS were released and it turned to be right. On the other hand all those that believed the myth spread in certain forums about future magic AGESA/BIOS fixing gaming are still awaiting for the miracle.

Infinity Fabric is a derivation of the old SEAMICRO 'supercomputer' fabric, which has very little of supercomputer and it was more focused to microservers

AMD purchase all that by huge amounts of money, loosing money due to another wrong decision.

Intel has been developing the mesh interconnect during many years as part o the advanced R&D they do on manycores. And started using it in the commercial Phi line. Meshes are also used in the manycores of small companies.

How was it a wrong decision for AMD to purchase the IP for Infinity Fabric it was a genius move ?

I agree with Yuka all the way, we will see it being used for heterogeneous compute with the first implementation being Vega and then in APU's and SOC's... As I say genius move on AMD's behalf.

 

[juanrga]

The average MHz value I quoted is from the hwbot database.

No one here said that the mesh interconnect is "flawless"; all the technological designs have both strong and weak points. What has been pointed here is that the original bad gaming performance of the SKL-X (with the i9-7900X chip loosing to the 10C Broadwell-E in all games) was a result of buggy BIOS launch, and the reason why retesting with the new BIOS changed the situation, as demonstrated in the benches given. It has been also demonstrated that overclocking the mesh gives nice performance gains of up to 25% on games, but even on stock settings, the 8C SKL-X CPU can be up to 27% faster than 8C RyZen on games

Spoiler

No magic AGESA/BIOS update for the AM4 platform is going to close that gap.

Look at the minimum FPS. Can you explain that?

At least you recognize the Mesh topology is not flawless, but you're selling short what AMD has with Infinity Fabric and you're not even looking at the good points of it's topology. What you see in the average/max FPS'es from Boiling Lake X is best case scenarios for the uArch. The big average of games is not that much better than Zen. Plus, I'd say the Zen uArch is holding it's own quite well being lower clocked and with less CPU resources in the in-core.

And no, I'm not implying AGESA is a "silver bullet", but the 3% uplift by fixing the buggy RAM support goes a long way when your *targeted* rival was within 5% or less from you. Put into context 2 BIG things: Intel has the better process AND more years in polishing the uArch. Even if this is the Mesh debut, the uArch behind the in-core is not.

Minimums look okay for one of the benches, one the other bench one of the datapoints could be an outlier or anything else.

AMD problem is not only on Infinity Fabric. Any other interconnect is not going to eliminate the latency penalty caused by a memory controller being optimized for throughput (as some of us said before launch), neither a new interconnect, is going to eliminate the huge die-die latencies on EPYC.

The latency gap wasn't 5%, but much more, that is why the new AGESAs with optimizations have almost changed nothing and recent reviews still put Zen well behind. On Ashes 8-core RyZen is still 20% behind 8-core Broadwell either on average or in minimum framerates.

So you have no explanation for that. Figures. But then you go and make a "solid" argument for Intel having the lead without an idea why those numbers come to be. You know, it's ironic how the same data set you display tells 2 stories simultaneously and you can't see both.

Also, when I mention 5%, I'm talking about performance difference and not latency differences. Don't forget the data has to travel in the IF BUS for all cores. The IMC's are centralized, so the IF is used for all purpose transport; hell they even go as far as to make it explicit it can replace PCIe for GPUs much like nVLink does. Even with all the """problems""" Infinity Fabric has in Zen 1, it's doing quite well and holding it own. You're even in the *need* to compare it with the new Mesh Intel had to put forth to replace the Ring BUS (QPI?) in their *next gen CPU* just released with God knows how much extra money behind for R&D.

And all of this is not even seeing software taking all the NUMA capabilities they introduced with Zen. I can guarantee Zen is still not showing all the cards on the table and Intel knows that. Have you seen the AES throughput metrics?

I don't have explanation for a single datapoint that seems to be an outlier. I have explanation why in general SKL-X is faster than BDW-E on games and much faster than RyZen.

I was also talking about performance. The original gap was about 20% clock for clock. A 5% improvement was not going to change things significantly. I said this before the new AGESA/BIOS were released and it turned to be right. On the other hand all those that believed the myth spread in certain forums about future magic AGESA/BIOS fixing gaming are still awaiting for the miracle.

Infinity Fabric is a derivation of the old SEAMICRO 'supercomputer' fabric, which has very little of supercomputer and it was more focused to microservers

AMD purchase all that by huge amounts of money, loosing money due to another wrong decision.

Intel has been developing the mesh interconnect during many years as part o the advanced R&D they do on manycores. And started using it in the commercial Phi line. Meshes are also used in the manycores of small companies.

How was it a wrong decision for AMD to purchase the IP for Infinity Fabric it was a genius move ?

I agree with Yuka all the way, we will see it being used for heterogeneous compute with the first implementation being Vega and then in APU's and SOC's... As I say genius move on AMD's behalf.

IF is not anything special or revolutionary and HSA continues being a marketing term with zero industry traction after many years of hype. Intel approach to heterogeneity is much better, and Intel is selling real production systems that solve real-life problems, whereas AMD only sell slides and hype.


The problem wasn't on acquiring SeaMicro IP. As stated above, the problem was on the huge cost of acquiring the whole of Seamicro ($334 million), then mismanaging the whole business, then loosing additional money on projects canceled latter (anyone remember Seattle, Cambridge, Skybridge?), and finally closing the Seamicro division because the business was not sustainable.

Add another $75 million lost related to the sale of the business as well as an additional $12 million charge related to a restructuring plan: Dense server subsidiary shut down after AMD posts a bigger-than-expected loss

 

[goldstone77]

IF is not anything special or revolutionary and HSA continues being a marketing term with zero industry traction after many years of hype. Intel approach to heterogeneity is much better, and Intel is selling real production systems that solve real-life problems, whereas AMD only sell slides and hype.


The problem wasn't on acquiring SeaMicro IP. As stated above, the problem was on the huge cost of acquiring the whole of Seamicro ($334 million), then mismanaging the whole business, then loosing additional money on projects canceled latter (anyone remember Seattle, Cambridge, Skybridge?), and finally closing the Seamicro division because the business was not sustainable.

Add another $75 million lost related to the sale of the business as well as an additional $12 million charge related to a restructuring plan: Dense server subsidiary shut down after AMD posts a bigger-than-expected loss

You want to talk about mismanagement and instability? How about 11 billion worth! I wonder where they got all that money from? Playing fair and competing on a level playing field through innovation? 10 years on the same design and multiple generations of re-branded CPU's, because they brought their closest competition to the brink of bankruptcy by fighting dirtier than anyone else on the block.

Intel Is Probably The Worst Aquirer In Tech History
Steve Cheney, Steve's Blog
Sep. 21, 2010, 12:01 PM
"At last week’s Intel Developer Forum, CEO Paul Otellini talked about this massive acquisition spree like it’s predestined to succeed. But if we look back in time, Intel’s pitiful M&A track record suggests this couldn’t be further from the truth."

"Intel spent over $11 billion buying about 40 companies, and the vast majority of these acquisitions failed. In fact, of the 15 largest acquisitions in its history, Intel has shut down or sold off the acquired products in every single case (aside from Wind River which is too recent to include)."

"This is an awe-inspiringly bad track record, and likely puts Intel as the worst acquirer in tech history. There isn’t even a Wikipedia acquisition page for Intel like there is for Google, Microsoft, Cisco, and Apple. If I were a conspiracy theorist, I’d say Intel’s corporate PR department had Wikipedia remove the page.

Only kidding. But the truth is, Intel doesn’t want the media and public to remember how bad they are at M&A. They want everyone to believe it will be different this time."
http://www.businessinsider.com/intel-is-probably-the-worst-aquirer-in-tech-history-2010-9

Edit: has anyone seen that movie "The Boy in the Plastic Bubble."

 

[juanrga]

IF is not anything special or revolutionary and HSA continues being a marketing term with zero industry traction after many years of hype. Intel approach to heterogeneity is much better, and Intel is selling real production systems that solve real-life problems, whereas AMD only sell slides and hype.


The problem wasn't on acquiring SeaMicro IP. As stated above, the problem was on the huge cost of acquiring the whole of Seamicro ($334 million), then mismanaging the whole business, then loosing additional money on projects canceled latter (anyone remember Seattle, Cambridge, Skybridge?), and finally closing the Seamicro division because the business was not sustainable.

Add another $75 million lost related to the sale of the business as well as an additional $12 million charge related to a restructuring plan: Dense server subsidiary shut down after AMD posts a bigger-than-expected loss

You want to talk about mismanagement and instability? How about 11 billion worth! I wonder where they got all that money from? Playing fair and competing on a level playing field through innovation? 10 years on the same design and multiple generations of re-branded CPU's, because they brought their closest competition to the brink of bankruptcy by fighting dirtier than anyone else on the block.

Intel Is Probably The Worst Aquirer In Tech History
Steve Cheney, Steve's Blog
Sep. 21, 2010, 12:01 PM
"At last week’s Intel Developer Forum, CEO Paul Otellini talked about this massive acquisition spree like it’s predestined to succeed. But if we look back in time, Intel’s pitiful M&A track record suggests this couldn’t be further from the truth."

"Intel spent over $11 billion buying about 40 companies, and the vast majority of these acquisitions failed. In fact, of the 15 largest acquisitions in its history, Intel has shut down or sold off the acquired products in every single case (aside from Wind River which is too recent to include)."

"This is an awe-inspiringly bad track record, and likely puts Intel as the worst acquirer in tech history. There isn’t even a Wikipedia acquisition page for Intel like there is for Google, Microsoft, Cisco, and Apple. If I were a conspiracy theorist, I’d say Intel’s corporate PR department had Wikipedia remove the page.

Only kidding. But the truth is, Intel doesn’t want the media and public to remember how bad they are at M&A. They want everyone to believe it will be different this time."
http://www.businessinsider.com/intel-is-probably-the-worst-aquirer-in-tech-history-2010-9

Edit: has anyone seen that movie "The Boy in the Plastic Bubble."

No one here said that Intel is perfect. No one here said that people working at Intel doesn't do mistakes. Intel mobile pretensions were a joke and that cost them billions in loses. Still Intel owns the 99% of server market, 3/4 of the PC market, and about a 10% of the HPC market (I mean the Phi line). Intel continues making profits each quarter, whereas in the other hand AMD just reported another lost this quarter.

 

[jaymc]

AMD turned a $25 million operating profit in Q2 2017

Operating income of $25 million, though it turned in a net loss of $16 million. Revenue was up 19% year-over year, and the company's operating profit marks its first time in the black by that metric in quite some time.

This is significant since it is a return to profitability and the first time AMD has posted a positive EPS in a long time.

Plus they raised their guidance for Q3.