Discussion Intel current and future Lakes & Rapids thread

[nerp]

I really don't see either AMD or Intel leaving each other in the dust anytime soon. This isn't like it was eight to ten years ago where Intel's foundries were hitting on all cylinders and Ad was stuck short funded and fighting with being multiple nodes behind. It seems like most of the IPC advances that we're seeing these days involve trade-offs, like cache restructuring that helps some loads, but hurts a few, or pipelinelength tweaks that help clocks but hurt branches. You get give of transistors thrown at things like machine leartgat is currently barely used, or super wide vector units tgare usable by maybe 5% of your customers, or so many cores that some essentially stay idle all the time.

The competition is going to be as much on system level optimization as it is on core improvements. While Intel is pushing big.LITTLE, AMD is pushing in other directions. Once Intel gets enough OS level buy in, AMD can look at big.LITTLE if they need to.

I just don't see either company performing a paradigm shift anytime soon.

Good perspective here. Sometimes we look at 5% advantage in something as a major win. Or adding 5 cores to come on top of synthetic multicore benchmark as dominance. But peel away the layers and look at the guts, we are so much more close and similar than different.

I'm older now, but there was a time that generational advantages were massive and could mean the difference between unplayable or playable. Now we are splitting hairs to some degree. Besides, GPUs and non-x86 architectures are making inroads into Intel and AMD's turf. Their future struggles might not come from their competition. I picture opposing soccer players stopping during the game and looking to sky to marvel at the large mushroom cloud rising amidst a suddenly orange sky.

 

[Panino Manino]

You get it. TL is a beast and Intel has a good product for their do-over. AMD cannot make one mistake or they're behind again. I will get flamed for this, but AMD is 1-2 years away from being left in the dust forever based on what we see with TL. I caution people to set reasonable expectations for Zen 3.

Just to expect AMD to not be competitive anymore, this will not happening until another "tragedy" strikes. What will happen is that both will trade blows annually as normal.

Don't celebrate too early. Yes, Gen12 GPU is good but we already know too well how much better Navi is. Keep the same CU but switch Vega for RDNA1 and AMD is far ahead again. But that is not the problem, the problem is not that Intel have a faster GPU, is that Intel now have a fast enough GPU. The problem with past Gen was that it was unplayable. If the consumer can still play his games, doesn't matter if it's slower than on the AMD competitor. The importance of AMD keeping their lead is that, to not let sink in the public opinion that Intel IGPs aren't gargabe anymore.

 

[mikk]

Making a driver more compliant does not make it faster - the contrary could very well be the case.

In general a new architecture has more room for future driver improvements because it's new for the driver team, both compatibility and for features/performance. The driver gains for a years old architecture are typically lower, this is not only specific to Intel but also for AMD and Nvidia. Vega has several years of optimizations in the bag.

 

[DrMrLordX]

But 8c TL will come,

My point is that in order for TigerLake-H to matter, it was supposed to be out by now, not later. And anything beyond 10nm is now up in the air since 7nm is apparently b0rked and Murthy was fired over it. Or have people forgotten?

 

[mikk]

It will matter next year, pretty sure. Why do you think it won't matter?

 

[DrMrLordX]

It will matter next year, pretty sure. Why do you think it won't matter?

8c TigerLake-H would have been a successful counter to Renoir. Against Cezanne, not so much. And that's just the mobile arena. If they could have packaged it as a 65W desktop CPU, it could anchor Intel's desktop lineup instead of Rocket Lake until Alder Lake was ready in 2021 . . . assuming it launches then.

Delaying 8c TigerLake raises all kinds of questions about yields that shouldn't even have to be asked in 2020 with respect to Intel 10nm, and yet here we are.

 

[mikk]

8c TigerLake-H would have been a successful counter to Renoir. Against Cezanne, not so much.

I don't think 8C Cezanne can match 8C Tigerlake-H. Cezanne needs to prove it can match the IPC of TGL and clock speeds. You can be sure they can go to 5 Ghz with TGL-H. And either way it will be a big improvement over CFL, so it will matter.

 

[DrMrLordX]

You can be sure they can go to 5 Ghz with TGL-H.

Wait, what? Not even the top-end 4c TigerLake CPUs do that. All-core turbo for the 1185G7 (had to copy-paste that, ugh) is 4.3 GHz. 4.8 GHz for one core.

 

[Shivansps]

Just to expect AMD to not be competitive anymore, this will not happening until another "tragedy" strikes. What will happen is that both will trade blows annually as normal.

Don't celebrate too early. Yes, Gen12 GPU is good but we already know too well how much better Navi is. Keep the same CU but switch Vega for RDNA1 and AMD is far ahead again. But that is not the problem, the problem is not that Intel have a faster GPU, is that Intel now have a fast enough GPU. The problem with past Gen was that it was unplayable. If the consumer can still play his games, doesn't matter if it's slower than on the AMD competitor. The importance of AMD keeping their lead is that, to not let sink in the public opinion that Intel IGPs aren't gargabe anymore.

Seeing Tigerlake i wonder if AMD got internal info and this is why they are trying to get Cezanne out ASAP, because as things stand now, once Tigerlake-H launches AMD will be again in pre-Renoir position, in fact it may even be worse as they dont longer have the IGP lead anymore.

As for Cezanne, AMD is using Vega again, we already know this, they will need to increase the CU count or go LPDDR5 to be in the lead again...

Im not sure whats the hold up as they can do RDNA2 apus already, by the time Cezanne is replaced by the RDNA2 APU, Vega may become the longest lived IGP arch in history, it is fighting hand to hand with Intel Gen9 but Vega is likely to win by a few months and this was a mistake.

 

[LightningZ71]

8c TigerLake-H would have been a successful counter to Renoir. Against Cezanne, not so much. And that's just the mobile arena. If they could have packaged it as a 65W desktop CPU, it could anchor Intel's desktop lineup instead of Rocket Lake until Alder Lake was ready in 2021 . . . assuming it launches then.

Delaying 8c TigerLake raises all kinds of questions about yields that shouldn't even have to be asked in 2020 with respect to Intel 10nm, and yet here we are.

Counter to Renoir?!?! Lets move past fantasy land here for a minute: The only Renoir part that is notably better than the existing 4C+96EU TGL in its full configuration is the 4800U. They are as scarce as hen's teeth the world over, and almost completely nonexistent in the US market. They don't beat TGL in single threaded tasks in anything but the barest few cases, and the win in multi-threaded tasks isn't always that great, and that's for a part that has twice as many cores and twice as many threads. An 8C TGL, with what would likely be 16-24MB of L3, in the typical higher TDP configuration of -H processors, would handily wipe the floor with any of the -H renoir variants, assuming that it had anything approaching proper cooling. And, yes, I'm well aware of the fact that it would have a very much scaled down iGPU. In the -H market, that's of little concern. There is precisely one currently shipping laptop platform that uses a -H Renoir processor without a dGPU. IT may be marketed by a few companies, but it's just a single model on the inside.

Cezanne has to find a way to increase it's single core throughput. 15% isn't going to fill the gap in many of the tests that we've seen over the last week. I do expect that the 15% IPC increase, coupled with the improved CCX design, will make a material improvement on the multi-threaded benchmarks, such that the uplift there may be 20% or more in some cases. And, I also grant that the Intel demonstration platform was setup for rather generous TDP targets, I also contend that most customers don't care one way or the other what the TDP target of their laptop is. If this one works better than that one, or if this one has a longer blue bar than that one has a red bar, then it's the one they're going to buy. 90% of the market is not as tech savy as we are. They won't care.

 

[DrMrLordX]

Seeing Tigerlake i wonder if AMD got internal info and this is why they are trying to get Cezanne out ASAP, because as things stand now, once Tigerlake-H launches AMD will be again in pre-Renoir position, in fact it may even be worse as they dont longer have the IGP lead anymore.

TigerLake-H has a smaller/weaker iGPU than TigerLake-U.

As for Cezanne, AMD is using Vega again, we already know this, they will need to increase the CU count or go LPDDR5 to be in the lead again... in fact they may even need to do both once Intel drivers are more mature.

Or AMD just doesn't care.

Counter to Renoir?!?!

Yes. TigerLake-H would be faster, making it a successful counter. A bit late to the party, mind you, but still. It would suffice. Aren't people still accustomed to Intel having the faster product? Or do we no longer expect that?

 

[mikk]

Wait, what? Not even the top-end 4c TigerLake CPUs do that. All-core turbo for the 1185G7 (had to copy-paste that, ugh) is 4.3 GHz. 4.8 GHz for one core.

TGL-U goes up to 4.8 Ghz which means 5.0 Ghz is expected for TGL-H. Obviously the all core turbo is limited to the power usage and cooling. I refer to the clock speed capability of a core. AMD is quite limited in comparison.

TigerLake-H has a smaller/weaker iGPU than TigerLake-U.

Which doesn't matter for the H parts, all of them will be bundled with a dGPU.

 

[Shivansps]

Or AMD just doesn't care.

I think it is a mistake not to care when Intel that never cared is giving it priority.

 

[LightningZ71]

Seeing Tigerlake i wonder if AMD got internal info and this is why they are trying to get Cezanne out ASAP, because as things stand now, once Tigerlake-H launches AMD will be again in pre-Renoir position, in fact it may even be worse as they dont longer have the IGP lead anymore.

As for Cezanne, AMD is using Vega again, we already know this, they will need to increase the CU count or go LPDDR5 to be in the lead again... in fact they may even need to do both once Intel drivers are more mature.

Im not sure whats the hold up as they can do RDNA2 apus already, by the time Cezanne is replaced by the RDNA2 APU, Vega may become the longest lived IGP arch in history, it is fighting hand to hand with Intel Gen9 but Vega is likely to win by a few months and this was a mistake.

It is not the same state as it used to be in either direction. In no case is a G7 configuration a clear knockout in performance as compared to the 4800u, and that's allowing for the fact that the demonstration platform has the twp turned all the way open on the TDP profile. At best, the 96EU G7 config is slightly better than the 8CU VEGA config in Renoir. There are precious few cases that a game is playable on one and not on the other with the same resolution and detail setup. Also, from a business perspective, Laptop makers are moving almost every sub $900 laptop they make in days once they hit retail shelves. Walk into your local BEst Buy, walmart, what have you, the shelves are rather bare. Time to turn on either product is not a problem.

As for improving things, for Cezanne, all AMD really needs to do to stay "in the game" is take advantage of further power management and process improvements to up the clock on the Vega iGPU a bit more and improve memory access latency some. With improved single threaded performance of the actual processor cores, that factor of game performance will improve, and the higher clocked Vega CUs will be there to keep up with it.

In the end, both products are going to be held back by absolute memory bandwidth. The next big boost to iGPU performance will be DDR5. Going from DDR4-3200 or LPDDR4X 4266 to DDR5 in the 5000+ range will provide at least a 50% boost to memory bandwidth, with full spec DDR5 capable of providing the same bandwidth as the RX 550/560 and the original XBox One's EDRAM cache. Assuming that both vendors can manage thermals, that should give solidly playable performance on almost anything with high quality 1080p settings.

 

[Shmee]

So when is the comet lake successor and what will it be called? I would consider a good gaming CPU from Intel, if they had PCIe gen 4 and enough lanes. Also, I would want at least 8 or 10 cores.

 

[Gideon]

8c TigerLake-H would have been a successful counter to Renoir. Against Cezanne, not so much.

This, and some previous replies expecting Cezanne to "stomp" Tiger lake IMO are setting themselves up to be dissaponted.

Tiger Lake-H will have 24MB of L3 and Tiger Lake has a very real and decent IPC and clock-speed advantage over Renoir. I can see Cezanne closing the gap mostly. Maybe being faster in some cases, but winning by a large enough magint to not consider Tiger Lake-H a counter, really ?!

I Highly highly doubt Cezanne will have more than 16MB L3 cache for instance, it might still be only 8MB (but unified).

I mean, just look at these (multithreaded) spec scores 8/16 Renoir vs 4/8 Tiger Lake U:

Intel’s Tiger Lake 11th Gen Core i7-1185G7 Review and Deep Dive: Baskin’ for the Exotic

www.anandtech.com

Even with 2x less cores, Tiger lake manages to win a few benches (probably avx-512) and be equal in a lot more. Add 2x the cores (with corresponding cache increase) to the graphs, and are you honestly claiming that a cut-down Zen 3 APU (Cezanne) will win. Not trade-blows but win, by a considerable margin?

I think that in reality these CPUs will be very close in perfomance, with AMD maybe having an edge on battery life, and Intel probably on single-threaded bust-perf.

 

[NTMBK]

Wait, did this thread get renamed? Is it now just going to cover all Intel architectures until the heat-death of the universe? It's already almost 300 pages long and covers multiple different topics simultaneously, maybe it's time to start a new thread.

 

[Shivansps]

It is not the same state as it used to be in either direction. In no case is a G7 configuration a clear knockout in performance as compared to the 4800u, and that's allowing for the fact that the demonstration platform has the twp turned all the way open on the TDP profile. At best, the 96EU G7 config is slightly better than the 8CU VEGA config in Renoir. There are precious few cases that a game is playable on one and not on the other with the same resolution and detail setup. Also, from a business perspective, Laptop makers are moving almost every sub $900 laptop they make in days once they hit retail shelves. Walk into your local BEst Buy, walmart, what have you, the shelves are rather bare. Time to turn on either product is not a problem.

As for improving things, for Cezanne, all AMD really needs to do to stay "in the game" is take advantage of further power management and process improvements to up the clock on the Vega iGPU a bit more and improve memory access latency some. With improved single threaded performance of the actual processor cores, that factor of game performance will improve, and the higher clocked Vega CUs will be there to keep up with it.

In the end, both products are going to be held back by absolute memory bandwidth. The next big boost to iGPU performance will be DDR5. Going from DDR4-3200 or LPDDR4X 4266 to DDR5 in the 5000+ range will provide at least a 50% boost to memory bandwidth, with full spec DDR5 capable of providing the same bandwidth as the RX 550/560 and the original XBox One's EDRAM cache. Assuming that both vendors can manage thermals, that should give solidly playable performance on almost anything with high quality 1080p settings.

Remember that Xe is running on very early drivers... you are not going to see consumers products running these drivers.

You need to realise that in some tests this thing is beating a desktop IGP, in Final Fantasy the new Intel IGP is beating the 4750G! At 28W... And while i did not see direct comparison i can assure you Witcher 3 performance is better than the 3400G probably matching the 4750G.

Intel needs to fix and optimise the drivers as this seems to be a alpha version with several issues on it.

So once the drivers are finished and optimised even 65W Renoir may be in trouble here.

 

[eek2121]

Tiger Lake H will have a weaker iGPU because Intel hopes to bundle it’s own discrete offering.

Tiger Lake isn’t meant to compete with Renoir. It is meant to compete with Zen 3.

As far as clocks, Intel’s engineers have stated Tiger Lake can go higher.

 

[Shmee]

Is Tiger lake H the successor to comet lake?

 

[tamz_msc]

Is Tiger lake H the successor to comet lake?

On the desktop its Rocket Lake(rumored), followed by Alder Lake(which is confirmed for 2021).

 

[tamz_msc]

Even with 2x less cores, Tiger lake manages to win a few benches (probably avx-512) and be equal in a lot more.

SPEC2017 doesn't have anything that makes use of AVX512 as far as I'm aware. The main reason why Renoir lags behind Tiger Lake and Ice Lake in most of the non-execution bound tests is because its memory subsystem sucks big time - downclocking the memory to 2666 MHz when memory-intensive programs are being run is most surely a design compromise to fit 8 cores in a 15W envelope.

 

[ondma]

My point is that in order for TigerLake-H to matter, it was supposed to be out by now, not later. And anything beyond 10nm is now up in the air since 7nm is apparently b0rked and Murthy was fired over it. Or have people forgotten?

So now it is you who decides what is relevant? Maybe you should let, you know, the marketplace decide, because that is the ultimate goal, to sell product.

 

[ondma]

It will matter next year, pretty sure. Why do you think it won't matter?

NVM

 

[ondma]

On the desktop its Rocket Lake(rumored), followed by Alder Lake(which is confirmed for 2021).

Cant find the article, but I did see a post with supposed GB5 results for Rocket Lake a couple of days ago. Didnt mention any time frame, but apparently Rocket Lake is still coming.
Actually, a higher TDP version of 8 core TL would seem to me to be a better option, but yield is probably still to low to bring out both mobile and desktop products.