Speculation: Ryzen 3000 series

[bryanW1995]

I havent built a new rig in 5 years, so I'm really out of the loop here...but how the heck did AMD catch up to intel, and even beat them to 7 nm??

 

[Ottonomous]

I havent built a new rig in 5 years, so I'm really out of the loop here...but how the heck did AMD catch up to intel, and even beat them to 7 nm??

Proper partitioning of the core (no shared components), beefier core components, design for broad power requirements and Intel messing up 10nm.

 

[CHADBOGA]

I havent built a new rig in 5 years, so I'm really out of the loop here...but how the heck did AMD catch up to intel, and even beat them to 7 nm??

AMD is using TSMC's "7nm" process, which will be much like Intel's 10nm process, if Intel ever gets it to yield.

Intel's Manufacturing Group(the Fabs) have done an appalling job on delivering a workable 10nm process and are 2 or more years behind their own schedule, thus causing Intel to use reworked versions of their 14nm process, for their mainstream desktop chips.

People seem to believe that Intel's CEO was fired for this stuff up, but I can't help but feel that not enough heads have rolled.

 

[Kedas]

Looks like AMD is also really going to have a best gaming CPU: (3600 faster than 8700 in game)
https://translate.google.com/translate?hl=nl&sl=auto&tl=en&u=https://www.chiphell.com/thread-2001351-1-1.html
Even the the 2600x or 2700x had trouble to get equal level with the 6 core 8700 for games,

It didn't make much of a difference while playing but there is a difference.
I wonder if the statement "if you want the fastest gaming PC buy an intel CPU" will hold.
I don't think so, it will be much game dependent who wins the test.

And the big question will be if 3800x is the best gaming CPU then how much fps do we lose with 12 cores, or maybe we can disable one die making it a faster 6 core. (like gamer mode)
In best case we don't lose anything and no need for a gamer mode.

 

[RTX2080]

Looks like AMD is also really going to have a best gaming CPU: (3600 faster than 8700 in game)
https://translate.google.com/translate?hl=nl&sl=auto&tl=en&u=https://www.chiphell.com/thread-2001351-1-1.html
Even the the 2600x or 2700x had trouble to get equal level with the 6 core 8700 for games,

The game he tested for comparing 3600&8700 is PUBG (googlish 'survive' means PUBG). Although just 1 game but still promising.
FPS 183 vs 172
Average FPS 6.3% higher
Highest FPS 10% higher

 

[exquisitechar]

New GB4 for Ryzen 5 3600 -
http://browser.geekbench.com/v4/cpu/13334011
This one has RAM running at 2666 MHz -
http://browser.geekbench.com/v4/cpu/13330826

According to Anandtech(https://www.anandtech.com/bench/CPU-2019/2203), ST score is right up there with Skylake.

Big improvements here, excluding memory and AES. Also, there is no regression in the HDR test like the previous samples had, as I thought.

 

[TheELF]

The game he tested for comparing 3600&8700 is PUBG (googlish 'survive' means PUBG). Although just 1 game but still promising.
FPS 183 vs 172
Average FPS 6.3% higher
Highest FPS 10% higher

He also says no overclocking and the 3600 is clocked 12% higher than the 8700 for base...
Without knowing at what clocks they both where running this could be very impressive or very disappointing.

 

[Abwx]

He also says no overclocking and the 3600 is clocked 12% higher than the 8700 for base...
Without knowing at what clocks they both where running this could be very impressive or very disappointing.

8700 is clocked exactly the same as a 8700K, 4.3GHz all cores and 4.7 single core, so actualy it s the 8700 that is clocked higher than the 3600.

https://www.computerbase.de/2017-12/intel-core-i7-8700-turbo-takt-oem-pc/

Same Cinebench score as a 8700K, and way more TDP than the ridiculous 65W stated Intel...

 

[Kedas]

He also says no overclocking and the 3600 is clocked 12% higher than the 8700 for base...
Without knowing at what clocks they both where running this could be very impressive or very disappointing.

Does there exist a Ryzen 2000 that can beat the 8700 in PUBG? (187 is not a low FPS)
Not that is will make a big difference while gaming but intel won't be able to claim best 6 core gaming CPU.

 

[moinmoin]

Zen2 apparently supports JEDEC DDR4-3200 speeds. Does anyone know if this only applies to the X570 chipset or if this will be available on X470 boards as well? (And is there a release date for unbuffered ECC DDR4-3200 DIMMs?)

The chipsets have nothing to do with RAM, the IMC is within the CPU. It will be up to the board manufacturers' BIOS support (possibly limited by the board layout) what range of speeds are officially supported respectively.

I havent built a new rig in 5 years, so I'm really out of the loop here...but how the heck did AMD catch up to intel, and even beat them to 7 nm??

AMD jumped onto the shoulder of giants that are pure play foundries primarily working for mobile devices (e.g. first major user of TSMC's 7nm was Apple with its millions of iPhones). Intel meanwhile choked on 10nm after biting off more than they can chew.

 

[RTX2080]

He also says no overclocking and the 3600 is clocked 12% higher than the 8700 for base...
Without knowing at what clocks they both where running this could be very impressive or very disappointing.

The author didn't say he OCed the 3600. And 8700 has higher turbo than 3600.
I'm afraid there's not much room to be disappoint.

 

[zinfamous]

I havent built a new rig in 5 years, so I'm really out of the loop here...but how the heck did AMD catch up to intel, and even beat them to 7 nm??

Lisa Su + Jim Keller + many many cockups by Intel over the last 5 or 6 years.

 

[PotatoWithEarsOnSide]

Base is pretty irrelevant.
No CPUs clock down to base when under load even after they've hit the turbo duration. They clock down to a 24/7 within TDP clock, typically a good few hundred MHz above base.

 

[Atari2600]

Base is pretty irrelevant.
No CPUs clock down to base when under load even after they've hit the turbo duration. They clock down to a 24/7 within TDP clock, typically a good few hundred MHz above base.

Unless you are talking about some of Intels CPUs running AVX512.

 

[davidbepo]

about the aida latency thing:

https://www.reddit.com/r/Amd/comments/bv6lf7

 

[Fanatical Meat]

Lisa Su + Jim Keller + many many cockups by Intel over the last 5 or 6 years.

I suspect next year intel will figure out the 10nm thing by simply throwing money at it, then they’ll release something huge that their “Ugandan team” was working on but no one thought it would work.
Simply put it will be something big, something ground breaking and something they could have done years ago but chose not to spend money doing it.

 

[Arkaign]

I suspect next year intel will figure out the 10nm thing by simply throwing money at it, then they’ll release something huge that their “Ugandan team” was working on but no one thought it would work.
Simply put it will be something big, something ground breaking and something they could have done years ago but chose not to spend money doing it.

Yep. It became pretty clear that Intel started sandbagging after Sandy Bridge. Almost all their efforts were in CPU status quo with extremely iterative improvement at best, with much more focus on all kinds of other things. IGP, SSD, etc, they're a bit all over the place.

Look at the 9700/9900 die. The Intel HD IGP takes up almost precisely the space of 4 full cores, yet is probably used by a fraction of buyers of that SKU. This is why I never agreed with the idea of putting it in every single desktop SKU like they did, it's wasteful at best.

A '9970k' would be possible with current design with 12C/24T with no IGP.

It was even more heinous with 2600k/3770k/4770k, where it was basically half the die size just for unnecessary IGP.

AMD on the other hand saw clearly that they could easily exploit that weakness by skipping the GPU on non-APUs to make the most of the potential with their CPU designs, and it paid off.

 

[moinmoin]

I suspect next year intel will figure out the 10nm thing by simply throwing money at it

You are quite optimistic. Do you remember process node roadmaps like this one?

Since that original planned date Intel is "simply throwing money at it" and look where that got them to.

Edit: Looking at https://en.wikichip.org/wiki/intel/process Intel continuously had a new smaller process node every two years between 1987 (1.0 µm) and 2011 (22nm)! The delay of 14nm to 2014 already spelled danger.

 

[maddie]

Yep. It became pretty clear that Intel started sandbagging after Sandy Bridge. Almost all their efforts were in CPU status quo with extremely iterative improvement at best, with much more focus on all kinds of other things. IGP, SSD, etc, they're a bit all over the place.

Look at the 9700/9900 die. The Intel HD IGP takes up almost precisely the space of 4 full cores, yet is probably used by a fraction of buyers of that SKU. This is why I never agreed with the idea of putting it in every single desktop SKU like they did, it's wasteful at best.

A '9970k' would be possible with current design with 12C/24T with no IGP.

It was even more heinous with 2600k/3770k/4770k, where it was basically half the die size just for unnecessary IGP.

AMD on the other hand saw clearly that they could easily exploit that weakness by skipping the GPU on non-APUs to make the most of the potential with their CPU designs, and it paid off.

Isn't one of the problems, power usage. More cores = more power relative to the IGP. Look at the Xeons with high clocks. Hungry fellas.

 

[Fanatical Meat]

You are quite optimistic. Do you remember process node roadmaps like this one?

Since that original planned date Intel is "simply throwing money at it" and look where that got them to.

Edit: Looking at https://en.wikichip.org/wiki/intel/process Intel continuously had a new smaller process node every two years between 1987 (1.0 µm) and 2011 (22nm)! The delay of 14nm to 2014 already spelled danger.

Yes but I suspect that problem can be solved by throwing a s ton of money at it.
There was no incentive post 2012 to spend that amount of money.
Maybe I’m wrong, just my opinion.

 

[Kenmitch]

Yes but I suspect that problem can be solved by throwing a s ton of money at it.
There was no incentive post 2012 to spend that amount of money.
Maybe I’m wrong, just my opinion.

The arrogance of the company is what got them in the predicament in the 1st place. Haven't they already thrown billions of dollars at the issue and haven't come up with a solution to get over their hurdles?

AMD used to be the kid that was wimpy and bullied, got their lunch money taken away before school. They've since worked out, got ripped, learned the martial arts, and are ready to kick some arse! They want their lunch money back!

 

[Mechanical Man]

If money would be all that is needed to fix it, it would have been already fixed. Its just wishful thinking they have solution but dont do it, that makes ZERO sense businesswise.

 

[Arkaign]

Isn't one of the problems, power usage. More cores = more power relative to the IGP. Look at the Xeons with high clocks. Hungry fellas.

Definitely. Obviously having the additional cores would come with a value somewhere above 0 and below 100 in terms of % of draw compared to the IGP being there instead.

However, tiering their product line, they could have easily locked the Z series chipsets to deliver the power necessary to run the non-IGP full scale products, while offering the IGP variants for mobile and for B/H series chipsets.

We can clearly see why they didn't do that, and it's the massively higher margins in their other sockets. Doing 6/8 core and beyond much earlier on 115x would have made a lot of Xeon 1366/2011 SKUs irrelevant much sooner, as well as offer fewer options for them to improve should push come to shove.

Once AMD got Ryzen out and started showing what potential was there for less compromised designs, only then did Intel move past 4C/8T for desktop. If AMD was trying to cram a half baked Radeon onto every Ryzen, they would have had a fraction of their success. But their concepts and execution have just been superior to Intel's, who have been the poster child of trudging, unfocused giant.

 

[Shivansps]

Yes, they have doubled the cache, which should also cut memory accesses significantly. Previously the L3 caches couldn't really share data between the two CCX. Geekbench seems to indicate (for a while now) that there are still 2 separate instances, but if they can snoop data from each other with low latency, we could be talking about up to 4x the effective L3 cache size.

Regardless, even 2x cache increase will diminish the need for memory accesses significantly

Even with a big cache there will be misses, we need to look at what happens in those cases.

 

[Fanatical Meat]

If money would be all that is needed to fix it, it would have been already fixed. Its just wishful thinking they have solution but dont do it, that makes ZERO sense businesswise.

No because investors and profit margins and such
Point is now intel has an incentive to produce something good. This is regular market stuff.