AMD 12 core 24 thread Zen2 engineering sample benchmark leaked

[RobbinM]

Source of the source of the source

 

APISAK tweeted he found a userbenchmark run of an engineering sample of a 12-core Zen 3 chip

 

The H2 code in the name of the sample points towards a 'Matisse' chip, AMD's code name for Zen 2

The motherboard is the yet unknown AMD Myrtle MTS.

It's base clock is 3.4 GHz and it turbo's to 3.6 GHz

The sample has a TDP of 105w and has 32MB of L3 cache.

It's presence should point towards AMD indeed using multiple chiplets.

 

Because it is a sample, all parameters can differ on the end product so take speeds with a grain of salt. It does however confirm AMD is testing with more than 8 cores, and that should be a good thing.

It is said to support PCI-e 4.0

 

[Jurrunio]

It's Zen 2, not Zen 3

[Fnige]

 

[r2724r16]

12 minutes ago, RobbinM said:

It's base clock is 3.4 GHz and it turbo's to 3.6 GHz

The sample has a TDP of 105w and has 32MB of L3 cache.

Rumours said that the R7 3700X would run at 4.2 - 5.0 GHz. This leak is kinda disappointing tbh.

 

Not that I expected a 12c/24t 5GHz CPU for $300 to actually come to the market anytime soon.

[Curufinwe_wins]

How do you know it isnt just a TR or TR2 ES?

[RobbinM]

On 1/24/2019 at 5:22 PM, Jurrunio said:

It's Zen 2, not Zen 3

You are absolutely right. Confused by Zen 2/ 3000 naming scheme.

 

On 1/24/2019 at 5:24 PM, Curufinwe_wins said:

How do you know it isnt just a TR or TR2 ES?

Socket AM4

[Lurick]

But....but.....my 5GHz boost

*eye roll*

 

If people really thought a 12 core/24 thread chip like that was going to magically jump up to 5GHz boost then I've got a few swamps and bridges to sell. People got their hopes way too high due to one or two news outlets hyping up a bunch of BS and now people are going to turn their anger to AMD directly because leaks by other people are going to somehow be AMD's fault.

 

For the record, I think AMD has done an awesome job with Ryzen and hopefully pushes things further with Ryzen 2, I'm just tried of all this smoke being blown about some magic chip that's going to save us from the darks of Intel. I'll wait for benchmarks but I could see it being that AMD might not even need to hit anywhere near a 5GHz boost to give Intel even more run for their money and I hope they keep pushing because I want them to make Intel sweat hard.

[SkyHound0202]

http://www.moepc.net/content/uploadfile/201811/8d7b1542898939.png

[ravenshrike]

RGT's video on it.

 

 

 

The really interesting thing is the chip is running on single channel memory. As for clock speed, it's an engineering sample. When was the last time you saw an engineering sample from AMD 5 months out have the same clock speed as the final product?

[BuckGup]

So basically 8 cores at 5ghz and the rest at 4ghz with improved IPC and power usage. Not bad considering game scaling is garbage and making 8 cores be the new quad core is a nice push. The higher core counts will be for HEDT use and be cheap too. So if don't have thread ripper kind of money or want a drop in upgrade makes sense. Also engineering sample

[porina]

32MB of L3 is kinda bordering on enough for me to worry less about ram bandwidth bottlenecking... question will remain if there is much latency penalty going between chiplets, if so this would be best treated as two halves.

 

Did they reveal how much cache per core there is either in the recent even, or previous Rome event? 32MB could be consistent with 8MB per chiplet, 2MB per core at 8 cores a chiplet??? Just trying to work out if they're going more or less on the cache.

[S w a t s o n]

39 minutes ago, Lurick said:

But....but.....my 5GHz boost

*eye roll*

 

If people really thought a 12 core/24 thread chip like that was going to magically jump up to 5GHz boost then I've got a few swamps and bridges to sell. People got their hopes way too high due to one or two news outlets hyping up a bunch of BS and now people are going to turn their anger to AMD directly because leaks by other people are going to somehow be AMD's fault.

 

For the record, I think AMD has done an awesome job with Ryzen and hopefully pushes things further with Ryzen 2, I'm just tried of all this smoke being blown about some magic chip that's going to save us from the darks of Intel. I'll wait for benchmarks but I could see it being that AMD might not even need to hit anywhere near a 5GHz boost to give Intel even more run for their money and I hope they keep pushing because I want them to make Intel sweat hard.

head room for sure on these numbers, you think the chips will really only do 3.6ghz?

[Taf the Ghost]

On 1/24/2019 at 6:05 PM, porina said:

32MB of L3 is kinda bordering on enough for me to worry less about ram bandwidth bottlenecking... question will remain if there is much latency penalty going between chiplets, if so this would be best treated as two halves.

 

Did they reveal how much cache per core there is either in the recent even, or previous Rome event? 32MB could be consistent with 8MB per chiplet, 2MB per core at 8 cores a chiplet??? Just trying to work out if they're going more or less on the cache.

It's 16 mb per CCX, which is still double of Zen1. Still no explicit confirmation that it's 4 cores per CCX, but it seems that hasn't been changed for Zen2. (Probably a Zen3 move, but we'll see.)

 

On 1/24/2019 at 6:08 PM, S w a t s o n said:

head room for sure on these numbers, you think the chips will really only do 3.6ghz?

It's a 2nd stage engineering sample at a motherboard company. If it's 3.4 Base and 3.7 Boost, it means there's a lot of room to move. The first Rome samples were clocking at 1 Ghz. Slightly more concerning is the latencies. First run at the numbers has some thoughts there's a 4-6 ns penalty for crossing the dies, but that'll take extensive testing to know. It's also possible that the IMC will come with 3200 memory as baseline, giving room to drive down some of the latency penalty that way.

[Jurrunio]

5 minutes ago, S w a t s o n said:

head room for sure on these numbers, you think the chips will really only do 3.6ghz?

Aye, not.like.AMD cant screw up

[S w a t s o n]

On 1/24/2019 at 6:14 PM, Jurrunio said:

Aye, not.like.AMD cant screw up

I can't wait for the laughing ristas memes when AMD drops the 16 core halo sku

 

On 1/24/2019 at 6:16 PM, Taf the Ghost said:

It's a 2nd stage engineering sample at a motherboard company. If it's 3.4 Base and 3.7 Boost, it means there's a lot of room to move. The first Rome samples were clocking at 1 Ghz. Slightly more concerning is the latencies. First run at the numbers has some thoughts there's a 4-6 ns penalty for crossing the dies, but that'll take extensive testing to know. It's also possible that the IMC will come with 3200 memory as baseline, giving room to drive down some of the latency penalty that way.

I'm sure windows scheduler will be updated again (even though they just said no optimization needed)

[Taf the Ghost]

On 1/24/2019 at 6:16 PM, S w a t s o n said:

I can't wait for the laughing ristas memes when AMD drops the 16 core halo sku

With the second sighting of a 12c model, we know there are 16c models coming. What we don't know is if AMD will run them out in the first go with Zen2. They could sandbag some SKUs like they did with Threadripper.

 

We have to remember that AMD is about to launch a GPU that they axed then put back on the schedule. Halo SKUs depend on a lot of factors, though I fully expect them to drop special editions for AMD's 50th anniversary. 

 

On 1/24/2019 at 6:17 PM, S w a t s o n said:

I'm sure windows scheduler will be updated again (even though they just said no optimization needed)

Program optimization is different than kicking Windows' Scheduler in the shins a few times. After the WX-model work around, AMD said they have an open ticket with Microsoft over the issue. It'll get addressed, it just depends how long. Linux won't have an issue with it.

[Salv8 (sam)]

1 hour ago, SlimyPython said:

 

0.5 speed is the best for this meme

[cj09beira]

2 hours ago, Geography said:

Rumours said that the R7 3700X would run at 4.2 - 5.0 GHz. This leak is kinda disappointing tbh.

 

Not that I expected a 12c/24t 5GHz CPU for $300 to actually come to the market anytime soon.

soon enough we will have a pretty good idea of how good the node is by looking at vega's frequency range, hopefully someone makes a Fmax/Vmin curve

[Suika]

4 hours ago, Geography said:

Rumours said that the R7 3700X would run at 4.2 - 5.0 GHz. This leak is kinda disappointing tbh.

 

Not that I expected a 12c/24t 5GHz CPU for $300 to actually come to the market anytime soon.

Some of y'all are forgetting that this is an engineering sample and it shows.

 

Numbers aren't final, we may still have higher stock frequencies. The only thing of value to take from this leak is that AMD may very well be testing dual Zen 2 dies on the AM4 platform.

[Gilgam3s1]

This vid by redgamingtech actualy classifies a lot about this leak and some of the hard where they used, like how in the test for this sample they only 0ne 4gb stick of ram at like 2666mhz. He also goes on to say while in the current state the single core score isnt all that impressive it does take a rather large leap over the 2700x current multi core performance. But that being said amd knows higher core speed is somewhat expected so id lean more to this as just a test of this chip just to make sure everything works as ot should rather then anywhere close to a final product

[Rauten]

TR2 2920X is a Zen+ 12/24 at 3.5Ghz base and 4.3Ghz boost

So logic dictates that with the 7nm process and the Zen2 improvements, those speeds should be quite far from final. It maybe not reach the leaked 4.2/5.0 speeds, but they should be higher than both this benchmark and the 2920x.

[Quadriplegic]

Engineering sample

[Stefan Payne]

On 1/24/2019 at 5:17 PM, RobbinM said:

It's base clock is 3.4 GHz and it turbo's to 3.6 GHz

That looks kinda low.

But it shows that AMD is going full on attack.

 

Lets hope that they are sucessful and can make a bit of money and strenthen their position...

 

On 1/24/2019 at 9:33 PM, Suika said:

Some of y'all are forgetting that this is an engineering sample and it shows.

Yeah and probably the lower wattage non X version as well.

Or maybe even a 65W Version for the "low clock chips"...

 

Te 3700 should be 3,8GHz base +800MHz Boost. (wich seems to be the Standard - 800MHz boost throughout the Mantisse range).

[CarlBar]

2 hours ago, Stefan Payne said:

Yeah and probably the lower wattage non X version as well.

Or maybe even a 65W Version for the "low clock chips"...

 

Te 3700 should be 3,8GHz base +800MHz Boost. (wich seems to be the Standard - 800MHz boost throughout the Mantisse range).

 

 

For that matter we got some info on the chip AMD showed off at CES and it had higher clocks than this. Given it uses the same socket, IO ie, and CPU CHiplets there's no reason whatsoever they couldn't produce a much more highly clocked version if they wanted to.But it's obvious where looking at somewhat early silicon being put through very basic tests due to the memory configuration used.

[porina]

8 hours ago, Taf the Ghost said:

It's 16 mb per CCX, which is still double of Zen1. Still no explicit confirmation that it's 4 cores per CCX, but it seems that hasn't been changed for Zen2. (Probably a Zen3 move, but we'll see.)

I made a typo in that post, I meant 16MB per chiplet of 8 core, which is 2MB/core, or about same as current. I'm making the assumption that they are only using 6 out of 8 potential cores but not disabling cache. The other possibility is they are doing 2.66MB/core which would be an odd number.