AMD 7nm Zen 2 CPUs Sampling This Year for 2019 Volume Launch

Megalith

Megalith

24-bit/48kHz
Staff member
Joined
Aug 20, 2006
Messages
13,000
Comments made by Lisa Su during a recent earnings call suggest that 7nm Zen 2 CPUs will not fall victim to a paper launch and see wide availability next year, as sampling has already begun. AMD has tasked both TSMC and GLOBALFOUNDRIES for their 7nm lineup, which will include samples of a 7nm GPU based on Vega and 7nm server CPU later this year.

"We have a 7nm GPU based on Vega that we'll sample later this year. We have a 7nm server CPU that we'll sample later this year. And then, obviously, we have a number of products that are planned for 2019 as well. So it's a very, very busy product season for us. But we're pleased with the sort of the execution on the product roadmap," Dr. Su said.
 
Good to see them keeping up the momentum with their products. If the new 7nm process gets the vega's power draw down to reasonable levels, that will be good news for AMD and users.
 
AMD shouldn't have any problem with 7nm even if glofo fucks up, with their multi-die chips, they can be the first to launch new node products, and fairly high performance for cpu and pro gpu
these multi die chips were a good move from AMD, Intel and Nvidia will end up copying them for sure, and we will see how good the intel glue will be...
 
Will this 7nm vega successor actually be available to purchase? Cause original Vega has been an embarrassing massive paper launch with it only ever being in stock and available at msrp for one week. One single week.
 
I really hope for AMDs sake, that this new architecture is not some die-shrink BS with bugfixed firmware (Zen+), and also that it offers considerably more than a 6-10% IPC performance uplift per core.

Intel are coming with a new architecture, and you can bet your ass it will be a good one.
 
Snowdensjacket said:
Will this 7nm vega successor actually be available to purchase? Cause original Vega has been an embarrassing massive paper launch with it only ever being in stock and available at msrp for one week. One single week.
Click to expand...

Interesting question considering this news is about the 7nm Zen 2 cpu, unless you were really just looking for an excuse to take a jab. The 7nm gpus wont be sampling till later, end of year?
 
I really wonder how that 7nm process compares to Intel's 10nm, considering most just stick whatever sticker they can find on their process.
 
Stimpy88 said:
I really hope for AMDs sake, that this new architecture is not some die-shrink BS with bugfixed firmware (Zen+), and also that it offers considerably more than a 6-10% IPC performance uplift per core.

Intel are coming with a new architecture, and you can bet your ass it will be a good one.
Click to expand...

Might be optimistic, but I'm hoping for a 25%-30% increase. They've said before there are clear improvements that can be made. I would be surprised if Jim Keller left them with an architecture that can't evolve and improve significantly - seems like that was the basis for a lot of the design choices.
 
I expect AMD will focus on 2 major things: Reducing latency and increasing clockspeed.

With those two boosts alone, you'll see a huge increase in raw power.

Imagine reducing the inter-CCX latency to below 50ns, and then boosting the clockspeed to 4.5+. You'll have a killer chip, with nothing else changed.

I do, however, expect AMD to squeeze more cores into their mainstream CPUs. If Intel is rumoured to be working on a mainstream i7 8-core CPU, expect AMD will be building a bigger, badder Ryzen to counter it.

I imagine a 12-core 24 thread AM4 chip with 3 CCX modules. Pricing will be in line with Intel's pricing. so if Intel charge more for their 8-core, AMD will charge more for their 12-core.
 
As stated above, is there really a need for 10-30% increase in ipc gains? I am not educated enough into knowing what all these cores and processing power are going to be used for.

Does a 15% increase in performance mean you will replace all your server cpus? Does having 200-500mhz more per core mean you will save 20-30min per day? Or do you want a 30% savings in electrical costs?

For the average consumer I believe none of this matters as long as AMD keeps giving out free PCIexpress lanes and undercuts Intel in pricing. For commercial, research, military, and business use; I have no idea what will cause them to scrap their chips so soon.
 
Saki630 said:
As stated above, is there really a need for 10-30% increase in ipc gains? I am not educated enough into knowing what all these cores and processing power are going to be used for.

Does a 15% increase in performance mean you will replace all your server cpus? Does having 200-500mhz more per core mean you will save 20-30min per day? Or do you want a 30% savings in electrical costs?

For the average consumer I believe none of this matters as long as AMD keeps giving out free PCIexpress lanes and undercuts Intel in pricing. For commercial, research, military, and business use; I have no idea what will cause them to scrap their chips so soon.
Click to expand...

I run solution sets for large stress analysis models almost daily at work. Runs can take from 30 minutes to several hours. Even a 10% reduction in runtime can mean 1 or 2 extra runs the same day. That's more than enough reason for my boss to throw a few grand down for a faster machine. I am one among dozens who do similar work, and my company is relatively small. For the average consumer, sure the increases are meaningless, but there are lots of CPU-hungry tasks still and every bit helps.
 
Saki630 said:
As stated above, is there really a need for 10-30% increase in ipc gains?
Click to expand...

In short: No.

But this is [H].

Generally speaking, IPC helps for stuff that isn't easily parallelizable, which tends to include much of what games do; outside of those types of workloads, it's just more processing throughput and other factors can be equally or more important.
 
Saki630 said:
As stated above, is there really a need for 10-30% increase in ipc gains? I am not educated enough into knowing what all these cores and processing power are going to be used for.

Does a 15% increase in performance mean you will replace all your server cpus? Does having 200-500mhz more per core mean you will save 20-30min per day? Or do you want a 30% savings in electrical costs?

For the average consumer I believe none of this matters as long as AMD keeps giving out free PCIexpress lanes and undercuts Intel in pricing. For commercial, research, military, and business use; I have no idea what will cause them to scrap their chips so soon.
Click to expand...


So, data centres don't usually rush out to upgrade ANYTHING, they run it for several years (3-5 years) and then sniff out the market for new stuff. Datacentres may have fleets of sever racks that they update periodically, but because they have so many fleets, they are ALWAYS looking for new servers. So its not about having hardware that tempts people to flock out and upgrade, its about having the better deal and the most appealing hardware ALL the time, to catch as many upgrade schedules as possible.
 
Snowdensjacket said:
Will this 7nm vega successor actually be available to purchase? Cause original Vega has been an embarrassing massive paper launch with it only ever being in stock and available at msrp for one week. One single week.
Click to expand...
Was at MC in Baltimore yesterday and they several Vega 64 for sale. They even got a instant rebate of $300! So they were only $850.
 
Snowdensjacket said:
Will this 7nm vega successor actually be available to purchase? Cause original Vega has been an embarrassing massive paper launch with it only ever being in stock and available at msrp for one week. One single week.
Click to expand...

Big_Rig_Stig said:
Not their fault, blame the miners.
Click to expand...
It wasn't a paper launch, it was available that first week, only because it took that long for miners to figure out the hash rates and such, so that the they will order planes full of them.. btw, i think under Lisa's direction AMD might be past paper launches cpu or gpu
 
Saki630 said:
As stated above, is there really a need for 10-30% increase in ipc gains? I am not educated enough into knowing what all these cores and processing power are going to be used for.

Does a 15% increase in performance mean you will replace all your server cpus? Does having 200-500mhz more per core mean you will save 20-30min per day? Or do you want a 30% savings in electrical costs?

For the average consumer I believe none of this matters as long as AMD keeps giving out free PCIexpress lanes and undercuts Intel in pricing. For commercial, research, military, and business use; I have no idea what will cause them to scrap their chips so soon.
Click to expand...
It seems like power savings would be a bigger selling point for non-enthusiasts than IPC gains. Most casual users already consider computers "fast enough" don't they?

I'm not averse to a bit of both. Maybe I'll actually be able to afford a GPU and memory by the time Zen 2 comes out...
 
vegeta535 said:
Was at MC in Baltimore yesterday and they several Vega 64 for sale. They even got a instant rebate of $300! So they were only $850.
Click to expand...

Man I miss living within a driving 3 hour driving distance of MC...I have spent sooooo much money with then over the last 15 years. Even with sales tax their PM policy and the fact I take my stuff home right then made me basically leave NE and to a lesser extend Amazon behind.

On topic, I am really looking forward to Zen 2. I just grabbed a 2700 to replace my golden 1600 @ 4.1 but haven't had time to see what I can squeeze out of it with 840mm of rad to help cool it. If Zen 2 offers 8/12 cores north of 4.5Ghz (would really perfer to see 4.7Ghz+ if IPC does not come up a ton) then I will jump yet again.

I am rewarding AMD with every purchase I can. I've always gone with the faster option in the past, but I am tired of the BS Intel and Nvidia have been pulling for waaaay too long. This is why I stayed on a 5Ghz 3770K for so long. AMD has been extremely good to me profit wise since my days of running 3 290Xs under water and making a ton of money. That money paid for a big chunk of my house and my current upgrades and 2 vacations!
 
  • Like
Reactions: N4CR
like this
deton8 said:
It seems like power savings would be a bigger selling point for non-enthusiasts than IPC gains.
Click to expand...

Intel agrees! This is where their development budget has been going to lately, and they've shown significant gains despite being handicapped by the pace of process improvements.
 
Stimpy88 said:
I really hope for AMDs sake, that this new architecture is not some die-shrink BS with bugfixed firmware (Zen+), and also that it offers considerably more than a 6-10% IPC performance uplift per core.

Intel are coming with a new architecture, and you can bet your ass it will be a good one.
Click to expand...
Like that P4? Intel makes mistakes too. Of course, when they do, they illegally make sure they maintain their marketing position.
When was the last time a followup gen of intel's reached 10%? Your evaluation of what is "BS" seems rather bigoted.
 
ChefJeff789 said:
I run solution sets for large stress analysis models almost daily at work. Runs can take from 30 minutes to several hours. Even a 10% reduction in runtime can mean 1 or 2 extra runs the same day. That's more than enough reason for my boss to throw a few grand down for a faster machine. I am one among dozens who do similar work, and my company is relatively small. For the average consumer, sure the increases are meaningless, but there are lots of CPU-hungry tasks still and every bit helps.
Click to expand...

I work with network video recorders and analytics. More cores, more speed, etc, is always better. We could do significantly more advanced processing. We still have to downsample some things to 320x240 to get reasonable speed or process at 2 fps.
 
My purely unscientific opinion about Zen2 - before juanrga shows up.

- Small IPC uplift. ~5%. Near-parity with Skylake IPC, but a pubic hair less - mainly due to edge cases. TheStilt said non-256b workload IPC difference between Zen+ and Skylake is 8%. I expect this to be reduced to ~3%. I also expect that this will be basically unnoticeable in day-to-day games and apps, on average.

- Clockspeed gain of ~400MHz over Zen+ at equivalent core counts. ~4.7-4.8GHz halo product clocks, after OC.

- And I wonder if we will see 6 core CCXs with Zen 2. I've seen some rumors and supposed leaks that suggest this. Buuuuuuuut lol. 90% of everything is bullshit, and we're so far out speculating is probably lol. I'll go with a big "maybe" on more cores per CCX. If this is true, I expect clockspeed improvement to be less impressive for such a product. Maybe more like 4.4-4.5GHz top OC instead of 4.7-4.8.

- Power consumption of X products will still suck (especially if more cores, I bet, while power consumption of non-X parts will be great. While motherboard/socket compatibility will remain, I expect shitty boards to have power delivery issues, regardless of more cores and/or more clocks.

Note, I think a 6 core CCX is much more likely than 3 CCXs as opposed to 2. There appears to be an architectural reason why CCX core counts must match, and I suspect that this may also cause trouble with odd numbers of multiple CCXs (1 CCX being the exception, of course - see Ryzen APUs).
 
  • Like
Reactions: N4CR
like this
DuronBurgerMan said:
My purely unscientific opinion about Zen2 - before juanrga shows up.

- Small IPC uplift. ~5%. Near-parity with Skylake IPC, but a pubic hair less - mainly due to edge cases. TheStilt said non-256b workload IPC difference between Zen+ and Skylake is 8%. I expect this to be reduced to ~3%. I also expect that this will be basically unnoticeable in day-to-day games and apps, on average.

- Clockspeed gain of ~400MHz over Zen+ at equivalent core counts. ~4.7-4.8GHz halo product clocks, after OC.

- And I wonder if we will see 6 core CCXs with Zen 2. I've seen some rumors and supposed leaks that suggest this. Buuuuuuuut lol. 90% of everything is bullshit, and we're so far out speculating is probably lol. I'll go with a big "maybe" on more cores per CCX. If this is true, I expect clockspeed improvement to be less impressive for such a product. Maybe more like 4.4-4.5GHz top OC instead of 4.7-4.8.

- Power consumption of X products will still suck (especially if more cores, I bet, while power consumption of non-X parts will be great. While motherboard/socket compatibility will remain, I expect shitty boards to have power delivery issues, regardless of more cores and/or more clocks.

Note, I think a 6 core CCX is much more likely than 3 CCXs as opposed to 2. There appears to be an architectural reason why CCX core counts must match, and I suspect that this may also cause trouble with odd numbers of multiple CCXs (1 CCX being the exception, of course - see Ryzen APUs).
Click to expand...
Mmmm 6-core CCX
 
DuronBurgerMan said:
My purely unscientific opinion about Zen2 - before juanrga shows up.

- Small IPC uplift. ~5%. Near-parity with Skylake IPC, but a pubic hair less - mainly due to edge cases. TheStilt said non-256b workload IPC difference between Zen+ and Skylake is 8%. I expect this to be reduced to ~3%. I also expect that this will be basically unnoticeable in day-to-day games and apps, on average.

- Clockspeed gain of ~400MHz over Zen+ at equivalent core counts. ~4.7-4.8GHz halo product clocks, after OC.

- And I wonder if we will see 6 core CCXs with Zen 2. I've seen some rumors and supposed leaks that suggest this. Buuuuuuuut lol. 90% of everything is bullshit, and we're so far out speculating is probably lol. I'll go with a big "maybe" on more cores per CCX. If this is true, I expect clockspeed improvement to be less impressive for such a product. Maybe more like 4.4-4.5GHz top OC instead of 4.7-4.8.

- Power consumption of X products will still suck (especially if more cores, I bet, while power consumption of non-X parts will be great. While motherboard/socket compatibility will remain, I expect shitty boards to have power delivery issues, regardless of more cores and/or more clocks.

Note, I think a 6 core CCX is much more likely than 3 CCXs as opposed to 2. There appears to be an architectural reason why CCX core counts must match, and I suspect that this may also cause trouble with odd numbers of multiple CCXs (1 CCX being the exception, of course - see Ryzen APUs).
Click to expand...
BTW, I think your guestimates sound right.. what I have been wondering about is, if there would be somehow some new x86 instruction sets down the pipe by AMD... maybe to optimize multi-core performance?
 
Snowdensjacket said:
Will this 7nm vega successor actually be available to purchase? Cause original Vega has been an embarrassing massive paper launch with it only ever being in stock and available at msrp for one week. One single week.
Click to expand...
Not in every country.
They were only OOS for a few weeks in NZ. Always been available and always around Ti price since recent boom/because of mining. For a month or so they were well under Ti price and a worthy contender.
 
DuronBurgerMan said:
My purely unscientific opinion about Zen2 - before juanrga shows up.

- Small IPC uplift. ~5%. Near-parity with Skylake IPC, but a pubic hair less - mainly due to edge cases. TheStilt said non-256b workload IPC difference between Zen+ and Skylake is 8%. I expect this to be reduced to ~3%. I also expect that this will be basically unnoticeable in day-to-day games and apps, on average.

- Clockspeed gain of ~400MHz over Zen+ at equivalent core counts. ~4.7-4.8GHz halo product clocks, after OC.

- And I wonder if we will see 6 core CCXs with Zen 2. I've seen some rumors and supposed leaks that suggest this. Buuuuuuuut lol. 90% of everything is bullshit, and we're so far out speculating is probably lol. I'll go with a big "maybe" on more cores per CCX. If this is true, I expect clockspeed improvement to be less impressive for such a product. Maybe more like 4.4-4.5GHz top OC instead of 4.7-4.8.

- Power consumption of X products will still suck (especially if more cores, I bet, while power consumption of non-X parts will be great. While motherboard/socket compatibility will remain, I expect shitty boards to have power delivery issues, regardless of more cores and/or more clocks.

Note, I think a 6 core CCX is much more likely than 3 CCXs as opposed to 2. There appears to be an architectural reason why CCX core counts must match, and I suspect that this may also cause trouble with odd numbers of multiple CCXs (1 CCX being the exception, of course - see Ryzen APUs).
Click to expand...

Looking at their recent work tweaking cache latency, I am hoping we see closer to a 8-10% IPC bump, not including the 300-500Mhz clock speed gain (at the same TDP of Zen+ designs or slightly higher). They have had a decent amount of time to tweak since they started on Zen2 concurrently with Zen.

I believe we could see the 10% bump but time will tell. Zen+ is a define step in the right direction. As long as they keep doing this then parity with Intel will be fully obtained, maybe even surpassing Intel a bit.
 
You must log in or register to reply here.
Back
Top