(16core added)AMD 3000 specs! 4.7 GHZ, R9 3950x, R7 3700x, 3800x.

[Trixanity]

27 minutes ago, GoldenLag said:

oh, yeah that is true. zen+ could be an outlier for 12nm as AMD probably didnt want to spend any time on it since it was never intended to be much of an improvement. 

12nm does have an area reduction. AMD just didn't want to use the resources required to shrink the die or to add transistors. So they just adapted the design for more space between transistors. Minimal effort.

[GoldenLag]

6 minutes ago, Trixanity said:

12nm does have an area reduction. AMD just didn't want to use the resources required to shrink the die or to add transistors. So they just adapted the design for more space between transistors. Minimal effort.

Thats what i said in that response.........

 

And 12nm was a node meant to just be copypasted from 14nm

[RejZoR]

52 minutes ago, Trixanity said:

12nm does have an area reduction. AMD just didn't want to use the resources required to shrink the die or to add transistors. So they just adapted the design for more space between transistors. Minimal effort.

I'm assuming having more space between transistors improves heat conduction, electrical leakage and improves yields on the waffers. Chip is not any smaller and you can't make any more of them from a single waffer, but you get more usable ones that can possibly clock a bit higher. If I'm understanding this right.

[Stefan Payne]

4 hours ago, leadeater said:

Would probably not be able to due to the interposer/package and the traces required for each die, no actual idea but my gut feel is too little space.

Well, until then, the CPU Die looks a bit different...

Because they will either be in "6nm" or even "5nm", wich is already in the Works.

The Differences are smaller than 14nm -> 7nm but they are still there.

But then again, the real question is: Does the I/O Die have 3 IF Interfaces or does it top out at two?

 

That's the real question...

With DDR-5 SDRAM, maybe possibly. Because they need a new I/O Die.

But with Ryzen 4000 I bet they'd still be using the same I/O Die as Matisse.

4 hours ago, leadeater said:

I'd rather see that, along with die stacking with HBM. That kind of dream really is not free though lol.

Agree.
Even the new, rumored "Low cost HBM" with only 512bit with would be better than nothing.

Though it goes back to Sidechannel of the good old AMD 78xG and 790GX that was used in like 2 or 3 Boards (I have one of those)...

But with HBCC it can be hidden from the software that the GPU has some "Buffer Memory"...

4 hours ago, leadeater said:

The functions of that die are better suited to 14nm which is why 7nm is not being used. Not everything responds well or even positively to node shrinks. 12nm (14nm+) is an option I believe though.

Yeah, because of the Power/Current Requirements of the PHY; you can't shrink them much these days.

HBM I believe still is shrinkable because that is basically an on chip connection that requires very little power compared to other interfaces that go off the chip or, worse, over a connector.

That's the main reason we don't see no upgradable GPU Memory. Because its just not possible/viable...

 

The Question of 12nm vs. 14nm is a question of Cost.

When 12nm becomes a bit cheaper than 14nm, they will switch to it, if it doesn't, they will stick to 14nm.

[CarlBar]

23 hours ago, System32.exe said:

Expecting the prices to come down this gen is kinda ridiculous tbh, not only is Zen 2 on 7nm, but they've got a separate I/O chip to deal with now as well.

 

Inflation and basic R&D economics is a bigger reason. The actual silicon cost of the chips is negligible, (another reason price being slaved to core count doesn't work, the production cost does not scale that sharply), most of the money per sale goes towards, (in somewhat simplified terms), paying off the R&D budget. That pretty much means whichever chip they think they're going to sell in the most quantity, (or chips in most quantities), has X price slapped on it. Of course what i said about market expectation is a factor too.

 

And to a very large degree many consumers and OEM's come in more with a specific budget than a specific performance target in mind so the chip that sells the best is going to be heavily influenced by people's average budget, (and also the price of other elements of the overall end product that may or may not be produced by AMD). 

[trevb0t]

Your 3600 core clock reads 3.3

According to the AMD site link, it's actually 3.6

 

(I'm sure it's been pointed out, but just in case.)

[cj09beira]

22 hours ago, Stefan Payne said:

Well, until then, the CPU Die looks a bit different...

Because they will either be in "6nm" or even "5nm", wich is already in the Works.

The Differences are smaller than 14nm -> 7nm but they are still there.

But then again, the real question is: Does the I/O Die have 3 IF Interfaces or does it top out at two?

 

That's the real question...

With DDR-5 SDRAM, maybe possibly. Because they need a new I/O Die.

But with Ryzen 4000 I bet they'd still be using the same I/O Die as Matisse.

Agree.
Even the new, rumored "Low cost HBM" with only 512bit with would be better than nothing.

Though it goes back to Sidechannel of the good old AMD 78xG and 790GX that was used in like 2 or 3 Boards (I have one of those)...

But with HBCC it can be hidden from the software that the GPU has some "Buffer Memory"...

Yeah, because of the Power/Current Requirements of the PHY; you can't shrink them much these days.

HBM I believe still is shrinkable because that is basically an on chip connection that requires very little power compared to other interfaces that go off the chip or, worse, over a connector.

That's the main reason we don't see no upgradable GPU Memory. Because its just not possible/viable...

 

The Question of 12nm vs. 14nm is a question of Cost.

When 12nm becomes a bit cheaper than 14nm, they will switch to it, if it doesn't, they will stick to 14nm.

they could also be looking at 12FDX which is planar but quite efficient and cheap 

i am more looking forward to hbm as a cpu cache than a gpu cache, its just a pitty that finding out what hbm's latency is, is quite hardl, google has failed me

though with it the lattency lather might become a little weird 

[Stefan Payne]

1 hour ago, cj09beira said:

they could also be looking at 12FDX which is planar but quite efficient and cheap 

That was the SOI one, wasn't it?
That might even be the best one due to Insulator.

1 hour ago, cj09beira said:

i am more looking forward to hbm as a cpu cache than a gpu cache, its just a pitty that finding out what hbm's latency is, is quite hardl, google has failed me

though with it the lattency lather might become a little weird 

Do you know someone fluent in Vulkan/OpenGL Programming?
If so, write a little Programm that does measure the Latency of the Memory...

 

But I doubt that there is a big difference in latency. If there is a difference its the Interface/protocoll, the DRAM Cells are the same...

[Ravendarat]

So I have a question I still cant wrap my head around completely. These new x570 boards are going to be more money, some people say a lot more money, but that term is all relative to the person speaking it. Top end boards for x470 are around $300 usd based of PCPartpicker. Now I know its all speculation at this point but are we expecting general equivalents in x570 to be like 50% more money? double? Triple? Im just trying to understand what constitutes a lot more money

[CiBi]

3 minutes ago, Ravendarat said:

So I have a question I still cant wrap my head around completely. These new x570 boards are going to be more money, some people say a lot more money, but that term is all relative to the person speaking it. Top end boards for x470 are around $300 usd based of PCPartpicker. Now I know its all speculation at this point but are we expecting general equivalents in x570 to be like 50% more money? double? Triple? Im just trying to understand what constitutes a lot more money

GamersNexus hinted at a 777USD motherboard from either Gigabyte or MSI I forgot which

[Origami Cactus]

1 hour ago, Ravendarat said:

So I have a question I still cant wrap my head around completely. These new x570 boards are going to be more money, some people say a lot more money, but that term is all relative to the person speaking it. Top end boards for x470 are around $300 usd based of PCPartpicker. Now I know its all speculation at this point but are we expecting general equivalents in x570 to be like 50% more money? double? Triple? Im just trying to understand what constitutes a lot more money

The top end motherboard from ASUS, Gigabyte, MSI, and ASRock are about 500, 600, 600 and 1000 respectivly.  But they are not the equivalents to x470 boards, they are a new class over the existing motherboards. That just shows that the Motherboards manufacturers are so confident in AMD this launch, that they have brought out their flagship motherboards.

 

The equivalents to their x470 boards should be about +50$ more expensive or so.

[Ravendarat]

4 minutes ago, Origami Cactus said:

The top end motherboard from ASUS, Gigabyte, MSI, and ASRock are about 500, 600, 600 and 1000 respectivly.  But they are not the equivalents to x470 boards, they are a new class over the existing motherboards. That just shows that the Motherboards manufacturers are so confident in AMD this launch, that they have brought out their flagship motherboards.

 

The equivalents to their x470 boards should be about +50$ more expensive or so.

Ya when I said equivalent I meant like too end to top end, I understand that the feature set is different. I’m going to go either msi or gigabyte so those prices sound inline to me

[cj09beira]

5 minutes ago, Ravendarat said:

Ya when I said equivalent I meant like too end to top end, I understand that the feature set is different. I’m going to go either msi or gigabyte so those prices sound inline to me

i expect 200 bucks to be the starting point for x570

[KarimEid]

So our best bet right now is to either build an amd based pc or wait for intel to acknowledge pcie gen 4 then build an intel pc. 

[Reytime]

I am very interested in the 3900X. Wonder how it stacks up against Intel CPUs 

[mr moose]

4 hours ago, KarimEid said:

So our best bet right now is to either build an amd based pc or wait for intel to acknowledge pcie gen 4 then build an intel pc. 

 

Best bet right now is to wait for actual benchmarks.  Even though it's looking really good for a ryzen build (personally excited because I'm ready too) we don't know what Intel will do or how long it will be before Gen4 PCIe is necessary (if current PCIe is not a bottle neck for the next 4-5 years then spending money on gen 4 now is kinda wasted). 

[ch3w2oy]

7 hours ago, KarimEid said:

So our best bet right now is to either build an amd based pc or wait for intel to acknowledge pcie gen 4 then build an intel pc. 

There is NO reason to wait for Intel. We're probably about to have the best processors ever, for cheaper than what Intel would charge.. Why is the world would you wait for Intel?

 

If you need a PC, it's about to be the best time to build one.. If you don't need one then sure, wait and see what's available then.. 

[trevb0t]

I watched the presentation, but wanna make sure I didn't miss it; Did AMD give any indication that they are currently working on a Zen2 APU?

[TahoeDust]

Nevermind

[The Benjamins]

40 minutes ago, trevb0t said:

I watched the presentation, but wanna make sure I didn't miss it; Did AMD give any indication that they are currently working on a Zen2 APU?

Not this one, but its been on the road map. keep in mind the APU line has been a arch behind.

[mr moose]

7 hours ago, ch3w2oy said:

There is NO reason to wait for Intel. We're probably about to have the best processors ever, for cheaper than what Intel would charge.. Why is the world would you wait for Intel?

 

If you need a PC, it's about to be the best time to build one.. If you don't need one then sure, wait and see what's available then.. 

Can I have your crystal ball when you're done?

[RejZoR]

Well, he's not wrong. Sunny Cove is still relatively far away and until then, Intel really doesn't have anything else to show than repackaging 9900K for the 10th time...  Unless you're willing to wait for another year or two, there really isn't much to think about. AMD even closed the clock gap with the IPC boost. So, even though they are only hitting 4.6GHz, the CPU's are basically as fast as Intel's 5GHz parts. Minus super specific loads like AVX...

[mr moose]

3 minutes ago, RejZoR said:

Well, he's not wrong. Sunny Cove is still relatively far away and until then, Intel really doesn't have anything else to show than repackaging 9900K for the 10th time...  Unless you're willing to wait for another year or two, there really isn't much to think about. AMD even closed the clock gap with the IPC boost. So, even though they are only hitting 4.6GHz, the CPU's are basically as fast as Intel's 5GHz parts. Minus super specific loads like AVX...

 

Can I have your crystal ball then.  Not sure if anyone's aware yet,  but we don't have any benchmarks, we literally only have marketing to go off at the moment.  Whilst I am pretty confident they will perform as expected (or at least within a bulls spit),  We still don't know enough to be telling people to go ahead and buy AM4 without knowing if PCIe4 is worth it right now or if Ryzen will actually perform sufficiently for whatever task it is that people buy it for.

[RejZoR]

Because AMD lied so much about Zen so far. One thing was being sceptical about first generation of Ryzen after Bulldozer disappointment. After Zen came Zen+ and delivered, I don't see a single reason why Zen2 couldn't or why AMD would have to make shit up to make it more interesting product. This isn't Vega or Navi. This is a real product that has proven itself.

 

As for PCIe... Who's even questioning if it's "worth it" or basing entire purchase around that? If you feel like you don't need it, then you go with X470. If you do, then you do. Either because you need bandwidth of it or because you want to future proof your build.

[mr moose]

Just now, RejZoR said:

Because AMD lied so much about Zen so far. One thing was being sceptical about first generation of Ryzen after Bulldozer disappointment. After Zen came Zen+ and delivered, I don't see a single reason why Zen2 couldn't or why AMD would have to make shit up to make it more interesting product.

So you have a benchmark, you know how it's going to perform?  If your happy to drop $600-700 on it without a benchamark then be my guest.  But I will NEVER recommend anyone buy a CPU until you can see 3rd party benchmarks of it under similar work tasks that you intend to use it for.  This has nothing to do with AMD lying or bulldozer or anything else fo that matter, it is a simple case of not recommending a product until we can see how it performs in independant testing.

 

 

Just now, RejZoR said:

As for PCIe... Who's even questioning if it's "worth it" or basing entire purchase around that? If you feel like you don't need it, then you go with X470. If you do, then you do. Either because you need bandwidth of it or because you want to future proof your build.

 

You haven't read the last few posts have you?

 

this was posted on the previous page:

 

Quote

So our best bet right now is to either build an amd based pc or wait for intel to acknowledge pcie gen 4 then build an intel pc. 

 

So yes, people are questioning and the answer is still the same:

12 hours ago, mr moose said:

 

Best bet right now is to wait for actual benchmarks.  Even though it's looking really good for a ryzen build (personally excited because I'm ready too) we don't know what Intel will do or how long it will be before Gen4 PCIe is necessary (if current PCIe is not a bottle neck for the next 4-5 years then spending money on gen 4 now is kinda wasted).