archsan wrote:

I was conceding my little 'slip of mind'

If that's what you meant to convey, you communicated that poorly. From your post I do not get the impression that "mainstream" was not in fact the word you meant to use, but rather that you meant to stretch up the meaning of "mainstream" to include things that really aren't considered mainstream.
And let's face it, mainstream is a pretty common and well-understood term. It's not very likely to make a mistake like that.

R7 1700 looks to be the star of the show! Granted, XFR has not really been explained properly yet and I fear AMD may be pulling an Intel here. Despite saying all of the CPUs will be unlocked, I think they might actually allow you to only raise the multiplier as far up as the boost clock and only really offer an unlocked multiplier on the XFR enabled models.
Hopefully, I will not be correct on this one.

It is a little unfortunate that they only launched the 8-core parts, the 6-core and 4-core parts should follow suit pretty soon, I am definitely in the market for one of these. I'd love to get a 6-core/12 threads, but it looks like I'll be going with a 4-core, unless the price really is irresistible on that 6-core 😁.

after over 10 years of disappointments, I think AMD is back.
even if they can't compete in single thread with Kaby Lake, Broadwell-e was launched less than 1 year ago and they are very competitive with it, the X99 platform seems dead now because of it.

also the 1700 being a 65W part is pretty impressive from AMD.

ElementalChaos wrote:

Bulldozer is the NetBurst of this decade - hot, inefficient, and outperformed by its predecessor the Phenom II. Expect to see them flooding recycling centers come 2026.

AMD has a history of promising the world and underdelivering. I really hope that's not the case this time. More than anything we need some real competition on the CPU front; if Intel had their way with things the future would be very bleak pricing and performance-improvement wise. They've already been stagnating since Sandy Bridge.

EDIT: From what I've seen of purportedly leaked Ryzen benchmarks, AMD may actually have a hit on their hands. Now they just have to price it properly.

Well, they hit it out of the park with pricing. And every day that passes by this is looking less and less like a fluke. It's looking pretty likely I will be procuring a Ryzen upgrade for my soggy old FX-6300 gaming PC.

Of course Intel still has a chokehold on the laptop/tablet market which is arguably where more of the money is nowadays. But I imagine this will be enough of a hit with gamers to bring AMD out of the red.

It certainly looks like AMD's going to make one hell of a comeback. Seriously, Lisa Su is the bee's knees!
I haven't been this excited about a CPU launch since...Pentium II?

F2bnp wrote:

It is a little unfortunate that they only launched the 8-core parts, the 6-core and 4-core parts should follow suit pretty soon, I am definitely in the market for one of these. I'd love to get a 6-core/12 threads, but it looks like I'll be going with a 4-core, unless the price really is irresistible on that 6-core 😁.

My guess is that they went for the 8-cores because it's the low-hanging fruit: As I already said, Intel's 6-core and 8-core are priced considerably higher than the 2/4-core ones.
So that's where AMD can get easy (temporary) wins on price/performance.
I have a feeling the 4-cores are going to be a lot less interesting.

Are you not happy that Intel has some competition again and might have to price more competitively?

gdjacobs wrote:

Are you not happy that Intel has some competition again and might have to price more competitively?

I'm not in the market for their CPUs, so I don't really care.
Also, I think it's a bit too early... I want to see what kind of competition we have here first.

Scali wrote:

My guess is that they went for the 8-cores because it's the low-hanging fruit: As I already said, Intel's 6-core and 8-core are […]
Show full quote

F2bnp wrote:

It is a little unfortunate that they only launched the 8-core parts, the 6-core and 4-core parts should follow suit pretty soon, I am definitely in the market for one of these. I'd love to get a 6-core/12 threads, but it looks like I'll be going with a 4-core, unless the price really is irresistible on that 6-core 😁.

My guess is that they went for the 8-cores because it's the low-hanging fruit: As I already said, Intel's 6-core and 8-core are priced considerably higher than the 2/4-core ones.
So that's where AMD can get easy (temporary) wins on price/performance.
I have a feeling the 4-cores are going to be a lot less interesting.

the quads actually look interesting, 4c/4t at $129 and 4c/8t at $175... but the 6c/12t one is quite interesting at $229
keep in mind they are all unlocked for OC and can be OCed with not so expensive B350 MBs... they are going to be competing with locked CPUs with less cores/threads at the same price, and with IPC around Haswell/Broadwell, the disadvantage is not huge there...

obviously, those are not even officially announced yet, but judging by the other leaks being spot on, I think it's safe to assume those are the prices...
also we need the reviews to see how it actually performs in a wider range of programs, and how it OCs, but it's all looking great compared to "Bulldozer", and they promised 40% and delivered 52%, so I'm feeling pretty positive about it, quads should be good for gaming.

SPBHM wrote:

also we need the reviews to see how it actually performs in a wider range of programs, and how it OCs, but it's all looking great compared to "Bulldozer", and they promised 40% and delivered 52%, so I'm feeling pretty positive about it, quads should be good for gaming.

Looking great compared to Bulldozer isn't that difficult, especially if you focus entirely on IPC and ignore clockspeeds.
Promising something not-even-that-impressive and then delivering more than your original promise is also not that difficult.

Why do they do this?
I don't recall Intel going all crazy on "Look at our IPC gains!!!!11oneoneeleven!" when they went from Pentium 4 to Core2 Duo, and that was more than 52% gain.
I mean, let's take Cinebench single-core here: http://www.anandtech.com/show/2045/11
Core2 Duo at 2.66 GHz scores 446.
Pentium D at 3.6 GHz scores 305.
So if we normalize that for clockspeed, that's 446/2.66 = 165.67 vs 305/3.6 = 84.72.
165.67 / 84.72 = 1.96. So they could have marketed it like: 96% higher IPC!!!!

They didn't, because they simply let the product and its benchmarks speak for itself.
With AMD I get the distinct impression that they've cherry-picked some benchmarks (and the Intel CPUs they're comparing against), and then used some very 'creative' number juggling to try and make it sound way impressive.
I'm expecting CPUs that are 'almost' as good as Intel's on IPC and consume 'almost' as little power.
Which you can get for 'almost' as much money, if you like. But not exactly CPUs that will take over the world or anything.
I'm not expecting something like the original Athlon, which was actually better than a Pentium III and Pentium 4 in virtually every way. More like an 'also ran'.

To be fair, K8 was a better architecture than P4 as well, and far better for wide shared memory machines. 😀 The IMC was a genuine advancement although AMD didn't invent the concept (props to Sun and IBM).

Waiting for independent benchmarks to be released is certainly the pragmatic approach. The only people who have performance numbers right now are embargoed from publishing anything, so pretty much every comment is rooted in guesswork. As such, why not treat any speculation as speculative and move on. Arguing for or against someone else's hopes isn't very constructive and likely not very fruitful. Plus it's all going to come out in the wash in a few days time.

gdjacobs wrote:

To be fair, K8 was a better architecture than P4 as well

Yes, although I don't really consider it a separate microarchitecture.
To me, it's basically K7-with-IMC-and-64bit-extensions.
Just like I don't see Pentium Pro/II/III as different architectures. They're the same basic architecture with some changes to the 'uncore' and some extensions.
You could argue 'but 64-bit', but then I'd just point to the Pentium 4, which clearly was just a 32-bit CPU at the start, and Intel put the 64-bit extensions in one of its refreshes there as well, without changing a whole lot to the main architecture. So apparently those extensions weren't that big of a deal in a modern x86 which already translates its code to internal microcode and internal register file anyway.

gdjacobs wrote:

The IMC was a genuine advancement although AMD didn't invent the concept (props to Sun and IBM).

The sad part is that they didn't quite get it right. Core2 Duo was easily competitive with a standard FSB and chipset (at least in single-socket systems).
Nehalem got more benefit from the IMC, and together with its HT, that's where Intel really took off from what AMD was doing, and never looked back... until now perhaps.
Likewise, AMD failed to capitalize on the fact that the Athlon X2 had two cores on a single die. Core2 Duo really showed them the way with a shared L2 cache. If you're going to put two or more cores on a single die, that's what you have to do. Otherwise it's no better than two single-core CPUs in separate sockets.

Scali wrote:

The sad part is that they didn't quite get it right. Core2 Duo was easily competitive with a standard FSB and chipset (at least […]
Show full quote

gdjacobs wrote:

The IMC was a genuine advancement although AMD didn't invent the concept (props to Sun and IBM).

The sad part is that they didn't quite get it right. Core2 Duo was easily competitive with a standard FSB and chipset (at least in single-socket systems).
Nehalem got more benefit from the IMC, and together with its HT, that's where Intel really took off from what AMD was doing, and never looked back... until now perhaps.
Likewise, AMD failed to capitalize on the fact that the Athlon X2 had two cores on a single die. Core2 Duo really showed them the way with a shared L2 cache. If you're going to put two or more cores on a single die, that's what you have to do. Otherwise it's no better than two single-core CPUs in separate sockets.

C2D did use some of the tricks they'd mastered with Netburst for latency hiding, but I think the biggest factor mitigating the advantages of the IMC was the relatively huge cache sizes Intel's process tech allowed them. In sum total, though, Core 2 was a really good architecture that got a lot of stuff right.

I think the cache architecture of Athlon 64 X2 had more to do with the fact that the basic design was still single core. AMD needed the two cores to share die space because of the IMC, but they could still leverage the single core design bolted on to the hypertransport and IMC blocks. Once the market was almost exclusively multicore, they did away with the native single core design and fulfilled that section of their market with chip harvests from their quad core lines.

gdjacobs wrote:

C2D did use some of the tricks they'd mastered with Netburst for latency hiding, but I think the biggest factor mitigating the advantages of the IMC was the relatively huge cache sizes Intel's process tech allowed them.

You have to give credit for Intel's cache efficiency as well. They had high associativity and low latency. Just slapping a bunch of transistors on there isn't enough (we've seen that with Phenom, where the added L3 barely had any effect at all, because it was just too slow).

gdjacobs wrote:

I think the cache architecture of Athlon 64 X2 had more to do with the fact that the basic design was still single core. AMD needed the two cores to share die space because of the IMC, but they could still leverage the single core design bolted on to the hypertransport and IMC blocks.

They basically did the same as the Pentium D: they copy-pasted two cores onto a single die (the first Pentium D was single-die, they only moved to MCM later). It just looked a bit more like a 'Siamese twin' than the Pentium D because the IMC was also on the die. But at a technical level it was basically the same as a two-socket machine with two single-core CPUs, where only one CPU had memory installed, so they'd both have to access memory through the same IMC and its hypertransport crossbar interface.

They didn't actually *design* it as a native dualcore (which Intel did with the CoreDuo, and later Core2).
Which isn't that much of a problem, until you try to push your shit as 'native multicore', trying to claim it has some advantages that it doesn't.

I couldn't remember if Yonah was an MCM, but yeah. It was essentially the memory controller, crossbar, and two single core CPUs on a die. Pentium D was a little simpler as it could share the parallel FSB so no additional logic was required.

gdjacobs wrote:

I couldn't remember if Yonah was an MCM, but yeah. It was essentially the memory controller, crossbar, and two single core CPUs on a die. Pentium D was a little simpler as it could share the parallel FSB so no additional logic was required.

There was no additional logic required for the Athlon64 either, because it was already designed as a multi-CPU NUMA solution.

I believe there's no memory locality between the two cores due to the crossbar. Multi CPU setups benefit from NUMA aware memory allocation and task scheduling due to sockets having local and non-local memory pools.

Scali wrote:

Looking great compared to Bulldozer isn't that difficult, especially if you focus entirely on IPC and ignore clockspeeds. Promis […]
Show full quote

SPBHM wrote:

also we need the reviews to see how it actually performs in a wider range of programs, and how it OCs, but it's all looking great compared to "Bulldozer", and they promised 40% and delivered 52%, so I'm feeling pretty positive about it, quads should be good for gaming.

Looking great compared to Bulldozer isn't that difficult, especially if you focus entirely on IPC and ignore clockspeeds.
Promising something not-even-that-impressive and then delivering more than your original promise is also not that difficult.

Why do they do this?
I don't recall Intel going all crazy on "Look at our IPC gains!!!!11oneoneeleven!" when they went from Pentium 4 to Core2 Duo, and that was more than 52% gain.
I mean, let's take Cinebench single-core here: http://www.anandtech.com/show/2045/11
Core2 Duo at 2.66 GHz scores 446.
Pentium D at 3.6 GHz scores 305.
So if we normalize that for clockspeed, that's 446/2.66 = 165.67 vs 305/3.6 = 84.72.
165.67 / 84.72 = 1.96. So they could have marketed it like: 96% higher IPC!!!!

They didn't, because they simply let the product and its benchmarks speak for itself.
With AMD I get the distinct impression that they've cherry-picked some benchmarks (and the Intel CPUs they're comparing against), and then used some very 'creative' number juggling to try and make it sound way impressive.
I'm expecting CPUs that are 'almost' as good as Intel's on IPC and consume 'almost' as little power.
Which you can get for 'almost' as much money, if you like. But not exactly CPUs that will take over the world or anything.
I'm not expecting something like the original Athlon, which was actually better than a Pentium III and Pentium 4 in virtually every way. More like an 'also ran'.

they ignored clockspeed you say, yet they are able to release a CPU with 3.6GHz base and boost up to 4GHz within a 95W TDP, while Intel's 8c alternative is clocked lower (and costs 2x)

IPC talks have always been a thing, even if Intel didn't use them in slides for their Core 2 Duo marketing,
yes C2D IPC gain was far greater, but, I think the improvements are overall harder to achieve nowadays (look at Intel incremental approach), and we are talking about AMD, which is a tiny company compared to Intel, haven't made a decent profit in a very long time, lost a lot of people, and no longer control the fabs,

when AMD publicized their goal of 40% IPC increase, I was skeptic, as most people were, they delivered more, I'm sorry but I'm impressed,
their CPU division looked hopeless with their Bulldozer cores for the past 5 years, this is a big change, their best since K8 (I think k7 and k8 were better at the time, but this is the best since),

Cinebench R15 was considered a totally Intel dominated benchmark and something to be avoided by AMD, and this is the main one they used to present Ryzen, to me it shows they really attacked their main deficiency and are confident with what they have...

I think 4c/6c Ryzen is going to be great for the mainstream (well, what's left of it building desktops, the PC world now is dominated by something else, 15W laptops)

Looks like Ryzen's pre-launch demand is nice and strong
http://techreport.com/news/31485/ryzen-cpus-e … e-launch-demand

And Vega also made an appearance recently:
http://techreport.com/news/31488/amd-early-ve … n-san-francisco

You know, I may be an i7-4930K/GTX970 user, but I am thoroughly enjoying this "new" revitalized AMD. 😊
I may have to purchase an R7-1700 on launch just because I absolutely dig that they're finally sticking it to Intel where it Hz.

Standard Def Steve wrote:

...where it Hz.

Nice 🤣