16 core AMD processor with HT

I doubt we'll see any consumer oriented products for this platform. AMD is shifting focus to two segments: enterprise and consumer APU/mobile.

Would be petty bad ass to see a dually setup running at 3.5+ GHz!
 
Very interesting. However, that is absolutely a 3rd party, possibly warranty voiding "solution" that is not, by any means, a simplistic solution...hell, your average John or Jane Doe computer buyer doesn't even understand what OC'ing is. :(

I want to see 16c/32t AMD processor spec'd at a native 3.5+ GHz. Then I want to see two to four of them running on the same motherboard. :cool:

But then again, I can't even begin to imagine the true reality of power and cooling requirements needed for such a system. FFS, I don't even think I have a wall outlet in my house that is coming off a big enough breaker to support such an electrical load. :p
 
Where did you guys read HT or 32 threads? it's 8 module, 16 core on a single die with 4 HT links.

the title is misleading. in the amd world ht is hyper transport. in the intel world it is hyper threading. the techs arent even close.

so this is a 8m/16cu cpu.
 
Where did you guys read HT or 32 threads? it's 8 module, 16 core on a single die with 4 HT links.

Yep, that's right. HT meaning HyperTransport for purposes of 4-way SMP configurations, not HyperThreading.

Still, a 64c/64t quad CPU SMP system...yes please!
 
Where did you guys read HT or 32 threads? it's 8 module, 16 core on a single die with 4 HT links.
This.

HyperTransport is usually abbreviated as HTT. If AMD ever added SMT (Intel calls its version HT), AMD would call it SMT or some marketing name. :p
 
Looks to be a successor to Abu Dhabi 6300 series. Another positive to look forward to after happily moving from multi-socket Xeon to single socket 6376. 100% stable, versatility of gaming workstation and ESXi server, ~40C under load with Noctua cooler and one system fan and lower system noise. Just picked up another 6328.
 
what cores? steamroller? no mcm sounds good. Here's to hoping AMD becomes relevant in things other than price.
 
Piledriver V2 I believe.

Thats what the roadmap said.

Piledriver V2 is basically Piledriver but lower power usage for the same performance.
 
It would be real nice to see an enthusiast G34 board with PCIe 3.0 support, USB 3.0, SLI/Crossfire to answer to Intel's X79 platform.
 
the title is misleading. in the amd world ht is hyper transport. in the intel world it is hyper threading. the techs arent even close.

so this is a 8m/16cu cpu.

let it be understood here and now that i meant Hyper Transport, a choice i deemed reasonable given that this is an AMD subforum.
 
http://support.amd.com/TechDocs/47414_15h_sw_opt_guide.pdf

(Page 197)

2014-01-18_11-50-59.png


Found via Guru3D.

I highly doubt this is coming to the desktop, but the documents do make mention of the Steamroller modules, so it's likely this will be a 16-core (8 Steamroller modules) server processor on 28nm process.
 
Yeah, that looks like it may be for up to 16-way (using 4 cHT links) per server high density applications. That configuration may be interesting as competition to other GPGPU devices like Xeon Phi and Tesla.
 
It would be real nice to see an enthusiast G34 board with PCIe 3.0 support, USB 3.0, SLI/Crossfire to answer to Intel's X79 platform.

G34 is not enthusiast, they're server parts. By the time X99 or X109 is out...maybe AMD will have something new on the desktop, otherwise it is as things appear and AMD has given up on developing traditional enthusiast CPUs.
 
G34 is not enthusiast, they're server parts. By the time X99 or X109 is out...maybe AMD will have something new on the desktop, otherwise it is as things appear and AMD has given up on developing traditional enthusiast CPUs.

If HSA takes off, all future AMD designs will be APUs.
 
G34 is not enthusiast, they're server parts. By the time X99 or X109 is out...maybe AMD will have something new on the desktop, otherwise it is as things appear and AMD has given up on developing traditional enthusiast CPUs.

true, but C32 [could] be an enthusiast platform, just as AMD tried with its s940 AthlonFX platform back in the day.
 
true, but C32 [could] be an enthusiast platform, just as AMD tried with its s940 AthlonFX platform back in the day.

Wow, I didn't think anyone even remembered the good ole Socket 940 setups...

On another note, I too would love to see a C32 High End Desktop setup based around one of these..If AMD would release a "Black Edition" or "FX" that is unlocked, it would a killer setup to feed 3~4 R9 290s for a workstation/gaming/mining/kick ass rig..
 
true, but C32 [could] be an enthusiast platform, just as AMD tried with its s940 AthlonFX platform back in the day.

Wow, I didn't think anyone even remembered the good ole Socket 940 setups...

On another note, I too would love to see a C32 High End Desktop setup based around one of these..If AMD would release a "Black Edition" or "FX" that is unlocked, it would a killer setup to feed 3~4 R9 290s for a workstation/gaming/mining/kick ass rig..

So that way AMD would spend money they don't have re-engineering a server platform into a desktop platform...and still only have 8-core CPUs? Like they have now?
 
Wow, I didn't think anyone even remembered the good ole Socket 940 setups...

I had dual socket 940s for around 4 years although on Black Friday 2008 I replaced a dual socket dual core with a single quadcore which was much faster and used less than 1/2 the power.
 
So that way AMD would spend money they don't have re-engineering a server platform into a desktop platform...and still only have 8-core CPUs? Like they have now?

Almost all desktop platforms are derived from server platforms, so if it's meant to be a desktop platform, then it's already largely factored into the overall dev budget.

Granted, AMD is a bit unique since they have separated their product lines into server parts and APU parts, but it wouldn't take much for them to shape said server parts into desktop parts.
 
Almost all desktop platforms are derived from server platforms, so if it's meant to be a desktop platform, then it's already largely factored into the overall dev budget.

Granted, AMD is a bit unique since they have separated their product lines into server parts and APU parts, but it wouldn't take much for them to shape said server parts into desktop parts.

They'd have to add USB3 support...redo PCIe for PCIe3....as well as any and all re-creating of the CPUs for black edition use and non-ECC memory...and all for what? CPUs that are no more advanced than what they already have (at least on the C32 front), and still get thrashed by Intel parts in the same price class?
 
You know that the FX line has always been made from re-purposed server parts. They were never designed for desktop.
 
So that way AMD would spend money they don't have re-engineering a server platform into a desktop platform...and still only have 8-core CPUs? Like they have now?

i did suggest this notion originally as the [successor] to C32 being used for an AthlonFX style consumer platform.

other than supporting dual-cpu's C32 suffers from exactly the same outdated architecture (PCIe on chipset) and outdated chipset (no native PCIe 3.0 or USB 3.0), as AM3+.

however, in principle, there is nothing wrong with C32 as a high-end consumer platform, particularly if that investment has already been amortized on a low-end server platform.

this only makes sense if you get the right answers to the following questions:
1. do we think AMD are going to replace AM3+ (with another high-end consumer platform)? [no]
2. do we think AMD can still sell high-end server chips to the desktop market (i.e. >£250 like socket 1366/2011)? [yes]
3. do we think AMD are going to continue with a split server platform with a successor to C32 (i.e. low cost)? [yes]

the point being; since there is nothing to stop them scaling up their offerings on FM2+ into products that could justify the "FX" tag (£150-£250 with more CPU cores), then anything beyond that is going to require something special to compete with Intel in the <£250 market.

the only thing that i can see is dual-sockets, but it would be niche in the same way that AthlonFX was niche...
 
not happening :D

Why not?

Because there will be no C32 successor (i.e. the end of a hi/lo split in server platforms), or;
Because they won't attempt another shot at workstation derived enthusiast platform (s940 Athlon FX)...
 
They'd have to add USB3 support...redo PCIe for PCIe3....as well as any and all re-creating of the CPUs for black edition use and non-ECC memory...and all for what? CPUs that are no more advanced than what they already have (at least on the C32 front), and still get thrashed by Intel parts in the same price class?

If they introduced a next-gen desktop platform that's derived from this new architecture, then it would already have PCIe 3.0 support. It wouldn't take much to remove ECC-only stipulations from the IMC, and I'm almost willing to bet money that this new platform will have USB 3.0 native support (I didn't see any mention about USB revision standards in any articles).
...A successor to Socket AM3+ and FX CPU platform would be easy for AMD to do.

But, your point is quite valid: all for what? We may see marginal increases in IPC (10-15%, maybe 20% TOPS...in benchmarks) over the current gen FX lineup, and that would mean that they would still trail Intel offerings at the same price points (or lower) just as we see today. Alas, an affordable 16 core desktop CPU running at 3.5+ GHz stock could very well be a lucrative venture for AMD. Especially if they kept the capability for 2 and 4 way SMP for us home users. Then there's the concern of power consumption and heat output...

Thinking about the sheer possibility of affordable SMP gives me fond flashbacks of my lower-budget dual PIII 1GHz system I had almost 15 years ago. I so badly wanted a dual Palomino based Athlon MP system, but couldn't afford it. Truly amazing that AMD and Intel are currently selling dual, quad, hex, and octa core consumer desktop CPUs at the price points that they are compared to the first dual cores of yesteryear.

If AMD can release a consumer SMP capable part based on this freshly announced platform, then I dream of a mock-parts list of the following:

FX 9165S (9 series, 16 core, 3.5GHz, SMP capable - yeah, creative huh?)
$300 (x2!)

Dual Socket AM4 (or whatever socket), 8 DIMM, SLI/XF PCIe 3.0
$300

4x8GB RAM (to start with)
$350

290X or 780ti
$700

Total: about $2000, give or take some

Not bad at all for a 32core rig!
 
What AMD has to realize, and what Intel already knows, is that even if the enthusiast PC market is considerably niche, there are still people willing to buy high-end/high-performing CPU parts.

Money is money, even if it's coated in chocolate.

Why should Intel make the -E series processors at all if they're not comparatively faster than the non-E series? They are essentially about as similar as their server versions. Even the upcoming Haswell-E are based on the same Haswell-EP/EN server processors with maybe a few alterations in core count and possibly memory support. The only reason to make an -E series processors is because not only that they know they can do it, but because they know people will buy them and that there is a market for these kind of processors. That's regardless of how small that market is.

Almost everyone laughed at AMD when the FX-60 came out so many years ago especially at the price tag it had. However, there were still people that bought them.

I'm not saying AMD should follow Intel in terms of pricing their processors, they just have to realize that the PC market is not just only entry-level, casual social media Netflix users. If AMD decides to ever make a desktop variant of a 4 to 8 module (8 to 16 core) Opteron processor, people will buy them.

AMD may not have the R&D funding nor the engineering strength as Intel, but they are about as capable of making a desktop processor for enthusiasts if they wanted to.

They still need to modernize or catch up first in a couple areas:

  • HyperTransport is good for multi-processor systems, but inadequate for today's high-bandwidth connections. Drop it for native PCI-e 3.0 controller like what they're doing with their newest APUs.
  • Add quad-channel memory support like they do with their high end Opteron models. Leave the dual channel memory to something like an AMD C6 650 (3 module, 6-core) CPU. (Yes, I'm trying to make something up that follows their new Radeon naming scheme.)
  • One-up Intel and add an on-die SATA3 and USB 3.0 if die space allows, even if it means one or two SATA3 and USB 3.0 connections.
  • Update their antiquated chipsets with PCI-e 3.0, USB 3.0 and SATA3 native support.
  • Allow it the ability to be as modular as their current CPU modules so that when PCI-e 4.0 and DDR4 become mainstream, they can swap out the controllers on both the CPU and chipset with ease.
  • Start with a 28nm Steamroller-based enthusiast line based on the Opterons then move onto smaller processes when they become available to use.
And, believe me, if it worked for Intel, people will buy it. There are still AMD customers around using AMD processors even if the market is smaller than Intel's. And, there is still an enthusiast PC market out there as well.
 
Why should Intel make the -E series processors at all if they're not comparatively faster than the non-E series?
You could have saved a whole lot of typing by knowing that the high end enthusiast Intel CPUs (E series) are just rebadged Xeons and even use the same Xeon 1P/2P server socket. ;) The (LGA1366/LGA2011) motherboards are basically workstation boards with some features more common on desktop motherboards (additional video card slots, more USB ports, extra SATA expandability, etc).

Intel isn't doing anything special that isn't also available on AMD platforms. You can buy AMD workstation boards which also contain desktop niceties to run higher end 1P/2P server CPUs.

On either platform, it's a niche product.
 
I got all excited for nothing. Its my understanding that AMD's new Warsaw server platform won't be release until 2015 or 2016. Since people seem to know what they're talking about in here, I have a question. Is it possible to game on a 16 core Opteron?
 
I have a question. Is it possible to game on a 16 core Opteron?

Yes. Having more cores will not slow down your game. They most likely won't help either since most games are written to use 2 to 4 cores so it will still be very important how well each core performs.
 
Yes. Having more cores will not slow down your game. They most likely won't help either since most games are written to use 2 to 4 cores so it will still be very important how well each core performs.

For streamers it would be awesome if you could dedicate cores to desktop / rendering / capture use as I use all 3 at the same time when streaming. If your transition time between switching sources means that your audience sees lag, freezes, or slowdowns in your stream, they will leave.

A typical high end streamer has:

A ton of browser windows open and multiple browsers open in case something happens to Chrome his 20 tabs are replicated in Firefox.
Is playing a stressful game like BF4 on all Ultra settings on a 64 man server.
Is maintaining a minimum 60fps frame rate.
Is transcoding video from his native resolution to 1080p for the audience. Many streamers have multiple monitor setups.
Is capturing the original video to his local hard drives for Youtube, DailyMotion, Vimeo, etc.
Is running a second game in the background so he can keep viewers as maps transition. Typically a still image screen means 1/10 of your viewers leave to see some live action.
Has a secondary monitor dedicated to chat, VOIP, video chat windows, music, etc that is being encoded into the stream on the fly.

Many high end streamers limit themselves to 1080p 30 fps, turn the game's setting down, and stream with Quad core Intels plus a dedicated capture card to do the rendering. To do 1080p 60 fps you really need a 6 core Intel. Your audience knows if you dip below 60 fps and will leave. You can build a secondary PC to capture the video from the 4770K and render it for your audience, but typically they stuff those with older Xeon processors and aren't cheap when you figure in the price of the motherboard.

A 12 or 16 core AMD chip with high performance per core, would be nice as you could let the game run on 8 cores, dedicate 4 to 1080p 60 fps transcoding, and still have cores left for the OS to switch sources, capture local video, and leave your stream quality seamless.
 
Sweet Jesus, I'm feeling my age:eek: When I gamed, we just had mics for conversations and in-game chat. I'm assuming "high performance per core" means a high clock rate. What do you mean by "switching sources"? Does anyone know if Linux is better at streaming than Windows? I know most software is tailored for Windows platforms.
 
^ That's actually very reasonable prices considering what you get.
 
Not necessarily. Remember that a 3.X GHz P4 got outperformed by a ~2 GHz Athlon.
Not really, but K7 was about 20% faster clock for clock than Northwood. AMD's PR ratings for XP models were pretty close in performance to Northwood running at a particular clock speed (XP 2000+ @ 1.67GHz ~= Northwood 2GHz). And AMD had a superior x87 FPU, which showed better performance in certain applications. http://www.anandtech.com/show/866/5

K8 was a different story. Prescott and Cedar Mill, even with a 1GHz advantage vs Athlon 64 clock speed, struggled to keep up in most applications and gaming.
 
Back
Top