[News] AMD Zen Engineering samples are floating around, specs leak online

[StyM]

http://www.kitguru.net/components/cp...s-leak-online/

It has been a while since we?ve had any interesting developments in regards to AMD?s upcoming Zen processors. However, this week someone managed to get a peek at a few Zen engineering samples, giving us a good idea of what spec AMD is currently working with for its new line of CPUs.

According to a leak that appeared on Guru3d, there are several Zen engineering samples floating around right now, a quad-core, an octa-core, a 24-core and a 32-core. The first two SKUs are going to run on the AM4 socket, while the last two are for server use, tying in to rumours we previously came across regarding AMD?s 32-core Zen plans.

The quad-core Zen CPU runs at a 65W TDP, while the eight-core model runs at a 95W TDP, so we can already start to see efficiency improvements over AMD?s last batch of desktop CPUs. The 24 core and 32 core server SKUs run at 150W and 180W respectively.

Quad-core Zen has eight threads and 2MB of L2 cache or 8MB of L3 cache, meanwhile the octa-core Zen has 16 threads and double the cache. Both of these CPUs currently run at 2.8GHz with a boost clock of 3.2GHz. However, while running in idle mode, both chips can clock down to just 550MHz and consume just 5 watts of power.

Apparently the source of this information has had a good track record with leaks in the past. However, we are not in touch with them, so we can?t independently verify the legitimacy of any of this information, so as with all leaks, take it with a grain of salt.

[Ket]

Zen is meant to be a "brainiac" CPU, doing more with each cycle, so those frequencies would match that philosophy. I hope the CPUs get at least a 300MHz speed bump by release though.

[zalbard]

Can't wait for Zen. Intel's pricing really annoys me.

[AliG]

I have a suspicion these are server samples based off the low power states they support. 5W idle is actually really impressive, but desktop users won't care one way or another.

It just seems like they would have opened up the power envelope for enthusiast parts. The FX-9590 has a *default* boost clock of 5GHz. Even with a 40% increase in IPC, quadcore Zen would need quite a bit more than just 3.2 GHz to surpass that level of performance, let alone catch up to Intel.

[vario]

I have a suspicion these are server samples based off the low power states they support. 5W idle is actually really impressive, but desktop users won't care one way or another.

It just seems like they would have opened up the power envelope for enthusiast parts. The FX-9590 has a *default* boost clock of 5GHz. Even with a 40% increase in IPC, quadcore Zen would need quite a bit more than just 3.2 GHz to surpass that level of performance, let alone catch up to Intel.

After Guru3d

The engineering samples currently are set at revision A0.

If thats true, then its a very good result, these are first batches not meant for sale, pretty much revision B or C goes into retail i gues maybe A3 or something along those lines.Theres room for improvement.If they are floating around now, they were made some time ago, it gives some time before retail production silicon.But yeah, they need 4ghz.

[madmaxx]

Give us some numbers already damnit! We been waiting long enough

[qcmadness]

2.8GHz is too low to actually beat Kaveri (2.8 * 1.4 ~ 4.0).
I think a minimum of 3.1GHz is needed. I hope it can go 3.5GHz in turbo.

[drmrlordx]

I have a suspicion these are server samples based off the low power states they support. 5W idle is actually really impressive, but desktop users won't care one way or another.

It just seems like they would have opened up the power envelope for enthusiast parts. The FX-9590 has a *default* boost clock of 5GHz. Even with a 40% increase in IPC, quadcore Zen would need quite a bit more than just 3.2 GHz to surpass that level of performance, let alone catch up to Intel.

I agree that they need higher clocks, though the cited +40% IPC improvement is versus XV, not PD. Also, by switching to an SMT uarch from a CMT uarch, it's hard to tell exactly what their IPC estimates mean. It may mean "40% faster than XV running a single thread", not "40% higher throughput for n number of Zen cores vs n number of XV modules". So it may not be the wunderkind we all expect depending on the workload. Or maybe it will be . . . we just don't know yet.

[AliG]

I don't think anyone expects that 40% increase to be across the board, just single threaded tasks. If anything, there will be a dropoff in multithreaded scaling simply because that is the one area that CMT is truly dominant.

In theory, CMT could have been the best of both worlds if they found a way to design a scheduler that could asynchronously assign tasks to both cores within a module. In that manner all 4 integer cores could have been combined into solving 1 single threaded task, or broken up into a pair of smaller multithreaded task. I think the reality is that became too difficult, and instead they found it was smarter to just do asynchronous SMT (which is far more flexible).

[RJARRRPCGP]

Can't wait for Zen. Intel's pricing really annoys me.

Same here! Soo QFT!

[zanzabar]

wait for benches, dont let amd disappoint you again.

[zalbard]

wait for benches, dont let amd disappoint you again.

It doesn't have to beat Intel's core.

If AMD gives us an affordable chip with 8 reasonably fast cores (say, within 20% of Intel's core), such a chip will be highly competitive with Intel's 4c/8t offers.

[qcmadness]

It doesn't have to beat Intel's core.

If AMD gives us an affordable chip with 8 reasonably fast cores (say, within 20% of Intel's core), such a chip will be highly competitive with Intel's 4c/8t offers.

I don't think Zen's single-core performance is within 20% of current Intel's offering.

http://www.anandtech.com/bench/CPU/1028

Cinebench R15 single-thread performance:
i7 6700 (4.0GHz @ Single-thread): 173
Athlonx4 845 (3.8GHz @ Single-thread): 92
Supposed Zen @ 3.8GHz: 129 (1.4 * 92)

It's not even close.

[Ket]

Generally speaking A1 or A2 revisions go retail unless the company are REALLY confident then even A0 revisions can creep in to retail channels in the early days but it seems like AMD are making absolutely sure Zen hits the ground running so I'd think A1 will go retail. I'd also think a 8 core Zen would be at least -20% of a 5960X if they are serious about competing with intel again but AMDs rep is so badly damaged due to incompetent management before Lisa Su I think they need to hit harder to inspire confidence again in AMD as a brand so I'd hope their aiming for a 8 core Zen to be -12% that of a 5960X at least and price it at around 250, 8 core is long overdue to be the norm. Lets hope AMD handle this right because they could seriously curb stomp intel with intel saying its going to be 2018 before even poxy 6 core becomes the norm for them.

[AliG]

Generally speaking A1 or A2 revisions go retail unless the company are REALLY confident then even A0 revisions can creep in to retail channels in the early days but it seems like AMD are making absolutely sure Zen hits the ground running so I'd think A1 will go retail. I'd also think a 8 core Zen would be at least -20% of a 5960X if they are serious about competing with intel again but AMDs rep is so badly damaged due to incompetent management before Lisa Su I think they need to hit harder to inspire confidence again in AMD as a brand so I'd hope their aiming for a 8 core Zen to be -12% that of a 5960X at least and price it at around 250, 8 core is long overdue to be the norm. Lets hope AMD handle this right because they could seriously curb stomp intel with intel saying its going to be 2018 before even poxy 6 core becomes the norm for them.

lolwut.

I get why you'd want that, but that's kinda ridiculous to expect a competitor to a chip that still retails for $1000 to be priced at $250. If it's within 12% of the 5960X, you can be sure the price will only 20-25% lower.

[FlanK3r]

I think A1 will be finals revision for customers, but Im thinking, the final clocks will be not much different than now at A0 (maybe +200 or 300 MHz)

[drmrlordx]

Sorry double post.

[drmrlordx]

It doesn't have to beat Intel's core.

No but it should beat XV on a per-thread basis (read: throughput and "single threaded" IPC). If it doesn't, it calls into question the wisdom of replacing the Construction cores.

Zen has enough computational resources per core to at least match XV in throughput. If the design team somehow managed to fall short of that feat, then they deserve a swift kick to the nether regions.

[zanzabar]

It doesn't have to beat Intel's core.

If AMD gives us an affordable chip with 8 reasonably fast cores (say, within 20% of Intel's core), such a chip will be highly competitive with Intel's 4c/8t offers.

i was not saying amd had to win outright, but everything has been a disappointment with amd since the 939/940 with the thulban being kind of fun (maybe since amd gave me one im a bit bias as liking it) but all disappointment.

lolwut.

I get why you'd want that, but that's kinda ridiculous to expect a competitor to a chip that still retails for $1000 to be priced at $250. If it's within 12% of the 5960X, you can be sure the price will only 20-25% lower.

intel charges $1100 for the 6900k but only $600 for the 6850k. the 6850k is about 10% slower, and they charge $1650 for the 6950x and that is only around 7% faster than the 6900k. amd is clearly not targeting that market and wants in on the 6600k and 6700k range to get market share back. even if they were competative in that $600+ range they could not sell anything there. hell in servers look at when the g34 launched. it stomped intel in performance per watt and in performance per U in clusters for highly threaded server applications but people still bought intel like amd was not even an option.

[AliG]

intel charges $1100 for the 6900k but only $600 for the 6850k. the 6850k is about 10% slower, and they charge $1650 for the 6950x and that is only around 7% faster than the 6900k. amd is clearly not targeting that market and wants in on the 6600k and 6700k range to get market share back. even if they were competative in that $600+ range they could not sell anything there. hell in servers look at when the g34 launched. it stomped intel in performance per watt and in performance per U in clusters for highly threaded server applications but people still bought intel like amd was not even an option.

I'll give you that price goes up quadratically for high end products, but again, let's be realistic. $250 is cheaper than a 6700k. You really think AMD would price their top end octocore at that level?

[zanzabar]

I'll give you that price goes up quadratically for high end products, but again, let's be realistic. $250 is cheaper than a 6700k. You really think AMD would price their top end octocore at that level?

with no cooler (like intel does) and needing a more expensive workstation kind of board i would guess they would put it around there for an overclockable but not top sku. also consider that microcenter sells the 6700k for $280.

[Ket]

lolwut.

I get why you'd want that, but that's kinda ridiculous to expect a competitor to a chip that still retails for $1000 to be priced at $250. If it's within 12% of the 5960X, you can be sure the price will only 20-25% lower.

Simply put, everything Zanzabar has said, plus intel prices are only as high as they are because AMD has been unable to compete.

[qcmadness]

i was not saying amd had to win outright, but everything has been a disappointment with amd since the 939/940 with the thulban being kind of fun (maybe since amd gave me one im a bit bias as liking it) but all disappointment.



intel charges $1100 for the 6900k but only $600 for the 6850k. the 6850k is about 10% slower, and they charge $1650 for the 6950x and that is only around 7% faster than the 6900k. amd is clearly not targeting that market and wants in on the 6600k and 6700k range to get market share back. even if they were competative in that $600+ range they could not sell anything there. hell in servers look at when the g34 launched. it stomped intel in performance per watt and in performance per U in clusters for highly threaded server applications but people still bought intel like amd was not even an option.

opposite with what i learned in various reviews

[AbortRetryFail?]

... intel prices are only as high as they are because AMD has been unable to compete.

I love this meme, false narrative that it is.

I realize we are *Extreme* but for 98.7% of computer users, their significant utilization is *CPU Idle*

Even the "Intel is 60% faster" in context is ridiculous. Yeah. It will save you 7/10ths of a second --- 1.7s v. 2.5s in Photoshop

OMG!

[AliG]

He's not wrong though, that's just basic supply and demand.

Laptop/Desktop OEMs do look at price:performance.