AMD New Horizon

[LeStringMan]

Well according to pcworld, they got to talk to some AMD representatives, pcworld claims that the Zen machine was running a prototype AM4 board with 16gbs of ram in dual channel, and it consumed 187.6W under load in the blender test, the 6900k machine was running an Asus ROG x99 board with 32gb of ram in quad channel (to avoid rumors that AMD handicapped it with subpar memory setup), it pulled 191.8W in the blender test. And for the Battlefield 1 demo the same setups were used, they were running at 4k in DX11 Ultra settings with SLI Titan Xs. ( I don't know if any of this is true, its just what pc worlds article claims the setups used were, this info is about halfway through the article for anyone interested.)

http://www.pcworld.com/article/3149051/components/amd-offers-more-proof-that-zen-now-renamed-ryzen-is-its-best-chip-in-a-decade.html

[porina]

8 minutes ago, LeStringMan said:

Well according to pcworld

...

the 6900k machine was running an Asus ROG x99 board with 32gb of ram in quad channel (to avoid rumors that AMD handicapped it with subpar memory setup)

Image below I screenshot at the end of the presentation yesterday. It states 2x8GB for the 6900K system.

 

 

[SpaceGhostC2C]

5 hours ago, Swatson said:

Absolutely, like I said 3rd party benchmarks trump all, but it's completely unfair to assume AMD skewed anything. They could have quite possibly but from the results we saw there is no reason to assume they did.

 

Waiting for 3rd party benchmarks and implying a company skewed results are two different things.

Yes, the word is "flawed", not skewed (at least so far). What we know is that it is not replicable with the information and files provided. We don't have evidence that it was biased against the 6900K yet.

 

Anyway, I don't care that much about trusting this or not, because we'll have better information by the time it actually matters (i.e., when we actually face the option to buy it ).

[done12many2]

3 minutes ago, SpaceGhostC2C said:

Yes, the word is "flawed", not skewed (at leas so far). What we know is that it is not replicable with the information and files provided. We don't have evidence that it was biased against the 6900K yet.

 

Anyway, I don't care that much about trusting this or not, because we'll have better information by the time it actually matters (i.e., when we actually face the option to buy it ).

 

The guys on overclock.net figured out how they pulled off the amazing render times that I was commenting on earlier.  There was a better close up of the render test during an event previous to the AMD New Horizon event and it clearly shows that AMD ran the test at 100 samples vice the 200 "default" setting that they challenged the public to use for their own evaluation.  Dr. Su stated that you just needed to download and run it as configured to compare, yet doing so would result in ANY system being slower than their results.

 

The fact remains that the 6900k and the Ryzen did complete the test in roughly the same time, but at least we can accurately compare our own setups to the capabilities of the new chip.  

 

I'll be rerunning the render test with their unmentioned settings, but I can already tell the it'll be easy to beat the 35 second run that they achieved, which makes a lot more sense.

 

[XenosTech]

3 minutes ago, done12many2 said:

 

The guys on overclock.net figured out how they pulled off the amazing render times that I was commenting on earlier.  There was a better close up of the render test during an event previous to the AMD New Horizon event and it clearly shows that AMD ran the test at 100 samples vice the 200 "default" setting that they challenged the public to use for their own evaluation.  Dr. Su stated that you just needed to download and run it as configured to compare, yet doing so would result in ANY system being slower than their results.

 

The fact remains that the 6900k and the Ryzen did complete the test in roughly the same time, but at least we can accurately compare our own setups to the capabilities of the new chip.  

 

I'll be rerunning the render test with their unmentioned settings, but I can already tell the it'll be easy to beat the 35 second run that they achieved, which makes a lot more sense.

And here is the whole world accusing AMD of skewing results to make Zen look good and howling for their blood

[DocSwag]

8 hours ago, That Norwegian Guy said:

6 core or GTFO. That's the optimized count for Star Citizen.

Well ATM for the majority of games 4 cores is the best, but in star citizen the future 6 cores will be the best.

[Pample]

14 hours ago, Trevor87 said:

They did over shot Zen features a bit in the video because decent Intel motherboards and cpu have had some of those features for years.       

I know other people have already debunked this already, but other than DDR4, etc, a LOT of new tech has gone into ZEN. I'm actually very excited about the auto overclocking feature, I think ity could really be impressive, if it indeed works the way they claim.

 

I really, REALLY, REALLY, REALLY hope AMD are back with a competative chip that gives Intel a run for their money.

[LeStringMan]

18 minutes ago, porina said:

Image below I screenshot at the end of the presentation yesterday. It states 2x8GB for the 6900K system.

-snip-

Lol I didn't watch all the way to the end, guess I should have, just thought I would post what pc world claimed in that article.

[Trevor87]

This answer my question OC

   

[done12many2]

29 minutes ago, Pample said:

I really, REALLY, REALLY, REALLY hope AMD are back with a competative chip that gives Intel a run for their money.

 

It only helps us all.

[Pample]

2 minutes ago, Trevor87 said:

This answer my question OC

   

Nope. That is the BIOS running stabality tests to overclock the processor, NOT the CPU determining it's own speed based on termal peramiters. NOT THE SAME.

[SpaceGhostC2C]

1 minute ago, Pample said:

Nope. That is the BIOS running stabality tests to overclock the processor, NOT the CPU determining it's own speed based on termal peramiters. NOT THE SAME.

I think the term "overclock" is confusing him. The best way to explain what AMD claimed to have is by comparing it with the way Pascal GPUs "clock themselves".

[Pample]

Just now, SpaceGhostC2C said:

I think the term "overclock" is confusing him. The best way to explain what AMD claimed to have is by comparing it with the way Pascal GPUs "clock themselves".

I hope your right, otherwise he is very off the mark.

[Pample]

6 minutes ago, done12many2 said:

 

It only helps us all.

That's true. It will be even better if VEGA give NVIDIA a run as well. We will be back to the days where we don't get ripped off (as much) for our hardware and have a decent choice when handing over large amount's of cash.

[S w a t s o n]

Not gonna lie, feels good to be right ;P

[XenosTech]

2 minutes ago, Swatson said:

Not gonna lie, feels good to be right ;P

Every time I see your name I always ask myself why am I posting from another account 

[That Norwegian Guy]

3 hours ago, Pample said:

I'm actually very excited about the auto overclocking feature, I think ity could really be impressive, if it indeed works the way they claim.

INB4 it means: dynamically overclocking cores the NSA is operating through backdoors while downclocking the ones you have available (if any) to fight back on your Windows 10-gestapo-mandated system

 

[LukaH]

i just figured it out 

 

AMD = Phoenix 

 

Keep making hot CPUs/GPUs until you burn out, then RyZE(n) out of the ashes 

[Mira Yurizaki]

The thing I'm trying to wrap my head around is AMD's marketing claiming the front-end of the processor has a neural network as if it's AI or something. The problem I have with this is if it's something like a neural network, it needs data from previous history (and probably lots of it) before it becomes useful, and I doubt this is something that can live in the confines of the CPU itself, unless they added on like 128MB or 256MB of L3 cache. Oh, and that's volatile memory. If you cut the power in any way (including going into S3 standby), that memory goes away, meaning the CPU would have to relearn everything.

 

At least with Project Denver, optimized code is stored in RAM. Even if the system goes into S3 standby or similar, it's not going to be flushed.

 

So the marketing here just... AMD please hire another marketing team.

[DocSwag]

50 minutes ago, M.Yurizaki said:

The thing I'm trying to wrap my head around is AMD's marketing claiming the front-end of the processor has a neural network as if it's AI or something. The problem I have with this is if it's something like a neural network, it needs data from previous history (and probably lots of it) before it becomes useful, and I doubt this is something that can live in the confines of the CPU itself, unless they added on like 128MB or 256MB of L3 cache. Oh, and that's volatile memory. If you cut the power in any way (including going into S3 standby), that memory goes away, meaning the CPU would have to relearn everything.

 

At least with Project Denver, optimized code is stored in RAM. Even if the system goes into S3 standby or similar, it's not going to be flushed.

 

So the marketing here just... AMD please hire another marketing team.

If I remember correctly they store the data in the cache. I don't remember where I heard it from; it might have been Anandtech or Paul's Hardware.

[Mira Yurizaki]

25 minutes ago, DocSwag said:

If I remember correctly they store the data in the cache. I don't remember where I heard it from; it might have been Anandtech or Paul's Hardware.

Anandtech thinks it's L3 cache.

 

I'm just skeptical about all this. I don't think something that learns should put its memory in something more volatile than RAM.

[leadeater]

5 hours ago, done12many2 said:

The guys on overclock.net figured out how they pulled off the amazing render times that I was commenting on earlier.  There was a better close up of the render test during an event previous to the AMD New Horizon event and it clearly shows that AMD ran the test at 100 samples vice the 200 "default" setting that they challenged the public to use for their own evaluation.  Dr. Su stated that you just needed to download and run it as configured to compare, yet doing so would result in ANY system being slower than their results.

 

The fact remains that the 6900k and the Ryzen did complete the test in roughly the same time, but at least we can accurately compare our own setups to the capabilities of the new chip.  

 

I'll be rerunning the render test with their unmentioned settings, but I can already tell the it'll be easy to beat the 35 second run that they achieved, which makes a lot more sense.

For a short press event I can see why they went with 100 samples so I can give some benefit of doubt that they weren't intentionally trying to mislead everyone by using that setting. Failing to mention the sample setting was extremely stupid however.

 

Let us know what your render times are now, also would be nice if you ran it at a lower frequency too (3.4-3.8 ish)

[done12many2]

1 minute ago, leadeater said:

For a short press event I can see why they went with 100 samples so I can give some benefit of doubt that they weren't intentionally trying to mislead everyone by using that setting. Failing to mention the sample setting was extremely stupid however.

 

Agreed on both.

 

1 minute ago, leadeater said:

 

Let us know what your render times are now, also would be nice if you ran it at a lower frequency too (3.4-3.8 ish)

 

I'm about to head out on a couple of errands.  I'll rerun it with stock and overclock as soon as I get back.

[Mihle]

My next system might be full AMD if the things are true.

[done12many2]

1 hour ago, leadeater said:

For a short press event I can see why they went with 100 samples so I can give some benefit of doubt that they weren't intentionally trying to mislead everyone by using that setting. Failing to mention the sample setting was extremely stupid however.

 

Let us know what your render times are now, also would be nice if you ran it at a lower frequency too (3.4-3.8 ish)

 

Well to further confuse things, AMD had an earlier event to the AMD New Horizon event in which they compared the same two processors (Ryzen and 6900k), but used a completely different version of Blender from what they used for the AMD New Horizon demo.  In this video, you can see that they ran it at a 100 sample, but it's a different version of Blender so nothing's comparable between the data from the two events.  The render times for the earlier event were in the 20's vice the 30 second range for the New Horizon event.  

 

It's also worth mentioning that AMD displayed total system draw during the render in the earlier event.  System configuration is unknown, but the idle draw on the AMD system was lower while the load draw was fairly even.

 

With all of this said, there's not enough information provided from AMD to even try to compare your own system per their challenge. They simple said here's one of the two files we said we'd provide and this is the program to use.  Figure out he rest on your own.  

 

Previous event:  Time stamped for Blender test

 

AMD New Horizon: Time stamped for Blender test