Intel engineers calls ryzen competitive

They are used in OEM systems however. Although OEMs tend to limit motherboard features.
 
Riptide_NVN said:
I'd be more concerned about bugs with the first generation motherboards than I would the CPU.
Click to expand...

You are not wrong. CPU's often have erratum that can be pretty scary but microcode updates and common BIOS workarounds employed by all manufacturers can usually resolve those things. What's more difficult to deal with are integrated features, chipset level problems and the various firmware(s) that come together on a motherboard.
 
Riptide_NVN said:
As of this morning you can't buy an AM4 motherboard off newegg. Maybe I'm missing something but to me that doesn't exactly spell out widespread adoption and testing.

Hoping for the best.
Click to expand...

QVL testing for any CPU (including associated motherboards) happens well ahead of announced release dates. Not being able to buy an AM4 motherboard before the CPU is released isn't a concern.
 
CentroX said:
Also revealed that Ryzen chips dies are smaller then skylake dies.

https://www.techpowerup.com/230446/amds-ryzen-chips-10-smaller-than-comparable-intel-skylake-dies
Click to expand...
If the Ryzen dies are indeed smaller (or even the same saize) as 4C8T Ivy Bridge or later, then that would rule out a soldered IHS. It will be interesting to see how close AMD are willing to place the IHS bottom to the die top: Intel's 0.06mm is conservative, trading off thermal performance for robustness. If AMD go for a finer tolerance they may gain some thermal headroom, at the risk of damage from direct die contact. Personally I'd expect the gap to if anything be slightly larger, as AMD's stock thermal solution requires the heatsink to be clamped on one side and then the other (giving a huge off-axis load to the IHS), while the LGA11xx design progressively tightens around the corners (as long as you follow the instructions).
 
EdZ said:
If the Ryzen dies are indeed smaller (or even the same saize) as 4C8T Ivy Bridge or later, then that would rule out a soldered IHS. It will be interesting to see how close AMD are willing to place the IHS bottom to the die top: Intel's 0.06mm is conservative, trading off thermal performance for robustness. If AMD go for a finer tolerance they may gain some thermal headroom, at the risk of damage from direct die contact. Personally I'd expect the gap to if anything be slightly larger, as AMD's stock thermal solution requires the heatsink to be clamped on one side and then the other (giving a huge off-axis load to the IHS), while the LGA11xx design progressively tightens around the corners (as long as you follow the instructions).
Click to expand...

The Ryzen dies isn't smaller. They are about 200mm2, close to SB 4C (216mm2).

IB 160mm2.
HW 177mm2.
SKL/KBL 122mm2.
 
EdZ said:
If the Ryzen dies are indeed smaller (or even the same saize) as 4C8T Ivy Bridge or later, then that would rule out a soldered IHS. It will be interesting to see how close AMD are willing to place the IHS bottom to the die top: Intel's 0.06mm is conservative, trading off thermal performance for robustness. If AMD go for a finer tolerance they may gain some thermal headroom, at the risk of damage from direct die contact. Personally I'd expect the gap to if anything be slightly larger, as AMD's stock thermal solution requires the heatsink to be clamped on one side and then the other (giving a huge off-axis load to the IHS), while the LGA11xx design progressively tightens around the corners (as long as you follow the instructions).
Click to expand...

It doesn't even need to be soldered, they could just use a better TIM.
 
This disappoints me:

http://wccftech.com/ryzen-smaller-die-intel-zen-architecture-not-good-hpc/

"Its also worth pointing out that Intel has a double precision IPC of 16 FLOPs per Clock with Skylake as well as 2x 256 bit FMA whereas Zen only has 8 FLOPs per clock and 2* 128 bit FMA."

I wouldn't call that competitive. I really, really wanted AMD to have good floating point power and no glaring weaknesses, but here we go....
 
"Its also worth pointing out that Intel has a double precision IPC of 16 FLOPs per Clock with Skylake as well as 2x 256 bit FMA whereas Zen only has 8 FLOPs per clock and 2* 128 bit FMA."
Click to expand...

This doesn't matter on the desktop, so it effectively doesn't matter to [H]
 
pgaster said:
This disappoints me:

http://wccftech.com/ryzen-smaller-die-intel-zen-architecture-not-good-hpc/

"Its also worth pointing out that Intel has a double precision IPC of 16 FLOPs per Clock with Skylake as well as 2x 256 bit FMA whereas Zen only has 8 FLOPs per clock and 2* 128 bit FMA."

I wouldn't call that competitive. I really, really wanted AMD to have good floating point power and no glaring weaknesses, but here we go....
Click to expand...

Wccftech...

Using info from middle of last year.
 
Riptide_NVN said:
I'd be more concerned about bugs with the first generation motherboards than I would the CPU.
Click to expand...

This is definitely my primary concern. My early P55 board was a total nightmare for the first 4-ish months.

*trying to avoid the hype-train

"Fingers crossed, applesauce, make my dreams come true!"
 
In addition, to obtain this type of performance (8 flops/clock for AMD or 16 flops/clock for Intel part) you need special libraries as intel MKL (for intel part) or OpenBlas, by example, where the mathematical kernels are written in assembler code.
 
yeah aren't most supercomputers these days moving computations to GPUs?
 
frunction said:
Wccftech...

Using info from middle of last year.
Click to expand...

That doesn't make it wrong tho since those they copied from is correct.

http://www.linleygroup.com/mpr/article.php?id=11666

upload_2017-2-8_21-11-41.png
 
Yes, the problem with GPUs are the applications that require a lot of memory and a lot of memory bandwidth at the same time. AMD has been working long time in line to integrate the GPU and CPU because AMD thinks that the future for scientific computing is that.
 
ManofGod said:
Anything to attempt to put AMD in a bad light, eh? No surprise there though, this is par for the course.
Click to expand...
Why even post this? Can you guys at least keep your bickering to one thread and not just bash on each other in every new Ryzen thread that pops up?
 
ManofGod said:
Anything to attempt to put AMD in a bad light, eh? No surprise there though, this is par for the course.
Click to expand...
AMD isn't an embattled politician, an oppressed historical figure, a refugee, or an underdog baseball team, they're not David and intel isn't Goliath. Why anyone is emotionally attached to AMD or intel makes no sense to me. They make chips and sell them to you. You don't get a pat on the back. You don't get a personal letter thanking you. They each just try to do things to stay relevant and appease shareholders and stay in business. So why the hell are people emotionally invested in the underdog story of AMD? Just choose whatever works for you and your budget and quit trying to convince someone if they choose intel they should be choosing AMD and vice versa.

Edit: But if we did all this what would be the point of coming to the forums to show off and argue our opinions. Never mind. As you were ;)
 
Shintai said:
In 256bit AVX/FMA loads? Yes. For everything else? No.

That's all you can conclude from it.
Click to expand...

Right, the AVX throughput is half that of Skylake. But how many consumer or server applications need that kind of AVX throughput? Only h.265 encoding (with the much larger blocks) and rendering/machine learning really benefit.

But the more typical mixed integer/float/SSE4 throughput is more competitive, and you get more cores at a lower power point as a result.

Basically, this is what Bulldozer SHOULD have been: not a clunky shared FPU/front-end unit, but instead a smaller split 2x128-bit dedicated FPU per-core, using HT to get the most out of it.

That load/store bandwidth advantage only comes into play when you have something embarrassingly parallelizable, and they're hoping you'll use a GPU for that. We'll have to wait and see how much this harms performance, but I'm betting it will be mostly transparent in real-world (like the Pentium Kaby Lake with AVX disabled is similar to i3 in most games, office and pro applications).

The most important thing you can conclude from that diagram is HOW CLOSE in performance Haswell is to Sandy Bridge, even though one has twice the bandwidth.
 
Last edited:
Ocellaris said:
It doesn't even need to be soldered, they could just use a better TIM.
Click to expand...
The TIM in Ivy Bridge and onward is just fine, it's the clearance gap between the die and the IHS that causes the 'issue':
[...]the only functional differences in these two test conditions is the CPU TIM. Notice that the Intel stock CPU TIM outperforms the NT-H1 replacement TIM once the CPU-to-IHS gap is identical
Click to expand...
i7-3770KtemperatureswhenshimmedwithNTH1.png


A gap is critical to prevent damage to the die when force is applied to the IHS. The smaller the gap, the less force can be applied before the IHS contacts the die and damage can occur.
 
I am trying to get blanket affirmations from a person i know that has published reviews globally, has been involved in hardware testing including with Ryzen and holds a few overclocking records. His last leak i made the mistake of verbatim and it was reverse searched. The leak went viral and AMD stepped in gave him a verbal warning so i cannot get direct answers only skirted ones but his information is very similar to Doc Terable from CPC Hardware.

IPC wise Zen is legitimately strong even told that it takes a highly overclocked 2600K to keep up. a 4.4ghz sandy barely matches a Haswell IPC at mid 3 per Cinebench which gives weight to the expectations of being somewhere between Haswell and Broadwell.

The IMC is not quite as beefy as one would expect and is a limiting factor on IPC/HPC loads however Zen+ likely features major upgrades there. This was more stage 2 development.

Boards have bugs and boost is still not working, overclocking is troublesome. He says it is a combination of new chipset and learning the process by Global Foundries. Expect better silicon each new stepping. They will have new BIOS soon hopefully fixing stability and giving a first taste of working turbos.

Thats more or less.
 
AV
whateverer said:
Right, the AVX throughput is half that of Skylake. But how many consumer or server applications need that kind of AVX throughput? Only h.265 encoding (with the much larger blocks) and rendering/machine learning really benefit.

But the normal integer/float throughput is more competitive, and you get more cores at a lower power point as a result.

Basically, this is what Bulldozer SHOULD have been: not a clunky shared FPU/front-end unit, but instead a smaller split 2x128-bit dedicated FPU per-core, using HT to get the most out of it.

That load/store bandwidth advantage only comes into play when you have something embarrassingly parallelizable, and they're hoping you'll use a GPU for that. We'll have to wait and see how much this harms performance, but I'm betting it will be mostly transparent in real-world (like the Pentium Kaby Lake with AVX disabled is similar to i3 in most games, office and pro applications).

The most important thing you can conclude from that diagram is HOW CLOSE in performance Haswell is to Sandy Bridge, even though one has twice the bandwidth.
Click to expand...

AVX is the new Cherry picked instruction code to move the goal posts to suit inuendo. AVX is more Intel proprietary standard hence why so little adoption. Testing ryzen in an intel design spec is just funny. Even then it got between ivy and Haswell.

Also when a person gives zero credit to a company for its efforts, normally you need to pay to cross that bridge.
 
OrangeKhrush said:
IPC wise Zen is legitimately strong even told that it takes a highly overclocked 2600K to keep up.
Click to expand...

If this is the case, given anecdotally Intel hasn't had a lot of luck convincing people with overclocked 2600K's to upgrade to their latest and greatest, what's going to be the draw here to get those people to move away from their 5-6 year old processors, let alone those released since then? I know chipset features are often cited as reason to, but that applies to both AMD and Intel so... extra cores? Lower pricing? Those looking for a means to stick it to Intel, "side-grade" or not? If its primarily going draw those that are already brand loyal, not sure that's going to help build up market share.
 
Intel also says that Kaby lake is 15% faster than Sky lake. I think whoever is in charge of metrics has been drinking a little. I wouldn't invest too much into this.
 
AltTabbins said:
Intel also says that Kaby lake is 15% faster than Sky lake. I think whoever is in charge of metrics has been drinking a little. I wouldn't invest too much into this.
Click to expand...
They are comparing mobile SKUs, maybe even ultra mobile SKUs, in which case Intel's words are most likely legitimate.
 
Dan_D said:
You are not wrong. CPU's often have erratum that can be pretty scary but microcode updates and common BIOS workarounds employed by all manufacturers can usually resolve those things. What's more difficult to deal with are integrated features, chipset level problems and the various firmware(s) that come together on a motherboard.
Click to expand...

Via KT133 rings a bell from the Socket A days. Regretted not waiting for the Abit KT7-A RAID instead of the KT7 RAID I suffered with.
 
Tzeh Pesh said:
If this is the case, given anecdotally Intel hasn't had a lot of luck convincing people with overclocked 2600K's to upgrade to their latest and greatest, what's going to be the draw here to get those people to move away from their 5-6 year old processors, let alone those released since then? I know chipset features are often cited as reason to, but that applies to both AMD and Intel so... extra cores? Lower pricing? Those looking for a means to stick it to Intel, "side-grade" or not? If its primarily going draw those that are already brand loyal, not sure that's going to help build up market share.
Click to expand...

Price has a lot to do with it. In the RoC he says the price on a 5960X has gone down substantially, and Ryzen is slightly faster at was half the price.

Unfortunately AMD motherboards cost the same as a identical spec intel board.
 
Intel announcing new CPUs and already promising Canon Lake to be 15% faster (in one bench that is) truly shows that panic is in order. When Ryzen isn't even out yet and they are already scrambling is evidence in itself. I have never seen intel trying to move so quick on something lol..
 
M
NKD said:
Intel announcing new CPUs and already promising Canon Lake to be 15% faster (in one bench at least) truly shows that panic is in order. When Ryzen isn't even out yet and they are already scrambling is evidence in itself. I have never seen intel trying to move so quick on something lol..
Click to expand...
15% with clockspeed * (IPC/SmT)
 
Are people still thinking desktop? Cannon Lake is not coming to desktop. For Desktop its Coffee Lake and they could claim up to 40-50% faster there due to 6 cores.

SKL->KBL in mobile ~20%.
KBL->CNL in mobile ~15%+

Remember what the biggest limitation in mobile is. Specially 15W parts.
 
They didn't mention mobile, they mentioned like with Kaby lake that it was desktop, except it was like a 7700K vs a 6600 at much lower clocks in claiming 20% which is a bit meh. They claim 20% in sysmark, intels numero uno paint in good light bench.
 
You must log in or register to reply here.
Back
Top