AMD Ryzen 9 4950X "Vermeer" Tested, the Sample Boosts to 4.8 GHz

[Xuper]

I prefer to use 5K instead of 5000.don't know why.

 

[ratirt]

I prefer to use 5K instead of 5000.don't know why.

You can just go 5Ghz 5k Mhz is kinda ridiculous just like 5kk Khz

 

[lexluthermiester]

No one has surpassed by any great margin the 5Ghz AMD launched first five years ago and no one will.

Not true. Substrate materials to replace silicon are and have been created. For example, Gallium Nitride has been demonstrated to be able to reach and exceed 6.4ghz when used as a base for an IC. Gallium Arsenide likewise reaches and exceeds 7ghz, depending on formulation. There are other semi-conductors being considered as well. Scientists are currently experimenting with silicon/arsenic/tellurium blend compounds to find a suitable way forward in mass production. We will get past 5ghz, it's a mathematical certainty. How that is achieved is the question.

 

[TheoneandonlyMrK]

Not true. Substrate materials to replace silicon are and have been created. For example, Gallium Nitride has been demonstrated to be able to reach and exceed 6.4ghz when used as a base for an IC. Gallium Arsenide likewise reaches and exceeds 7ghz, depending on formulation. There are other semi-conductors being considered as well. Scientists are currently experimenting with silicon/arsenic/tellurium blend compounds to find a suitable way forward in mass production. We will get past 5ghz, it's a mathematical certainty. How that is achieved is the question.

I agree , I should have been specific I meant near term ,next five to ten years.
Materials research does and is taking time, but at the end of the day market realities will truly decide when we see such things.

 

[lexluthermiester]

I agree , I should have been specific I meant near term ,next five to ten years.

That's kinda what I mean, advances are coming soon, both in the chemistry of semi-conductors and in the way they're made.

For example, TSMC's 5nm will very likely exceed 5ghz easily. Intel is at 5ghz and just beyond right now with 14nm. 12nm/10nm/7nm is going to produce IC's with stable speeds in excess of the 5ghz mark. This is the next 1, 2 or 3 years. Intel will get their butt's in gear and answer AMD's Ryzen. The next few years for CPU's/chipsets/GPU's are going be interesting if not flat out exciting!

 

[Makaveli]

Essentially the same boost and base speeds as Zen 2. That is a slight disappointment.

(There seems to be a mistake in the last sentence. Those OPN codes should indicate a 3.7GHz base, not 3.5.)

They are still on 7nm I wasn't expecting much of an increase in clock speed. Most of the performance will come from the changes in the arch.

Everyone wanting to see the mythical 5Ghz Boost clocks will most likely have to wait until Zen 4 on 5nm.

 

[TheoneandonlyMrK]

That's kinda what I mean, advances are coming soon, both in the chemistry of semi-conductors and in the way they're made.

For example, TSMC's 5nm will very likely exceed 5ghz easily. Intel is at 5ghz and just beyond right now with 14nm. 12nm/10nm/7nm is going to produce IC's with stable speeds in excess of the 5ghz mark. This is the next 1, 2 or 3 years. Intel will get their butt's in gear and answer AMD's Ryzen. The next few years for CPU's/chipsets/GPU's are going be interesting if not flat out exciting!

I disagree with your timescales then , Intel and AMD can dance around the 5Ghz mark for the next five years, changes nothing they have been already for year's, you mentioned chemical doped silicon helping to get to 6Ghz, I repeat that's years off IMHO.
Most of the advantages of nanowire gates will take many years of continuous process improvement to realise too.
They're are also still too many process constraints to making EUV chips that need to be resolved and worked through too see Intel for proof.
Again an opinion.

I ran my fx8350 at 5Kmhz and I swapped that out after five years service years ago now!?.

 

[Frick]

You can just go 5Ghz 5k Mhz is kinda ridiculous just like 5kk Khz

5KKKHz.

 

[PowerPC]

I prefer to use 5K instead of 5000.don't know why.

I prefer using 5K monitors.

 

[Xuper]

I prefer using 5K monitors.

Oh shut it.

 

[PowerPC]

Oh shut it.

 

[[h+]³]

It sure does; I bought my ATi X800XT back then to play HL2. The only thing we are now missing is... the competitive prices of those times. Hopefully, we won't get screwed over too badly.

GTFO, I bought ATi X800XT PE at the same time. Great times.

 

[Alpha_Lyrae]

So long as there isn't clock speed regression, that's fine. Vermeer, though, has a unified 32MB L3 per 8 core CCD. That's much more important. Now every core (within CCD) can access the same L3 pool without hitting IO die/UMC (unlike the dual 4-core CCX per CCD).

I wonder if any L2 cache increases are on the horizon (for Zen4+). Seems somewhat unnecessary with such a large L3 victim pool, but there's always a latency cost when fetching evicted data from L3 and L2 is more localized to each core. Prefetchers are likely updated, since they're going through 2x total L3 cache space vs Zen 2. MicroOp cache increased too? Guess we'll find out soon. Should be a decent overhaul.

 

[net2007]

I own a 3950x watercooled and never it hits the 4.7. I would say more marketing BS.

 

[nemesis.ie]

It sure does; I bought my ATi X800XT back then to play HL2. The only thing we are now missing is... the competitive prices of those times. Hopefully, we won't get screwed over too badly.

And HL3 too!

 

[ARF]

This is an ES. Final production CPUs could have higher clocks.
Coupled with high enough IPC improvement, this could turn out to be really exciting launch.

But still... can it beat Renoir's efficiency and performance at the same TDP?

 

[midnightoil]

That's kinda what I mean, advances are coming soon, both in the chemistry of semi-conductors and in the way they're made.

For example, TSMC's 5nm will very likely exceed 5ghz easily. Intel is at 5ghz and just beyond right now with 14nm. 12nm/10nm/7nm is going to produce IC's with stable speeds in excess of the 5ghz mark. This is the next 1, 2 or 3 years. Intel will get their butt's in gear and answer AMD's Ryzen. The next few years for CPU's/chipsets/GPU's are going be interesting if not flat out exciting!

Intel's 10nm can't get anywhere near the clocks of its 14nm+++++. Apart from very poor yields, which reportedly have improved slightly now, clock speed is what's stopping them committing to it. Their desktop and server chips are even less competitive on it than 14nm++++++. The little that's leaked out about 7nm, the yields are so bad that it's unlikely to be commercially viable even in big margin HPC or server in the next 18 months, and clocks are no better than their 10nm. Hence the recent announcement of delay (again). They're royally screwed.

They're probably going to have to build a lot of stuff at Samsung in the coming years, with their fabs either doing mostly or exclusively solid state memory (which they make massive profits from). If they stuck to their own fabs for processors & GPUs, realistically they might only just beat AMD's 3nm products to market with their 7nm, since further delays to their 7nm are almost inevitable ... and as you'd expect, the performance and efficiency delta would be laughable.

 

[TheoneandonlyMrK]

This is an ES. Final production CPUs could have higher clocks.
Coupled with high enough IPC improvement, this could turn out to be really exciting launch.

But still... can it beat Renoir's efficiency and performance at the same TDP?

Why would a three die 32 core desktop chip ever match a mobile chips efficiency, or performance and not even with eco mode is it doing 35 watts loaded
Why not throw in half the heat output for the win too.

 

[midnightoil]

But still... can it beat Renoir's efficiency and performance at the same TDP?

Of course, since it has much higher IPC and it'll be on 7nm+. Renoir's Zen2 and 7nm, just with a more mature process and some improvements to the silicon.

 

[Alpha_Lyrae]

Of course, since it has much higher IPC and it'll be on 7nm+. Renoir's Zen2 and 7nm, just with a more mature process and some improvements to the silicon.

Renoir is monolithic, which has power efficiency advantages. IO die and critical IF links need to remain powered on desktop with some light power gating for accessories (USB/SATA/NVMe) and PCIe lanes and non-critical IF links. You can't turn off the UMC in either, but it seems Renoir's UMC has more aggressive downclocking and power reductions. It's simply unnecessary in desktop, as battery life isn't a factor.

 

[ARF]

Renoir is monolithic, which has power efficiency advantages. IO die and critical IF links need to remain powered on desktop with some light power gating for accessories (USB/SATA/NVMe) and PCIe lanes and non-critical IF links. You can't turn off the UMC in either, but it seems Renoir's UMC has more aggressive downclocking and power reductions. It's simply unnecessary in desktop, as battery life isn't a factor.

Monthly electricity bills are a real thing, though.

 

[Imsochobo]

I think I am going to upgrade to Ryzen 4000 with an iGPU, I am very curious on how they will perform in gaming, I mostly play older generation games with the odd new AAA title and a Ryzen iGPU will tie me over until I can replace this GTX660Ti, sadly my GTX1070 died and only got half my money back for it.

For now, I can only choose one thing to upgrade, a new CPU (With RAM and Mobo) or a new shiny GPU. I have always wanted to try the AMD camp, this seems a good time as any to do so, I am very excited to see the benchies.

renoir (4800H) as I have in my laptop is certainly pretty darn good.
gpu is a bit disappointing but it's clearly not the emphasis on the renoir platform, hopefully they will come with one

 

[Deleted member 190774]

I own a 3950x watercooled and never it hits the 4.7. I would say more marketing BS.

What does it hit? I have a 2950X with AU$1000 of hard-line water cooling strapped to it and it never hit 4.4 GHz (even with PBO and everything maxed out). I decided to up the base clock to 100.5 and now it jumps over 4.4 all the time (CB20 7918).

That said, I'm mildly optimistic about the 4000 series CPUs and looking forward to the 4960/70X Threadrippers - providing I can get a decent board.

 

[lexluthermiester]

Intel's 10nm can't get anywhere near the clocks of its 14nm+++++.

You don't and can't know that.

clock speed is what's stopping them committing to it.

No, what's stopping them from full committal to it is the yield problems.

However, we're off-topic. Let's rope it in shall we...

 

[Totally]

I prefer to use 5K instead of 5000.don't know why.

You are literally saying five thousand million hertz.