• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Why there won’t be 5700x CPU

Soldato
Joined
29 May 2005
Posts
5,011
Many people are hoping for a cut down cost of the new 8 core or 5700x to bring some kind of cost reconciliation of the new zen3 8c part to the old zen2 8c part. Below is why I don’t think there will be a 3700x equivalent to the new zen3 8c part.

the simple reason is that there is no incentive for AMD to charge less for their 8c sku. The zen3 ccd is a different animal to zen2. Zen2 is two 4c CCX out together so you can have poor quality CCX being lumbered together with a high quality 4c and to cut cost. In the zen3 8c CCX you don’t have that option. All the cores in that 8c needs to be quality so that certainly means the fabrication is more heavily binned against 5600x and 5900x where 6c per CCX means a less successful CCX can still be made into product therefore less “throw away” and more cost effective.

I think the pricing of 5800x is higher per core than the 5950x also makes sense as they have to use really high binned CCX for 5800x to ensure the boosts are hit on a single CCX where in 5950x they have 2 CCX/CCD so one of them is your high bin one and the other can be slightly lower ranked one and still able to sell as a 16c part without falling foul of the advertised boosting.

I think as the TSMC 7nm process matures and their yields get better there may well be a situation there are less less 6c/CCX skus available as there are more successful 8c/CCX part and that’s why you will see potentially diminishing supply of 5600x and 5900x and pricing of 5800x starts to reduce.

on 5600(non x) or some kind of 5400 (4c) sku, those parts will most likely to exist again in maybe relatively small numbers due to the actual cost of bringing those skus to consumer. And they are there purely to counter intel’s i3 and i5 non-k offerings.

what you think?

Hardware unboxed has a section in the 5800x review which I very much agree on in terms of the pricing of 5800x.
 
Does the unified CCX make any difference in terms of binning?

With the dual 4 core design or unified 8 core design only some of those cores need to hit the peak advertised boost clock.
I would imagine any failed (failed means those can’t hit the supposed boost clock or actual non-functional cores within the 8c CCX) 8c CCX will be binned to 5600x and 5900x.

as you can see from that logical stand point, given statistical probability, a perfect 8core CCX is much harder to come by.
 
If probability of hitting a correctly binned CCX is the same on the 7nm fab for zen2 and zen3 then you have twice as many CCX in zen2 that will be successful due to its smaller 4c CCX than 8c CCX. So you have less 5800x sku. Then you have to yank the prices up to reflect that.

That’s the thinking behind it
 
I think so for the first gen AM5 at least. 5nm will likely see the move to 16/32 but I have a hunch AMD will at some point fork the desktop parts away.
I think the next AM5 CPU will still be on 7nm albeit the more advanced N7+. The 5nm will be generation after that.

Warhol (after Vermeer) is still meant to be 7nm according to AMD’s road maps.

warhol you will probably see great power efficiency from the N7+ so the chips can clock even faster. And some single digit IPC gain from refinement of architecture. Together delivering double digit performance boost compared with the current zen3. But that performance maybe hindered by DDR5 being immature technology.
 
if AMD end up with enough defective CCXs with only 4 or 5 operable cores they could start sticking them together as a 4/8+4/8 combination 5700X without too much of a performance hit vs the 5800X. Any CCXs with 6 cores functional will be used for 5900X or 5600X parts.

thats an interesting proposition regarding 4c ccx usage. this will effectively make the 4+4 the same as current zen 2 which will likely to expose the chip to some IF latencies. I think AMD proabbly wont want to do that. they may be pushing those 4c part into ryzen 3 sku such as the 3300x currently.

but never say never...
 
i think it will make sense to create market segmentation with SMT switched on and switched off only if AMD end up with a lot of poor cores or low bin cores that cant run 2T

otherwise arbitarily switching off SMT whilst the base cost of producing that piece of silicon remains the same makes very little financial sense. they would have been better just out of the box advertise lower clock speeds for 6c and 8c parts and ask for less money but still keeping their margin
 
Most will be able to run 2T.

It's about creating a different line of products where there currently isn't one. You don't want a 5900x customer buying a 5600G. You do want a 10400F customer buying the 5600G. How else do you get there but by offering a better product than the competition. Clock speeds let you win in gaming, one of the easiest ways to increase those speeds is to turn off SMT.

It doesn't matter that the cores can do it. What matters is the absolute cost of the individual bits of silicon and if it's profitable to split the SKU in that way. Which I think is true. £250 5600G would win 6 core 6 thread with high base clock
I can see your point but I don’t think AMD will switch off SMT even for the low end parts. Even the 3100 has SMT and competes with the 10400F on pricing fairly aggressively.

People looking at 4c cheapy parts for gaming will want SMT.

there will however be a huge market for 4c 1T APUs for thinks like HTPC, NAS builds or tiny ITX on all the time low power builds even barebone systems. Currently that market is pure intel atm. I would personally love to have a 4C (with SMT) 25w APU tho. That would be epic.
 
I honestly don't think you are listening to the argument I am making.

I am NOT saying that AMD should remove SMT from it's entire line up. If you want SMT then pay extra for it.

disagree there mate. The most popular CPU is the 3600 and 3700x from AMD and within the intel camp it is the i5 and i7 K parts. Don’t think anyone is buying i3 F parts at all and very few are buying the other F and non K parts. Basically people want SMT for whatever reason.
 
Sorry, but your point doesn't exist. Why? Because I am not advocating for the removal of the parts that have thus far been announced.
The point is that why create something (like intel does arbitrarily) for a market that no one wants or cares about.

for years intel didn’t give us HT and reserved HT for high end i7 and i9. That’s because there is no alternative. Now there is competition, doing something like that isn’t going to help a company to build market share. It will just be slated for being stupid. OEM won’t even be interested in those SKUs. Maybe barebone systems only.
 
Please argue against what I'm actually saying.

SMT off allows for better clock speeds. This is an undisputable fact.

If AMD sell No SMT cpu's they would run FASTER (than the X varient).

All you are doing is going "BUT INTEL" without actually going further.

Are you suggesting that the i5 6600 had a faster clock than the i7 6700... Because it didn't.

Did intel have a 6600 that was HT and one that wasn't... NOOO.

So please actually debate on what's being discussed, instead of just going "but intel" as if it's an argument.
No one is going to buy a 4C4T or even 6C6T possibly 8C8T sku even if it can clock 100MHz or 200Mhz higher unless it is ridiculously cheap

4C4T - £60-£70
6C6T - £80-£90
8C8T - £100-£130

AMD ain’t gonna do that from a financial stand point. A) it will take sales away from their PREMIUM SKU which they want to sell to make a tonne of money B) they aren’t making any significant margins on those so why bother.

they may well bring out something similar to the above down the line but the availability is like non-existent (3300X for example) to counter whatever intel may have. But that’s a completely different ball game
 
Market Share is why

Again you are not thinking how actual consumers think but how an avid enthusiast thinks.

If a benchmark shows a 5300G beating a 5300X in gaming then gamers will buy it. It's that simple. Quoting a frequency difference doesn't matter.

those 2 skus dont exist so hard for me to imagine it. say AMD puts out an 8C no SMT SKU at half of the MRP of the 5600X that is able to offer 5% lift on single core perf. then I think there will be loads of poeple want to buy it.

but why would AMD do that...they could just offer 4C8T with no single core perf boost at half of the cost of the 5600X and it wont take away market share of 5600x and it will satisfy those want budget and reasonable perf. intel cant offer anything to compete atm so it is AMD's market really.
 
Every game that benefits from a faster CPU basically benefits from SMT being off.

Ryzen dynamically increases it's clock in line with thermal headroom. SMT increases heat generated. So yes... It's a performance increase in basically every game... Not one or two as you suggest.

I mean we already know this. But for some reason you guys are playing coy. This has been understood for SMT since ryzen 1. And for years in regards to HT in intel.

It uses more power, it makes the core slower for


Dude are you serious. Imagination kinda relies on the thing not existing. I mean... What am I even debating against now.

Who said ANYTHING about 8c t at half the msrp as the 5600x

All you have just done is create a scenario where it wouldn't work.

5300X - 4 Core 8 Thread - £110 (3.4ghz base)
5300G - 4 Core 4 thread - £90 (4.2ghz base)

5600X - 6 Core 12 thread - £299 (3.7 ghz base)
5600G - 6 Core 6 thread - £250 (4.5 ghz base)

5800X - 8 core 16 thread - £450 (3.8ghz base)
5800G - 8 core 8 thread - £390 (4.7ghz base)

In the above lineup with hardly any thinking too it I can see that gamers would go for the 5800G over the 5600X.

Some 5800X users would go to the G. But by and large if you CURRENTLY want 8 cores with ryzen then you also want the threads.

However, a large gaming only audience don't give two ***** about SMT if it makes their game fps lower.
Right I am gonna forget about those 4c skus cos they make little sense to me in terms of your proposed pricing.

Anyway £50 off from MRRP and loss of 50% logical cores while gaining say another 5% in performance seems a steep price to pay. I mean if on those pricing then why not. Clearly huge profit to make.

My point is that very few people will want a 6C6T sku. 8C8T has some arguments tbh as most games aren’t core heavy only a few titles currently enjoy decent performance lift from having more cores.
 
Turning SMT off with my 3600 makes gaming performance decrease in max FPS and especially the 1% lows so I'm not sure why you think SMT off is better?.
He is saying SMT off you can clock the CPU higher. So SMT on - lower single core clocks and SMT off - higher single core clocks (which you probably need to manually change some setting in bios to allow the CPU to boost even more I suppose).
 
He's saying that turning SMT off reduces heat.

But shouldn't you see gains in every CPU (with a boost function) by turning SMT off if it was a guaranteed win?

You'd then have more thermal headroom and the ability to boost higher on more cores.

So just turning SMT off should benefit existing CPUs as well as his proposed G CPUs.

But the testing does not reveal this to be the case, so far.

he is not techinically wrong, but i suspect there is far more to it than just switching it off and let the CPU manage its thermal headroom. the PBO is done based on some kind of ceiling value and also some sort of in-built boost table.

I dont know if there is any way of altering those to suit SMT off situation other than BIOS changing the thermal envelop or disabling PBO and do per core manual clocking via Ryzen Master.

its all hyperthetical anyways, and technically AMD can product non-SMT chips that can clock way more than current crop to market it as gaming skus. but the appeal of those to people is another debate.
 
Turning off SMT doesn't seem to effect boost or max manual OC with my sample.
i dont think it will under PBO as the boost table is built into the CPU and BIOS. the only way to really see if that makes a different is to do per core overclocking.

too complicated and time consuming to test really.

i think most poeple will find 8 thread is really the minimum requirement for gaming these days to ensure that 1% FPS isnt suffering @ 1080p for instance. switching SMT off on lower core count sku doesnt give a good gaming experience @ 1080p, just ask those with older haswell and skylake i5s...
 
Last edited:
Back
Top Bottom